4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/namei.h>
35 #include <linux/proc_fs.h>
36 #include <linux/vs_base.h>
37 #include <linux/vserver/inode.h>
38 #include <linux/vserver/debug.h>
39 #include <asm/namei.h>
40 #include <asm/uaccess.h>
42 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
44 /* [Feb-1997 T. Schoebel-Theuer]
45 * Fundamental changes in the pathname lookup mechanisms (namei)
46 * were necessary because of omirr. The reason is that omirr needs
47 * to know the _real_ pathname, not the user-supplied one, in case
48 * of symlinks (and also when transname replacements occur).
50 * The new code replaces the old recursive symlink resolution with
51 * an iterative one (in case of non-nested symlink chains). It does
52 * this with calls to <fs>_follow_link().
53 * As a side effect, dir_namei(), _namei() and follow_link() are now
54 * replaced with a single function lookup_dentry() that can handle all
55 * the special cases of the former code.
57 * With the new dcache, the pathname is stored at each inode, at least as
58 * long as the refcount of the inode is positive. As a side effect, the
59 * size of the dcache depends on the inode cache and thus is dynamic.
61 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
62 * resolution to correspond with current state of the code.
64 * Note that the symlink resolution is not *completely* iterative.
65 * There is still a significant amount of tail- and mid- recursion in
66 * the algorithm. Also, note that <fs>_readlink() is not used in
67 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
68 * may return different results than <fs>_follow_link(). Many virtual
69 * filesystems (including /proc) exhibit this behavior.
72 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
73 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
74 * and the name already exists in form of a symlink, try to create the new
75 * name indicated by the symlink. The old code always complained that the
76 * name already exists, due to not following the symlink even if its target
77 * is nonexistent. The new semantics affects also mknod() and link() when
78 * the name is a symlink pointing to a non-existant name.
80 * I don't know which semantics is the right one, since I have no access
81 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
82 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
83 * "old" one. Personally, I think the new semantics is much more logical.
84 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
85 * file does succeed in both HP-UX and SunOs, but not in Solaris
86 * and in the old Linux semantics.
89 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
90 * semantics. See the comments in "open_namei" and "do_link" below.
92 * [10-Sep-98 Alan Modra] Another symlink change.
95 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
96 * inside the path - always follow.
97 * in the last component in creation/removal/renaming - never follow.
98 * if LOOKUP_FOLLOW passed - follow.
99 * if the pathname has trailing slashes - follow.
100 * otherwise - don't follow.
101 * (applied in that order).
103 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
104 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
105 * During the 2.4 we need to fix the userland stuff depending on it -
106 * hopefully we will be able to get rid of that wart in 2.5. So far only
107 * XEmacs seems to be relying on it...
110 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
111 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
112 * any extra contention...
115 /* In order to reduce some races, while at the same time doing additional
116 * checking and hopefully speeding things up, we copy filenames to the
117 * kernel data space before using them..
119 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
120 * PATH_MAX includes the nul terminator --RR.
122 static int do_getname(const char __user *filename, char *page)
125 unsigned long len = PATH_MAX;
127 if (!segment_eq(get_fs(), KERNEL_DS)) {
128 if ((unsigned long) filename >= TASK_SIZE)
130 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
131 len = TASK_SIZE - (unsigned long) filename;
134 retval = strncpy_from_user(page, filename, len);
138 return -ENAMETOOLONG;
144 char * getname(const char __user * filename)
148 result = ERR_PTR(-ENOMEM);
151 int retval = do_getname(filename, tmp);
156 result = ERR_PTR(retval);
159 audit_getname(result);
163 #ifdef CONFIG_AUDITSYSCALL
164 void putname(const char *name)
166 if (unlikely(current->audit_context))
171 EXPORT_SYMBOL(putname);
176 * generic_permission - check for access rights on a Posix-like filesystem
177 * @inode: inode to check access rights for
178 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
179 * @check_acl: optional callback to check for Posix ACLs
181 * Used to check for read/write/execute permissions on a file.
