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 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
239 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
240 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
241 vxd_path(nd->dentry, nd->mnt));
245 int permission(struct inode *inode, int mask, struct nameidata *nd)
249 if (mask & MAY_WRITE) {
250 umode_t mode = inode->i_mode;
253 * Nobody gets write access to a read-only fs.
255 if (IS_RDONLY(inode) &&
256 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
260 * Nobody gets write access to an immutable file.
262 if (IS_IMMUTABLE(inode))
267 /* Ordinary permission routines do not understand MAY_APPEND. */
268 submask = mask & ~MAY_APPEND;
269 if ((retval = xid_permission(inode, mask, nd)))
271 if (inode->i_op && inode->i_op->permission)
272 retval = inode->i_op->permission(inode, submask, nd);
274 retval = generic_permission(inode, submask, NULL);
278 return security_inode_permission(inode, mask, nd);
282 * get_write_access() gets write permission for a file.
283 * put_write_access() releases this write permission.
284 * This is used for regular files.
285 * We cannot support write (and maybe mmap read-write shared) accesses and
286 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
287 * can have the following values:
288 * 0: no writers, no VM_DENYWRITE mappings
289 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
290 * > 0: (i_writecount) users are writing to the file.
292 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
293 * except for the cases where we don't hold i_writecount yet. Then we need to
294 * use {get,deny}_write_access() - these functions check the sign and refuse
295 * to do the change if sign is wrong. Exclusion between them is provided by
296 * the inode->i_lock spinlock.
299 int get_write_access(struct inode * inode)
301 spin_lock(&inode->i_lock);
302 if (atomic_read(&inode->i_writecount) < 0) {
303 spin_unlock(&inode->i_lock);
306 atomic_inc(&inode->i_writecount);
307 spin_unlock(&inode->i_lock);
312 int deny_write_access(struct file * file)
314 struct inode *inode = file->f_dentry->d_inode;
316 spin_lock(&inode->i_lock);
317 if (atomic_read(&inode->i_writecount) > 0) {
318 spin_unlock(&inode->i_lock);
321 atomic_dec(&inode->i_writecount);
322 spin_unlock(&inode->i_lock);
327 void path_release(struct nameidata *nd)
334 * umount() mustn't call path_release()/mntput() as that would clear
337 void path_release_on_umount(struct nameidata *nd)
344 * Internal lookup() using the new generic dcache.
347 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
349 struct dentry * dentry = __d_lookup(parent, name);
351 /* lockess __d_lookup may fail due to concurrent d_move()
352 * in some unrelated directory, so try with d_lookup
355 dentry = d_lookup(parent, name);
357 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
358 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
367 * Short-cut version of permission(), for calling by
368 * path_walk(), when dcache lock is held. Combines parts
369 * of permission() and generic_permission(), and tests ONLY for
370 * MAY_EXEC permission.
372 * If appropriate, check DAC only. If not appropriate, or
373 * short-cut DAC fails, then call permission() to do more
374 * complete permission check.
376 static inline int exec_permission_lite(struct inode *inode,
377 struct nameidata *nd)
379 umode_t mode = inode->i_mode;
381 if (inode->i_op && inode->i_op->permission)
384 if (current->fsuid == inode->i_uid)
386 else if (in_group_p(inode->i_gid))
392 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
395 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
398 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
403 return security_inode_permission(inode, MAY_EXEC, nd);
407 * This is called when everything else fails, and we actually have
408 * to go to the low-level filesystem to find out what we should do..
410 * We get the directory semaphore, and after getting that we also
411 * make sure that nobody added the entry to the dcache in the meantime..
414 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
416 struct dentry * result;
417 struct inode *dir = parent->d_inode;
421 * First re-do the cached lookup just in case it was created
422 * while we waited for the directory semaphore..
424 * FIXME! This could use version numbering or similar to
425 * avoid unnecessary cache lookups.
427 * The "dcache_lock" is purely to protect the RCU list walker
428 * from concurrent renames at this point (we mustn't get false
429 * negatives from the RCU list walk here, unlike the optimistic
432 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
434 result = d_lookup(parent, name);
436 struct dentry * dentry = d_alloc(parent, name);
437 result = ERR_PTR(-ENOMEM);
439 result = dir->i_op->lookup(dir, dentry, nd);
450 * Uhhuh! Nasty case: the cache was re-populated while
451 * we waited on the semaphore. Need to revalidate.
454 if (result->d_op && result->d_op->d_revalidate) {
455 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
457 result = ERR_PTR(-ENOENT);
463 static int __emul_lookup_dentry(const char *, struct nameidata *);
467 walk_init_root(const char *name, struct nameidata *nd)
469 read_lock(¤t->fs->lock);
470 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
471 nd->mnt = mntget(current->fs->altrootmnt);
472 nd->dentry = dget(current->fs->altroot);
473 read_unlock(¤t->fs->lock);
474 if (__emul_lookup_dentry(name,nd))
476 read_lock(¤t->fs->lock);
478 nd->mnt = mntget(current->fs->rootmnt);
479 nd->dentry = dget(current->fs->root);
480 read_unlock(¤t->fs->lock);
484 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
493 if (!walk_init_root(link, nd))
494 /* weird __emul_prefix() stuff did it */
497 res = link_path_walk(link, nd);
499 if (nd->depth || res || nd->last_type!=LAST_NORM)
502 * If it is an iterative symlinks resolution in open_namei() we
503 * have to copy the last component. And all that crap because of
504 * bloody create() on broken symlinks. Furrfu...
507 if (unlikely(!name)) {
511 strcpy(name, nd->last.name);
512 nd->last.name = name;
516 return PTR_ERR(link);
520 struct vfsmount *mnt;
521 struct dentry *dentry;
524 static inline int __do_follow_link(struct path *path, struct nameidata *nd)
527 struct dentry *dentry = path->dentry;
529 touch_atime(path->mnt, dentry);
530 nd_set_link(nd, NULL);
532 if (path->mnt == nd->mnt)
534 error = dentry->d_inode->i_op->follow_link(dentry, nd);
536 char *s = nd_get_link(nd);
538 error = __vfs_follow_link(nd, s);
539 if (dentry->d_inode->i_op->put_link)
540 dentry->d_inode->i_op->put_link(dentry, nd);
549 * This limits recursive symlink follows to 8, while
550 * limiting consecutive symlinks to 40.
552 * Without that kind of total limit, nasty chains of consecutive
553 * symlinks can cause almost arbitrarily long lookups.
555 static inline int do_follow_link(struct path *path, struct nameidata *nd)
558 if (current->link_count >= MAX_NESTED_LINKS)
560 if (current->total_link_count >= 40)
562 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
564 err = security_inode_follow_link(path->dentry, nd);
567 current->link_count++;
568 current->total_link_count++;
570 err = __do_follow_link(path, nd);
571 current->link_count--;
576 if (path->mnt != nd->mnt)
582 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
584 struct vfsmount *parent;
585 struct dentry *mountpoint;
586 spin_lock(&vfsmount_lock);
587 parent=(*mnt)->mnt_parent;
588 if (parent == *mnt) {
589 spin_unlock(&vfsmount_lock);
593 mountpoint=dget((*mnt)->mnt_mountpoint);
594 spin_unlock(&vfsmount_lock);
596 *dentry = mountpoint;
602 /* no need for dcache_lock, as serialization is taken care in
605 static int __follow_mount(struct path *path)
608 while (d_mountpoint(path->dentry)) {
609 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
616 path->dentry = dget(mounted->mnt_root);
622 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
624 while (d_mountpoint(*dentry)) {
625 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
631 *dentry = dget(mounted->mnt_root);
635 /* no need for dcache_lock, as serialization is taken care in
638 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
640 struct vfsmount *mounted;
642 mounted = lookup_mnt(*mnt, *dentry);
647 *dentry = dget(mounted->mnt_root);
653 static inline void follow_dotdot(struct nameidata *nd)
656 struct vfsmount *parent;
657 struct dentry *old = nd->dentry;
659 read_lock(¤t->fs->lock);
660 if (nd->dentry == current->fs->root &&
661 nd->mnt == current->fs->rootmnt) {
662 read_unlock(¤t->fs->lock);
663 /* for sane '/' avoid follow_mount() */
666 read_unlock(¤t->fs->lock);
667 spin_lock(&dcache_lock);
668 if (nd->dentry != nd->mnt->mnt_root) {
669 nd->dentry = dget(nd->dentry->d_parent);
670 spin_unlock(&dcache_lock);
674 spin_unlock(&dcache_lock);
675 spin_lock(&vfsmount_lock);
676 parent = nd->mnt->mnt_parent;
677 if (parent == nd->mnt) {
678 spin_unlock(&vfsmount_lock);
682 nd->dentry = dget(nd->mnt->mnt_mountpoint);
683 spin_unlock(&vfsmount_lock);
688 follow_mount(&nd->mnt, &nd->dentry);
692 * It's more convoluted than I'd like it to be, but... it's still fairly
693 * small and for now I'd prefer to have fast path as straight as possible.
694 * It _is_ time-critical.
696 static int do_lookup(struct nameidata *nd, struct qstr *name,
699 struct vfsmount *mnt = nd->mnt;
700 struct dentry *dentry = __d_lookup(nd->dentry, name);
705 if (dentry->d_op && dentry->d_op->d_revalidate)
706 goto need_revalidate;
707 inode = dentry->d_inode;
710 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
712 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
713 struct proc_dir_entry *de = PDE(inode);
715 if (de && !vx_hide_check(0, de->vx_flags))
720 path->dentry = dentry;
721 __follow_mount(path);
724 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
725 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
726 vxd_path(dentry, mnt));
731 dentry = real_lookup(nd->dentry, name, nd);
737 if (dentry->d_op->d_revalidate(dentry, nd))
739 if (d_invalidate(dentry))
745 return PTR_ERR(dentry);
750 * This is the basic name resolution function, turning a pathname into
751 * the final dentry. We expect 'base' to be positive and a directory.
753 * Returns 0 and nd will have valid dentry and mnt on success.
754 * Returns error and drops reference to input namei data on failure.
756 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
761 unsigned int lookup_flags = nd->flags;
768 inode = nd->dentry->d_inode;
770 lookup_flags = LOOKUP_FOLLOW;
772 /* At this point we know we have a real path component. */
778 err = exec_permission_lite(inode, nd);
779 if (err == -EAGAIN) {
780 err = permission(inode, MAY_EXEC, nd);
786 c = *(const unsigned char *)name;
788 hash = init_name_hash();
791 hash = partial_name_hash(c, hash);
792 c = *(const unsigned char *)name;
793 } while (c && (c != '/'));
794 this.len = name - (const char *) this.name;
795 this.hash = end_name_hash(hash);
797 /* remove trailing slashes? */
800 while (*++name == '/');
802 goto last_with_slashes;
805 * "." and ".." are special - ".." especially so because it has
806 * to be able to know about the current root directory and
807 * parent relationships.
809 if (this.name[0] == '.') switch (this.len) {
813 if (this.name[1] != '.')