182 * We use "fsuid" for this, letting us set arbitrary permissions
183 * for filesystem access without changing the "normal" uids which
184 * are used for other things..
186 int generic_permission(struct inode *inode, int mask,
187 int (*check_acl)(struct inode *inode, int mask))
189 umode_t mode = inode->i_mode;
191 if (current->fsuid == inode->i_uid)
194 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
195 int error = check_acl(inode, mask);
196 if (error == -EACCES)
197 goto check_capabilities;
198 else if (error != -EAGAIN)
202 if (in_group_p(inode->i_gid))
207 * If the DACs are ok we don't need any capability check.
209 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
214 * Read/write DACs are always overridable.
215 * Executable DACs are overridable if at least one exec bit is set.
217 if (!(mask & MAY_EXEC) ||
218 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
219 if (capable(CAP_DAC_OVERRIDE))
223 * Searching includes executable on directories, else just read.
225 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
226 if (capable(CAP_DAC_READ_SEARCH))
232 static inline int vx_barrier(struct inode *inode)
234 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN)) {
235 vxwprintk(1, "xid=%d did hit the barrier.",
242 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
244 if (vx_barrier(inode))
246 if (inode->i_xid == 0)
248 #ifdef CONFIG_VSERVER_FILESHARING
249 /* MEF: PlanetLab FS module assumes that any file that can be
250 * named (e.g., via a cross mount) is not hidden from another
251 * context or the admin context.
253 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC))
256 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
259 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
260 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
265 int permission(struct inode *inode, int mask, struct nameidata *nd)
269 if (mask & MAY_WRITE) {
270 umode_t mode = inode->i_mode;
273 * Nobody gets write access to a read-only fs.
275 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
276 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
280 * Nobody gets write access to an immutable file.
282 if (IS_IMMUTABLE(inode))
287 /* Ordinary permission routines do not understand MAY_APPEND. */
288 submask = mask & ~MAY_APPEND;
289 if ((retval = xid_permission(inode, mask, nd)))
291 if (inode->i_op && inode->i_op->permission)
292 retval = inode->i_op->permission(inode, submask, nd);
294 retval = generic_permission(inode, submask, NULL);
298 return security_inode_permission(inode, mask, nd);
302 * vfs_permission - check for access rights to a given path
303 * @nd: lookup result that describes the path
304 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
306 * Used to check for read/write/execute permissions on a path.
307 * We use "fsuid" for this, letting us set arbitrary permissions
308 * for filesystem access without changing the "normal" uids which
309 * are used for other things.
311 int vfs_permission(struct nameidata *nd, int mask)
313 return permission(nd->dentry->d_inode, mask, nd);
317 * file_permission - check for additional access rights to a given file
318 * @file: file to check access rights for
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
321 * Used to check for read/write/execute permissions on an already opened
325 * Do not use this function in new code. All access checks should
326 * be done using vfs_permission().
328 int file_permission(struct file *file, int mask)
330 return permission(file->f_dentry->d_inode, mask, NULL);
334 * get_write_access() gets write permission for a file.
335 * put_write_access() releases this write permission.
336 * This is used for regular files.
337 * We cannot support write (and maybe mmap read-write shared) accesses and
338 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
339 * can have the following values:
340 * 0: no writers, no VM_DENYWRITE mappings
341 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
342 * > 0: (i_writecount) users are writing to the file.
344 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
345 * except for the cases where we don't hold i_writecount yet. Then we need to
346 * use {get,deny}_write_access() - these functions check the sign and refuse
347 * to do the change if sign is wrong. Exclusion between them is provided by
348 * the inode->i_lock spinlock.