816 inode = nd->dentry->d_inode;
822 * See if the low-level filesystem might want
823 * to use its own hash..
825 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
826 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
830 nd->flags |= LOOKUP_CONTINUE;
831 /* This does the actual lookups.. */
832 err = do_lookup(nd, &this, &next);
837 inode = next.dentry->d_inode;
844 if (inode->i_op->follow_link) {
845 err = do_follow_link(&next, nd);
849 inode = nd->dentry->d_inode;
857 if (nd->mnt != next.mnt)
860 nd->dentry = next.dentry;
863 if (!inode->i_op->lookup)
866 /* here ends the main loop */
869 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
871 nd->flags &= ~LOOKUP_CONTINUE;
872 if (lookup_flags & LOOKUP_PARENT)
874 if (this.name[0] == '.') switch (this.len) {
878 if (this.name[1] != '.')
881 inode = nd->dentry->d_inode;
886 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
887 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
891 err = do_lookup(nd, &this, &next);
894 inode = next.dentry->d_inode;
895 if ((lookup_flags & LOOKUP_FOLLOW)
896 && inode && inode->i_op && inode->i_op->follow_link) {
897 err = do_follow_link(&next, nd);
900 inode = nd->dentry->d_inode;
903 if (nd->mnt != next.mnt)
906 nd->dentry = next.dentry;
911 if (lookup_flags & LOOKUP_DIRECTORY) {
913 if (!inode->i_op || !inode->i_op->lookup)
919 nd->last_type = LAST_NORM;
920 if (this.name[0] != '.')
923 nd->last_type = LAST_DOT;
924 else if (this.len == 2 && this.name[1] == '.')
925 nd->last_type = LAST_DOTDOT;
930 * We bypassed the ordinary revalidation routines.
931 * We may need to check the cached dentry for staleness.
933 if (nd->dentry && nd->dentry->d_sb &&
934 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
936 /* Note: we do not d_invalidate() */
937 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
944 if (nd->mnt != next.mnt)
954 * Wrapper to retry pathname resolution whenever the underlying
955 * file system returns an ESTALE.
957 * Retry the whole path once, forcing real lookup requests
958 * instead of relying on the dcache.
960 int fastcall link_path_walk(const char *name, struct nameidata *nd)
962 struct nameidata save = *nd;
965 /* make sure the stuff we saved doesn't go away */
969 result = __link_path_walk(name, nd);
970 if (result == -ESTALE) {
974 nd->flags |= LOOKUP_REVAL;
975 result = __link_path_walk(name, nd);
984 int fastcall path_walk(const char * name, struct nameidata *nd)
986 current->total_link_count = 0;
987 return link_path_walk(name, nd);
991 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
992 * everything is done. Returns 0 and drops input nd, if lookup failed;
994 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
996 if (path_walk(name, nd))
997 return 0; /* something went wrong... */
999 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1000 struct dentry *old_dentry = nd->dentry;
1001 struct vfsmount *old_mnt = nd->mnt;
1002 struct qstr last = nd->last;
1003 int last_type = nd->last_type;
1005 * NAME was not found in alternate root or it's a directory. Try to find
1006 * it in the normal root:
1008 nd->last_type = LAST_ROOT;
1009 read_lock(¤t->fs->lock);
1010 nd->mnt = mntget(current->fs->rootmnt);
1011 nd->dentry = dget(current->fs->root);
1012 read_unlock(¤t->fs->lock);
1013 if (path_walk(name, nd) == 0) {
1014 if (nd->dentry->d_inode) {
1021 nd->dentry = old_dentry;
1024 nd->last_type = last_type;
1029 void set_fs_altroot(void)
1031 char *emul = __emul_prefix();
1032 struct nameidata nd;
1033 struct vfsmount *mnt = NULL, *oldmnt;
1034 struct dentry *dentry = NULL, *olddentry;
1039 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1045 write_lock(¤t->fs->lock);
1046 oldmnt = current->fs->altrootmnt;
1047 olddentry = current->fs->altroot;
1048 current->fs->altrootmnt = mnt;
1049 current->fs->altroot = dentry;
1050 write_unlock(¤t->fs->lock);
1057 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1058 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1062 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1066 read_lock(¤t->fs->lock);
1068 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1069 nd->mnt = mntget(current->fs->altrootmnt);
1070 nd->dentry = dget(current->fs->altroot);
1071 read_unlock(¤t->fs->lock);
1072 if (__emul_lookup_dentry(name,nd))
1073 goto out; /* found in altroot */
1074 read_lock(¤t->fs->lock);
1076 nd->mnt = mntget(current->fs->rootmnt);
1077 nd->dentry = dget(current->fs->root);
1079 nd->mnt = mntget(current->fs->pwdmnt);
1080 nd->dentry = dget(current->fs->pwd);
1082 read_unlock(¤t->fs->lock);
1083 current->total_link_count = 0;
1084 retval = link_path_walk(name, nd);
1086 if (unlikely(current->audit_context
1087 && nd && nd->dentry && nd->dentry->d_inode))
1088 audit_inode(name, nd->dentry->d_inode);
1093 * Restricted form of lookup. Doesn't follow links, single-component only,
1094 * needs parent already locked. Doesn't follow mounts.
1097 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1099 struct dentry * dentry;
1100 struct inode *inode;
1103 inode = base->d_inode;
1104 err = permission(inode, MAY_EXEC, nd);
1105 dentry = ERR_PTR(err);
1110 * See if the low-level filesystem might want
1111 * to use its own hash..
1113 if (base->d_op && base->d_op->d_hash) {
1114 err = base->d_op->d_hash(base, name);
1115 dentry = ERR_PTR(err);
1120 dentry = cached_lookup(base, name, nd);
1122 struct dentry *new = d_alloc(base, name);
1123 dentry = ERR_PTR(-ENOMEM);
1126 dentry = inode->i_op->lookup(inode, new, nd);
1136 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1138 return __lookup_hash(name, base, NULL);
1142 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1153 hash = init_name_hash();
1155 c = *(const unsigned char *)name++;
1156 if (c == '/' || c == '\0')
1158 hash = partial_name_hash(c, hash);
1160 this.hash = end_name_hash(hash);
1162 return lookup_hash(&this, base);
1164 return ERR_PTR(-EACCES);
1170 * is used by most simple commands to get the inode of a specified name.
1171 * Open, link etc use their own routines, but this is enough for things
1174 * namei exists in two versions: namei/lnamei. The only difference is
1175 * that namei follows links, while lnamei does not.
1178 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1180 char *tmp = getname(name);
1181 int err = PTR_ERR(tmp);
1184 err = path_lookup(tmp, flags, nd);
1191 * It's inline, so penalty for filesystems that don't use sticky bit is
1194 static inline int check_sticky(struct inode *dir, struct inode *inode)
1196 if (!(dir->i_mode & S_ISVTX))
1198 if (inode->i_uid == current->fsuid)
1200 if (dir->i_uid == current->fsuid)
1202 return !capable(CAP_FOWNER);
1206 * Check whether we can remove a link victim from directory dir, check
1207 * whether the type of victim is right.
1208 * 1. We can't do it if dir is read-only (done in permission())
1209 * 2. We should have write and exec permissions on dir
1210 * 3. We can't remove anything from append-only dir
1211 * 4. We can't do anything with immutable dir (done in permission())
1212 * 5. If the sticky bit on dir is set we should either
1213 * a. be owner of dir, or
1214 * b. be owner of victim, or
1215 * c. have CAP_FOWNER capability
1216 * 6. If the victim is append-only or immutable we can't do antyhing with
1217 * links pointing to it.
1218 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1219 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1220 * 9. We can't remove a root or mountpoint.
1221 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1222 * nfs_async_unlink().
1224 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1228 if (!victim->d_inode)
1231 BUG_ON(victim->d_parent->d_inode != dir);
1233 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1238 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1239 IS_IXORUNLINK(victim->d_inode))
1242 if (!S_ISDIR(victim->d_inode->i_mode))
1244 if (IS_ROOT(victim))
1246 } else if (S_ISDIR(victim->d_inode->i_mode))
1248 if (IS_DEADDIR(dir))
1250 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1255 /* Check whether we can create an object with dentry child in directory
1257 * 1. We can't do it if child already exists (open has special treatment for
1258 * this case, but since we are inlined it's OK)
1259 * 2. We can't do it if dir is read-only (done in permission())
1260 * 3. We should have write and exec permissions on dir
1261 * 4. We can't do it if dir is immutable (done in permission())
1263 static inline int may_create(struct inode *dir, struct dentry *child,
1264 struct nameidata *nd)
1268 if (IS_DEADDIR(dir))
1270 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1274 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1277 * O_DIRECTORY translates into forcing a directory lookup.
1279 static inline int lookup_flags(unsigned int f)
1281 unsigned long retval = LOOKUP_FOLLOW;
1284 retval &= ~LOOKUP_FOLLOW;
1286 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1287 retval &= ~LOOKUP_FOLLOW;
1289 if (f & O_DIRECTORY)
1290 retval |= LOOKUP_DIRECTORY;
1296 * p1 and p2 should be directories on the same fs.
1298 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1303 down(&p1->d_inode->i_sem);
1307 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1309 for (p = p1; p->d_parent != p; p = p->d_parent) {
1310 if (p->d_parent == p2) {
1311 down(&p2->d_inode->i_sem);
1312 down(&p1->d_inode->i_sem);
1317 for (p = p2; p->d_parent != p; p = p->d_parent) {
1318 if (p->d_parent == p1) {
1319 down(&p1->d_inode->i_sem);
1320 down(&p2->d_inode->i_sem);
1325 down(&p1->d_inode->i_sem);
1326 down(&p2->d_inode->i_sem);
1330 void unlock_rename(struct dentry *p1, struct dentry *p2)
1332 up(&p1->d_inode->i_sem);
1334 up(&p2->d_inode->i_sem);
1335 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1339 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1340 struct nameidata *nd)
1342 int error = may_create(dir, dentry, nd);
1347 if (!dir->i_op || !dir->i_op->create)
1348 return -EACCES; /* shouldn't it be ENOSYS? */
1351 error = security_inode_create(dir, dentry, mode);
1355 error = dir->i_op->create(dir, dentry, mode, nd);
1357 inode_dir_notify(dir, DN_CREATE);
1358 security_inode_post_create(dir, dentry, mode);
1363 int may_open(struct nameidata *nd, int acc_mode, int flag)
1365 struct dentry *dentry = nd->dentry;
1366 struct inode *inode = dentry->d_inode;
1372 if (S_ISLNK(inode->i_mode))
1375 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1378 error = permission(inode, acc_mode, nd);
1383 * FIFO's, sockets and device files are special: they don't
1384 * actually live on the filesystem itself, and as such you
1385 * can write to them even if the filesystem is read-only.
1387 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1389 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1390 if (nd->mnt->mnt_flags & MNT_NODEV)
1394 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1397 * An append-only file must be opened in append mode for writing.