351 int get_write_access(struct inode * inode)
353 spin_lock(&inode->i_lock);
354 if (atomic_read(&inode->i_writecount) < 0) {
355 spin_unlock(&inode->i_lock);
358 atomic_inc(&inode->i_writecount);
359 spin_unlock(&inode->i_lock);
364 int deny_write_access(struct file * file)
366 struct inode *inode = file->f_dentry->d_inode;
368 spin_lock(&inode->i_lock);
369 if (atomic_read(&inode->i_writecount) > 0) {
370 spin_unlock(&inode->i_lock);
373 atomic_dec(&inode->i_writecount);
374 spin_unlock(&inode->i_lock);
379 void path_release(struct nameidata *nd)
386 * umount() mustn't call path_release()/mntput() as that would clear
389 void path_release_on_umount(struct nameidata *nd)
392 mntput_no_expire(nd->mnt);
396 * release_open_intent - free up open intent resources
397 * @nd: pointer to nameidata
399 void release_open_intent(struct nameidata *nd)
401 if (nd->intent.open.file->f_dentry == NULL)
402 put_filp(nd->intent.open.file);
404 fput(nd->intent.open.file);
408 * Internal lookup() using the new generic dcache.
411 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
413 struct dentry * dentry = __d_lookup(parent, name);
415 /* lockess __d_lookup may fail due to concurrent d_move()
416 * in some unrelated directory, so try with d_lookup
419 dentry = d_lookup(parent, name);
421 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
422 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
431 * Short-cut version of permission(), for calling by
432 * path_walk(), when dcache lock is held. Combines parts
433 * of permission() and generic_permission(), and tests ONLY for
434 * MAY_EXEC permission.
436 * If appropriate, check DAC only. If not appropriate, or
437 * short-cut DAC fails, then call permission() to do more
438 * complete permission check.
440 static int exec_permission_lite(struct inode *inode,
441 struct nameidata *nd)
443 umode_t mode = inode->i_mode;
445 if (vx_barrier(inode))
447 if (inode->i_op && inode->i_op->permission)
450 if (current->fsuid == inode->i_uid)
452 else if (in_group_p(inode->i_gid))
458 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
461 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
464 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
469 return security_inode_permission(inode, MAY_EXEC, nd);
473 * This is called when everything else fails, and we actually have
474 * to go to the low-level filesystem to find out what we should do..
476 * We get the directory semaphore, and after getting that we also
477 * make sure that nobody added the entry to the dcache in the meantime..
480 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
482 struct dentry * result;
483 struct inode *dir = parent->d_inode;
485 mutex_lock(&dir->i_mutex);
487 * First re-do the cached lookup just in case it was created
488 * while we waited for the directory semaphore..
490 * FIXME! This could use version numbering or similar to
491 * avoid unnecessary cache lookups.
493 * The "dcache_lock" is purely to protect the RCU list walker
494 * from concurrent renames at this point (we mustn't get false
495 * negatives from the RCU list walk here, unlike the optimistic
498 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
500 result = d_lookup(parent, name);
502 struct dentry * dentry = d_alloc(parent, name);
503 result = ERR_PTR(-ENOMEM);
505 result = dir->i_op->lookup(dir, dentry, nd);
511 mutex_unlock(&dir->i_mutex);
516 * Uhhuh! Nasty case: the cache was re-populated while
517 * we waited on the semaphore. Need to revalidate.
519 mutex_unlock(&dir->i_mutex);
520 if (result->d_op && result->d_op->d_revalidate) {
521 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
523 result = ERR_PTR(-ENOENT);
529 static int __emul_lookup_dentry(const char *, struct nameidata *);
532 static __always_inline int
533 walk_init_root(const char *name, struct nameidata *nd)
535 read_lock(¤t->fs->lock);
536 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
537 nd->mnt = mntget(current->fs->altrootmnt);
538 nd->dentry = dget(current->fs->altroot);
539 read_unlock(¤t->fs->lock);
540 if (__emul_lookup_dentry(name,nd))
542 read_lock(¤t->fs->lock);
544 nd->mnt = mntget(current->fs->rootmnt);
545 nd->dentry = dget(current->fs->root);
546 read_unlock(¤t->fs->lock);
550 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
559 if (!walk_init_root(link, nd))
560 /* weird __emul_prefix() stuff did it */
563 res = link_path_walk(link, nd);
565 if (nd->depth || res || nd->last_type!=LAST_NORM)
568 * If it is an iterative symlinks resolution in open_namei() we
569 * have to copy the last component. And all that crap because of
570 * bloody create() on broken symlinks. Furrfu...