1399 if (IS_APPEND(inode)) {
1400 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1406 /* O_NOATIME can only be set by the owner or superuser */
1407 if (flag & O_NOATIME)
1408 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1412 * Ensure there are no outstanding leases on the file.
1414 error = break_lease(inode, flag);
1418 if (flag & O_TRUNC) {
1419 error = get_write_access(inode);
1424 * Refuse to truncate files with mandatory locks held on them.
1426 error = locks_verify_locked(inode);
1430 error = do_truncate(dentry, 0);
1432 put_write_access(inode);
1436 if (flag & FMODE_WRITE)
1445 * namei for open - this is in fact almost the whole open-routine.
1447 * Note that the low bits of "flag" aren't the same as in the open
1448 * system call - they are 00 - no permissions needed
1449 * 01 - read permission needed
1450 * 10 - write permission needed
1451 * 11 - read/write permissions needed
1452 * which is a lot more logical, and also allows the "no perm" needed
1453 * for symlinks (where the permissions are checked later).
1456 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1458 int acc_mode, error = 0;
1463 acc_mode = ACC_MODE(flag);
1465 /* Allow the LSM permission hook to distinguish append
1466 access from general write access. */
1467 if (flag & O_APPEND)
1468 acc_mode |= MAY_APPEND;
1470 /* Fill in the open() intent data */
1471 nd->intent.open.flags = flag;
1472 nd->intent.open.create_mode = mode;
1475 * The simplest case - just a plain lookup.
1477 if (!(flag & O_CREAT)) {
1478 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1485 * Create - we need to know the parent.
1487 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1492 * We have the parent and last component. First of all, check
1493 * that we are not asked to creat(2) an obvious directory - that
1497 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1501 nd->flags &= ~LOOKUP_PARENT;
1502 down(&dir->d_inode->i_sem);
1503 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1507 error = PTR_ERR(path.dentry);
1508 if (IS_ERR(path.dentry)) {
1509 up(&dir->d_inode->i_sem);
1513 /* Negative dentry, just create the file */
1514 if (!path.dentry->d_inode) {
1515 if (!IS_POSIXACL(dir->d_inode))
1516 mode &= ~current->fs->umask;
1517 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1518 up(&dir->d_inode->i_sem);
1520 nd->dentry = path.dentry;
1523 /* Don't check for write permission, don't truncate */
1530 * It already exists.
1532 up(&dir->d_inode->i_sem);
1538 if (__follow_mount(&path)) {
1540 if (flag & O_NOFOLLOW)
1544 if (!path.dentry->d_inode)
1546 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1550 nd->dentry = path.dentry;
1551 if (nd->mnt != path.mnt)
1555 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1558 error = may_open(nd, acc_mode, flag);
1565 if (nd->mnt != path.mnt)
1573 if (flag & O_NOFOLLOW)
1576 * This is subtle. Instead of calling do_follow_link() we do the
1577 * thing by hands. The reason is that this way we have zero link_count
1578 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1579 * After that we have the parent and last component, i.e.
1580 * we are in the same situation as after the first path_walk().
1581 * Well, almost - if the last component is normal we get its copy
1582 * stored in nd->last.name and we will have to putname() it when we
1583 * are done. Procfs-like symlinks just set LAST_BIND.
1585 nd->flags |= LOOKUP_PARENT;
1586 error = security_inode_follow_link(path.dentry, nd);
1589 error = __do_follow_link(&path, nd);
1592 nd->flags &= ~LOOKUP_PARENT;
1593 if (nd->last_type == LAST_BIND)
1596 if (nd->last_type != LAST_NORM)
1598 if (nd->last.name[nd->last.len]) {
1599 putname(nd->last.name);
1604 putname(nd->last.name);
1608 down(&dir->d_inode->i_sem);
1609 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1611 putname(nd->last.name);
1616 * lookup_create - lookup a dentry, creating it if it doesn't exist
1617 * @nd: nameidata info
1618 * @is_dir: directory flag
1620 * Simple function to lookup and return a dentry and create it
1621 * if it doesn't exist. Is SMP-safe.
1623 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1625 struct dentry *dentry;
1627 down(&nd->dentry->d_inode->i_sem);
1628 dentry = ERR_PTR(-EEXIST);
1629 if (nd->last_type != LAST_NORM)
1631 nd->flags &= ~LOOKUP_PARENT;
1632 dentry = lookup_hash(&nd->last, nd->dentry);
1635 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1640 dentry = ERR_PTR(-ENOENT);
1644 EXPORT_SYMBOL_GPL(lookup_create);
1646 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1648 int error = may_create(dir, dentry, NULL);
1653 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1656 if (!dir->i_op || !dir->i_op->mknod)
1659 error = security_inode_mknod(dir, dentry, mode, dev);
1664 error = dir->i_op->mknod(dir, dentry, mode, dev);
1666 inode_dir_notify(dir, DN_CREATE);
1667 security_inode_post_mknod(dir, dentry, mode, dev);
1672 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1676 struct dentry * dentry;
1677 struct nameidata nd;
1681 tmp = getname(filename);
1683 return PTR_ERR(tmp);
1685 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1688 dentry = lookup_create(&nd, 0);
1689 error = PTR_ERR(dentry);
1691 if (!IS_POSIXACL(nd.dentry->d_inode))
1692 mode &= ~current->fs->umask;
1693 if (!IS_ERR(dentry)) {
1694 switch (mode & S_IFMT) {
1695 case 0: case S_IFREG:
1696 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1698 case S_IFCHR: case S_IFBLK:
1699 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1700 new_decode_dev(dev));
1702 case S_IFIFO: case S_IFSOCK:
1703 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1713 up(&nd.dentry->d_inode->i_sem);
1721 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1723 int error = may_create(dir, dentry, NULL);
1728 if (!dir->i_op || !dir->i_op->mkdir)
1731 mode &= (S_IRWXUGO|S_ISVTX);
1732 error = security_inode_mkdir(dir, dentry, mode);
1737 error = dir->i_op->mkdir(dir, dentry, mode);
1739 inode_dir_notify(dir, DN_CREATE);
1740 security_inode_post_mkdir(dir,dentry, mode);
1745 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1750 tmp = getname(pathname);
1751 error = PTR_ERR(tmp);
1753 struct dentry *dentry;
1754 struct nameidata nd;
1756 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1759 dentry = lookup_create(&nd, 1);
1760 error = PTR_ERR(dentry);
1761 if (!IS_ERR(dentry)) {
1762 if (!IS_POSIXACL(nd.dentry->d_inode))
1763 mode &= ~current->fs->umask;
1764 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1767 up(&nd.dentry->d_inode->i_sem);
1777 * We try to drop the dentry early: we should have
1778 * a usage count of 2 if we're the only user of this
1779 * dentry, and if that is true (possibly after pruning
1780 * the dcache), then we drop the dentry now.
1782 * A low-level filesystem can, if it choses, legally
1785 * if (!d_unhashed(dentry))
1788 * if it cannot handle the case of removing a directory
1789 * that is still in use by something else..
1791 void dentry_unhash(struct dentry *dentry)
1794 if (atomic_read(&dentry->d_count))
1795 shrink_dcache_parent(dentry);
1796 spin_lock(&dcache_lock);
1797 spin_lock(&dentry->d_lock);
1798 if (atomic_read(&dentry->d_count) == 2)
1800 spin_unlock(&dentry->d_lock);
1801 spin_unlock(&dcache_lock);
1804 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1806 int error = may_delete(dir, dentry, 1);
1811 if (!dir->i_op || !dir->i_op->rmdir)
1816 down(&dentry->d_inode->i_sem);
1817 dentry_unhash(dentry);
1818 if (d_mountpoint(dentry))
1821 error = security_inode_rmdir(dir, dentry);
1823 error = dir->i_op->rmdir(dir, dentry);
1825 dentry->d_inode->i_flags |= S_DEAD;
1828 up(&dentry->d_inode->i_sem);
1830 inode_dir_notify(dir, DN_DELETE);
1838 asmlinkage long sys_rmdir(const char __user * pathname)
1842 struct dentry *dentry;
1843 struct nameidata nd;
1845 name = getname(pathname);
1847 return PTR_ERR(name);
1849 error = path_lookup(name, LOOKUP_PARENT, &nd);
1853 switch(nd.last_type) {
1864 down(&nd.dentry->d_inode->i_sem);
1865 dentry = lookup_hash(&nd.last, nd.dentry);
1866 error = PTR_ERR(dentry);
1867 if (!IS_ERR(dentry)) {
1868 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1871 up(&nd.dentry->d_inode->i_sem);
1879 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1881 int error = may_delete(dir, dentry, 0);
1886 if (!dir->i_op || !dir->i_op->unlink)
1891 down(&dentry->d_inode->i_sem);
1892 if (d_mountpoint(dentry))
1895 error = security_inode_unlink(dir, dentry);
1897 error = dir->i_op->unlink(dir, dentry);
1899 up(&dentry->d_inode->i_sem);
1901 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1902 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1904 inode_dir_notify(dir, DN_DELETE);
1910 * Make sure that the actual truncation of the file will occur outside its
1911 * directory's i_sem. Truncate can take a long time if there is a lot of
1912 * writeout happening, and we don't want to prevent access to the directory
1913 * while waiting on the I/O.
1915 asmlinkage long sys_unlink(const char __user * pathname)
1919 struct dentry *dentry;
1920 struct nameidata nd;
1921 struct inode *inode = NULL;
1923 name = getname(pathname);
1925 return PTR_ERR(name);
1927 error = path_lookup(name, LOOKUP_PARENT, &nd);
1931 if (nd.last_type != LAST_NORM)
1933 down(&nd.dentry->d_inode->i_sem);
1934 dentry = lookup_hash(&nd.last, nd.dentry);
1935 error = PTR_ERR(dentry);
1936 if (!IS_ERR(dentry)) {
1937 /* Why not before? Because we want correct error value */
1938 if (nd.last.name[nd.last.len])
1940 inode = dentry->d_inode;
1942 atomic_inc(&inode->i_count);
1943 error = vfs_unlink(nd.dentry->d_inode, dentry);
1947 up(&nd.dentry->d_inode->i_sem);
1949 iput(inode); /* truncate the inode here */
1957 error = !dentry->d_inode ? -ENOENT :
1958 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1962 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1964 int error = may_create(dir, dentry, NULL);
1969 if (!dir->i_op || !dir->i_op->symlink)
1972 error = security_inode_symlink(dir, dentry, oldname);
1977 error = dir->i_op->symlink(dir, dentry, oldname);
1979 inode_dir_notify(dir, DN_CREATE);
1980 security_inode_post_symlink(dir, dentry, oldname);
1985 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1991 from = getname(oldname);
1993 return PTR_ERR(from);
1994 to = getname(newname);
1995 error = PTR_ERR(to);
1997 struct dentry *dentry;
1998 struct nameidata nd;
2000 error = path_lookup(to, LOOKUP_PARENT, &nd);
2003 dentry = lookup_create(&nd, 0);
2004 error = PTR_ERR(dentry);
2005 if (!IS_ERR(dentry)) {
2006 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2009 up(&nd.dentry->d_inode->i_sem);
2018 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2020 struct inode *inode = old_dentry->d_inode;
2026 error = may_create(dir, new_dentry, NULL);
2030 if (dir->i_sb != inode->i_sb)
2034 * A link to an append-only or immutable file cannot be created.