573 if (unlikely(!name)) {
577 strcpy(name, nd->last.name);
578 nd->last.name = name;
582 return PTR_ERR(link);
586 struct vfsmount *mnt;
587 struct dentry *dentry;
590 static inline void dput_path(struct path *path, struct nameidata *nd)
593 if (path->mnt != nd->mnt)
597 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
600 if (nd->mnt != path->mnt)
603 nd->dentry = path->dentry;
606 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
610 struct dentry *dentry = path->dentry;
612 touch_atime(path->mnt, dentry);
613 nd_set_link(nd, NULL);
615 if (path->mnt != nd->mnt) {
616 path_to_nameidata(path, nd);
620 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
621 error = PTR_ERR(cookie);
622 if (!IS_ERR(cookie)) {
623 char *s = nd_get_link(nd);
626 error = __vfs_follow_link(nd, s);
627 if (dentry->d_inode->i_op->put_link)
628 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
637 * This limits recursive symlink follows to 8, while
638 * limiting consecutive symlinks to 40.
640 * Without that kind of total limit, nasty chains of consecutive
641 * symlinks can cause almost arbitrarily long lookups.
643 static inline int do_follow_link(struct path *path, struct nameidata *nd)
646 if (current->link_count >= MAX_NESTED_LINKS)
648 if (current->total_link_count >= 40)
650 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
652 err = security_inode_follow_link(path->dentry, nd);
655 current->link_count++;
656 current->total_link_count++;
658 err = __do_follow_link(path, nd);
659 current->link_count--;
668 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
670 struct vfsmount *parent;
671 struct dentry *mountpoint;
672 spin_lock(&vfsmount_lock);
673 parent=(*mnt)->mnt_parent;
674 if (parent == *mnt) {
675 spin_unlock(&vfsmount_lock);
679 mountpoint=dget((*mnt)->mnt_mountpoint);
680 spin_unlock(&vfsmount_lock);
682 *dentry = mountpoint;
688 /* no need for dcache_lock, as serialization is taken care in
691 static int __follow_mount(struct path *path)
694 while (d_mountpoint(path->dentry)) {
695 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
702 path->dentry = dget(mounted->mnt_root);
708 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
710 while (d_mountpoint(*dentry)) {
711 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
717 *dentry = dget(mounted->mnt_root);
721 /* no need for dcache_lock, as serialization is taken care in
724 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
726 struct vfsmount *mounted;
728 mounted = lookup_mnt(*mnt, *dentry);
733 *dentry = dget(mounted->mnt_root);
739 static __always_inline void follow_dotdot(struct nameidata *nd)
742 struct vfsmount *parent;
743 struct dentry *old = nd->dentry;
745 read_lock(¤t->fs->lock);
746 if (nd->dentry == current->fs->root &&
747 nd->mnt == current->fs->rootmnt) {
748 read_unlock(¤t->fs->lock);
749 /* for sane '/' avoid follow_mount() */
752 read_unlock(¤t->fs->lock);
753 spin_lock(&dcache_lock);
754 if (nd->dentry != nd->mnt->mnt_root) {
755 nd->dentry = dget(nd->dentry->d_parent);
756 spin_unlock(&dcache_lock);
760 spin_unlock(&dcache_lock);
761 spin_lock(&vfsmount_lock);
762 parent = nd->mnt->mnt_parent;
763 if (parent == nd->mnt) {
764 spin_unlock(&vfsmount_lock);
768 nd->dentry = dget(nd->mnt->mnt_mountpoint);
769 spin_unlock(&vfsmount_lock);
774 follow_mount(&nd->mnt, &nd->dentry);
778 * It's more convoluted than I'd like it to be, but... it's still fairly
779 * small and for now I'd prefer to have fast path as straight as possible.