2036 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2038 if (!dir->i_op || !dir->i_op->link)
2040 if (S_ISDIR(old_dentry->d_inode->i_mode))
2043 error = security_inode_link(old_dentry, dir, new_dentry);
2047 down(&old_dentry->d_inode->i_sem);
2049 error = dir->i_op->link(old_dentry, dir, new_dentry);
2050 up(&old_dentry->d_inode->i_sem);
2052 inode_dir_notify(dir, DN_CREATE);
2053 security_inode_post_link(old_dentry, dir, new_dentry);
2059 * Hardlinks are often used in delicate situations. We avoid
2060 * security-related surprises by not following symlinks on the
2063 * We don't follow them on the oldname either to be compatible
2064 * with linux 2.0, and to avoid hard-linking to directories
2065 * and other special files. --ADM
2067 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2069 struct dentry *new_dentry;
2070 struct nameidata nd, old_nd;
2074 to = getname(newname);
2078 error = __user_walk(oldname, 0, &old_nd);
2081 error = path_lookup(to, LOOKUP_PARENT, &nd);
2085 if (old_nd.mnt != nd.mnt)
2087 new_dentry = lookup_create(&nd, 0);
2088 error = PTR_ERR(new_dentry);
2089 if (!IS_ERR(new_dentry)) {
2090 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2093 up(&nd.dentry->d_inode->i_sem);
2097 path_release(&old_nd);
2105 * The worst of all namespace operations - renaming directory. "Perverted"
2106 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2108 * a) we can get into loop creation. Check is done in is_subdir().
2109 * b) race potential - two innocent renames can create a loop together.
2110 * That's where 4.4 screws up. Current fix: serialization on
2111 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2113 * c) we have to lock _three_ objects - parents and victim (if it exists).
2114 * And that - after we got ->i_sem on parents (until then we don't know
2115 * whether the target exists). Solution: try to be smart with locking
2116 * order for inodes. We rely on the fact that tree topology may change
2117 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2118 * move will be locked. Thus we can rank directories by the tree
2119 * (ancestors first) and rank all non-directories after them.
2120 * That works since everybody except rename does "lock parent, lookup,
2121 * lock child" and rename is under ->s_vfs_rename_sem.
2122 * HOWEVER, it relies on the assumption that any object with ->lookup()
2123 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2124 * we'd better make sure that there's no link(2) for them.
2125 * d) some filesystems don't support opened-but-unlinked directories,
2126 * either because of layout or because they are not ready to deal with
2127 * all cases correctly. The latter will be fixed (taking this sort of
2128 * stuff into VFS), but the former is not going away. Solution: the same
2129 * trick as in rmdir().
2130 * e) conversion from fhandle to dentry may come in the wrong moment - when
2131 * we are removing the target. Solution: we will have to grab ->i_sem
2132 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2133 * ->i_sem on parents, which works but leads to some truely excessive
2136 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2137 struct inode *new_dir, struct dentry *new_dentry)
2140 struct inode *target;
2143 * If we are going to change the parent - check write permissions,
2144 * we'll need to flip '..'.
2146 if (new_dir != old_dir) {
2147 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2152 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2156 target = new_dentry->d_inode;
2158 down(&target->i_sem);
2159 dentry_unhash(new_dentry);
2161 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2164 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2167 target->i_flags |= S_DEAD;
2169 if (d_unhashed(new_dentry))
2170 d_rehash(new_dentry);
2174 d_move(old_dentry,new_dentry);
2175 security_inode_post_rename(old_dir, old_dentry,
2176 new_dir, new_dentry);
2181 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2182 struct inode *new_dir, struct dentry *new_dentry)
2184 struct inode *target;
2187 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2192 target = new_dentry->d_inode;
2194 down(&target->i_sem);
2195 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2198 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2200 /* The following d_move() should become unconditional */
2201 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2202 d_move(old_dentry, new_dentry);
2203 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2211 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2212 struct inode *new_dir, struct dentry *new_dentry)
2215 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2217 if (old_dentry->d_inode == new_dentry->d_inode)
2220 error = may_delete(old_dir, old_dentry, is_dir);
2224 if (!new_dentry->d_inode)
2225 error = may_create(new_dir, new_dentry, NULL);
2227 error = may_delete(new_dir, new_dentry, is_dir);
2231 if (!old_dir->i_op || !old_dir->i_op->rename)
2234 DQUOT_INIT(old_dir);
2235 DQUOT_INIT(new_dir);
2238 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2240 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2242 if (old_dir == new_dir)
2243 inode_dir_notify(old_dir, DN_RENAME);
2245 inode_dir_notify(old_dir, DN_DELETE);
2246 inode_dir_notify(new_dir, DN_CREATE);
2252 static inline int do_rename(const char * oldname, const char * newname)
2255 struct dentry * old_dir, * new_dir;
2256 struct dentry * old_dentry, *new_dentry;
2257 struct dentry * trap;
2258 struct nameidata oldnd, newnd;
2260 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2264 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2269 if (oldnd.mnt != newnd.mnt)
2272 old_dir = oldnd.dentry;
2274 if (oldnd.last_type != LAST_NORM)
2277 new_dir = newnd.dentry;
2278 if (newnd.last_type != LAST_NORM)
2281 trap = lock_rename(new_dir, old_dir);
2283 old_dentry = lookup_hash(&oldnd.last, old_dir);
2284 error = PTR_ERR(old_dentry);
2285 if (IS_ERR(old_dentry))
2287 /* source must exist */
2289 if (!old_dentry->d_inode)
2291 /* unless the source is a directory trailing slashes give -ENOTDIR */
2292 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2294 if (oldnd.last.name[oldnd.last.len])
2296 if (newnd.last.name[newnd.last.len])
2299 /* source should not be ancestor of target */
2301 if (old_dentry == trap)
2303 new_dentry = lookup_hash(&newnd.last, new_dir);
2304 error = PTR_ERR(new_dentry);
2305 if (IS_ERR(new_dentry))
2307 /* target should not be an ancestor of source */
2309 if (new_dentry == trap)
2312 error = vfs_rename(old_dir->d_inode, old_dentry,
2313 new_dir->d_inode, new_dentry);
2319 unlock_rename(new_dir, old_dir);
2321 path_release(&newnd);
2323 path_release(&oldnd);
2328 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2334 from = getname(oldname);
2336 return PTR_ERR(from);
2337 to = getname(newname);
2338 error = PTR_ERR(to);
2340 error = do_rename(from,to);
2347 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2351 len = PTR_ERR(link);
2356 if (len > (unsigned) buflen)
2358 if (copy_to_user(buffer, link, len))
2365 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2366 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2367 * using) it for any given inode is up to filesystem.
2369 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2371 struct nameidata nd;
2374 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2376 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2377 if (dentry->d_inode->i_op->put_link)
2378 dentry->d_inode->i_op->put_link(dentry, &nd);
2383 int vfs_follow_link(struct nameidata *nd, const char *link)
2385 return __vfs_follow_link(nd, link);
2388 /* get the link contents into pagecache */
2389 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2392 struct address_space *mapping = dentry->d_inode->i_mapping;
2393 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2397 wait_on_page_locked(page);
2398 if (!PageUptodate(page))
2404 page_cache_release(page);
2405 return ERR_PTR(-EIO);
2411 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2413 struct page *page = NULL;
2414 char *s = page_getlink(dentry, &page);
2415 int res = vfs_readlink(dentry,buffer,buflen,s);
2418 page_cache_release(page);
2423 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2426 nd_set_link(nd, page_getlink(dentry, &page));
2430 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2432 if (!IS_ERR(nd_get_link(nd))) {
2434 page = find_get_page(dentry->d_inode->i_mapping, 0);
2438 page_cache_release(page);
2439 page_cache_release(page);
2443 int page_symlink(struct inode *inode, const char *symname, int len)
2445 struct address_space *mapping = inode->i_mapping;
2446 struct page *page = grab_cache_page(mapping, 0);
2452 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2455 kaddr = kmap_atomic(page, KM_USER0);
2456 memcpy(kaddr, symname, len-1);
2457 kunmap_atomic(kaddr, KM_USER0);
2458 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2460 * Notice that we are _not_ going to block here - end of page is
2461 * unmapped, so this will only try to map the rest of page, see
2462 * that it is unmapped (typically even will not look into inode -
2463 * ->i_size will be enough for everything) and zero it out.
2464 * OTOH it's obviously correct and should make the page up-to-date.
2466 if (!PageUptodate(page)) {
2467 err = mapping->a_ops->readpage(NULL, page);
2468 wait_on_page_locked(page);
2472 page_cache_release(page);
2475 mark_inode_dirty(inode);
2479 page_cache_release(page);
2484 struct inode_operations page_symlink_inode_operations = {
2485 .readlink = generic_readlink,
2486 .follow_link = page_follow_link_light,
2487 .put_link = page_put_link,
2490 EXPORT_SYMBOL(__user_walk);
2491 EXPORT_SYMBOL(follow_down);
2492 EXPORT_SYMBOL(follow_up);
2493 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2494 EXPORT_SYMBOL(getname);
2495 EXPORT_SYMBOL(lock_rename);
2496 EXPORT_SYMBOL(lookup_hash);
2497 EXPORT_SYMBOL(lookup_one_len);
2498 EXPORT_SYMBOL(page_follow_link_light);
2499 EXPORT_SYMBOL(page_put_link);
2500 EXPORT_SYMBOL(page_readlink);
2501 EXPORT_SYMBOL(page_symlink);
2502 EXPORT_SYMBOL(page_symlink_inode_operations);
2503 EXPORT_SYMBOL(path_lookup);
2504 EXPORT_SYMBOL(path_release);
2505 EXPORT_SYMBOL(path_walk);
2506 EXPORT_SYMBOL(permission);
2507 EXPORT_SYMBOL(unlock_rename);
2508 EXPORT_SYMBOL(vfs_create);
2509 EXPORT_SYMBOL(vfs_follow_link);
2510 EXPORT_SYMBOL(vfs_link);
2511 EXPORT_SYMBOL(vfs_mkdir);
2512 EXPORT_SYMBOL(vfs_mknod);
2513 EXPORT_SYMBOL(generic_permission);
2514 EXPORT_SYMBOL(vfs_readlink);
2515 EXPORT_SYMBOL(vfs_rename);
2516 EXPORT_SYMBOL(vfs_rmdir);
2517 EXPORT_SYMBOL(vfs_symlink);
2518 EXPORT_SYMBOL(vfs_unlink);
2519 EXPORT_SYMBOL(dentry_unhash);
2520 EXPORT_SYMBOL(generic_readlink);