780 * It _is_ time-critical.
782 static int do_lookup(struct nameidata *nd, struct qstr *name,
783 struct path *path, int atomic)
785 struct vfsmount *mnt = nd->mnt;
786 struct dentry *dentry = __d_lookup(nd->dentry, name);
791 if (dentry->d_op && dentry->d_op->d_revalidate)
792 goto need_revalidate;
793 inode = dentry->d_inode;
796 #ifdef CONFIG_VSERVER_FILESHARING
797 /* MEF: PlanetLab FS module assumes that any file that can be
798 * named (e.g., via a cross mount) is not hidden from another
799 * context or the admin context.
801 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC|VX_ADMIN)) {
804 else /* do the following check */
806 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
808 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
809 struct proc_dir_entry *de = PDE(inode);
811 if (de && !vx_hide_check(0, de->vx_flags))
816 path->dentry = dentry;
817 __follow_mount(path);
820 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
821 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
822 vxd_path(dentry, mnt));
828 return -EWOULDBLOCKIO;
829 dentry = real_lookup(nd->dentry, name, nd);
836 return -EWOULDBLOCKIO;
837 if (dentry->d_op->d_revalidate(dentry, nd))
839 if (d_invalidate(dentry))
845 return PTR_ERR(dentry);
850 * This is the basic name resolution function, turning a pathname into
851 * the final dentry. We expect 'base' to be positive and a directory.
853 * Returns 0 and nd will have valid dentry and mnt on success.
854 * Returns error and drops reference to input namei data on failure.
856 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
861 unsigned int lookup_flags = nd->flags;
863 atomic = (lookup_flags & LOOKUP_ATOMIC);
870 inode = nd->dentry->d_inode;
872 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
874 /* At this point we know we have a real path component. */
880 nd->flags |= LOOKUP_CONTINUE;
881 err = exec_permission_lite(inode, nd);
883 err = vfs_permission(nd, MAY_EXEC);
888 c = *(const unsigned char *)name;
890 hash = init_name_hash();
893 hash = partial_name_hash(c, hash);
894 c = *(const unsigned char *)name;
895 } while (c && (c != '/'));
896 this.len = name - (const char *) this.name;
897 this.hash = end_name_hash(hash);
899 /* remove trailing slashes? */
902 while (*++name == '/');
904 goto last_with_slashes;
907 * "." and ".." are special - ".." especially so because it has
908 * to be able to know about the current root directory and
909 * parent relationships.
911 if (this.name[0] == '.') switch (this.len) {
915 if (this.name[1] != '.')
918 inode = nd->dentry->d_inode;
924 * See if the low-level filesystem might want
925 * to use its own hash..
927 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
928 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
932 /* This does the actual lookups.. */
933 err = do_lookup(nd, &this, &next, atomic);
938 inode = next.dentry->d_inode;
945 if (inode->i_op->follow_link) {
946 err = do_follow_link(&next, nd);
950 inode = nd->dentry->d_inode;
957 path_to_nameidata(&next, nd);
959 if (!inode->i_op->lookup)
962 /* here ends the main loop */
965 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
967 /* Clear LOOKUP_CONTINUE iff it was previously unset */
968 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
969 if (lookup_flags & LOOKUP_PARENT)
971 if (this.name[0] == '.') switch (this.len) {
975 if (this.name[1] != '.')
978 inode = nd->dentry->d_inode;
983 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
984 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
988 err = do_lookup(nd, &this, &next, atomic);
991 inode = next.dentry->d_inode;
992 if ((lookup_flags & LOOKUP_FOLLOW)
993 && inode && inode->i_op && inode->i_op->follow_link) {
994 err = do_follow_link(&next, nd);
997 inode = nd->dentry->d_inode;
999 path_to_nameidata(&next, nd);
1003 if (lookup_flags & LOOKUP_DIRECTORY) {
1005 if (!inode->i_op || !inode->i_op->lookup)
1011 nd->last_type = LAST_NORM;
1012 if (this.name[0] != '.')
1015 nd->last_type = LAST_DOT;
1016 else if (this.len == 2 && this.name[1] == '.')
1017 nd->last_type = LAST_DOTDOT;
1022 * We bypassed the ordinary revalidation routines.
1023 * We may need to check the cached dentry for staleness.
1025 if (nd->dentry && nd->dentry->d_sb &&
1026 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1028 /* Note: we do not d_invalidate() */
1029 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
1035 dput_path(&next, nd);
1044 * Wrapper to retry pathname resolution whenever the underlying
1045 * file system returns an ESTALE.
1047 * Retry the whole path once, forcing real lookup requests
1048 * instead of relying on the dcache.
1050 int fastcall link_path_walk(const char *name, struct nameidata *nd)
1052 struct nameidata save = *nd;
1055 /* make sure the stuff we saved doesn't go away */
1059 result = __link_path_walk(name, nd);
1060 if (result == -ESTALE) {
1064 nd->flags |= LOOKUP_REVAL;
1065 result = __link_path_walk(name, nd);
1074 int fastcall path_walk(const char * name, struct nameidata *nd)
1076 current->total_link_count = 0;
1077 return link_path_walk(name, nd);
1081 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1082 * everything is done. Returns 0 and drops input nd, if lookup failed;
1084 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1086 if (path_walk(name, nd))
1087 return 0; /* something went wrong... */
1089 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1090 struct dentry *old_dentry = nd->dentry;
1091 struct vfsmount *old_mnt = nd->mnt;
1092 struct qstr last = nd->last;
1093 int last_type = nd->last_type;
1095 * NAME was not found in alternate root or it's a directory. Try to find
1096 * it in the normal root:
1098 nd->last_type = LAST_ROOT;
1099 read_lock(¤t->fs->lock);
1100 nd->mnt = mntget(current->fs->rootmnt);
1101 nd->dentry = dget(current->fs->root);
1102 read_unlock(¤t->fs->lock);
1103 if (path_walk(name, nd) == 0) {
1104 if (nd->dentry->d_inode) {
1111 nd->dentry = old_dentry;
1114 nd->last_type = last_type;
1119 void set_fs_altroot(void)
1121 char *emul = __emul_prefix();
1122 struct nameidata nd;
1123 struct vfsmount *mnt = NULL, *oldmnt;
1124 struct dentry *dentry = NULL, *olddentry;
1129 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1135 write_lock(¤t->fs->lock);
1136 oldmnt = current->fs->altrootmnt;
1137 olddentry = current->fs->altroot;
1138 current->fs->altrootmnt = mnt;
1139 current->fs->altroot = dentry;
1140 write_unlock(¤t->fs->lock);
1147 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1148 static int fastcall do_path_lookup(int dfd, const char *name,
1149 unsigned int flags, struct nameidata *nd)
1155 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1160 read_lock(¤t->fs->lock);
1161 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1162 nd->mnt = mntget(current->fs->altrootmnt);
1163 nd->dentry = dget(current->fs->altroot);
1164 read_unlock(¤t->fs->lock);
1165 if (__emul_lookup_dentry(name,nd))
1166 goto out; /* found in altroot */
1167 read_lock(¤t->fs->lock);
1169 nd->mnt = mntget(current->fs->rootmnt);
1170 nd->dentry = dget(current->fs->root);
1171 read_unlock(¤t->fs->lock);
1172 } else if (dfd == AT_FDCWD) {
1173 read_lock(¤t->fs->lock);
1174 nd->mnt = mntget(current->fs->pwdmnt);
1175 nd->dentry = dget(current->fs->pwd);
1176 read_unlock(¤t->fs->lock);
1178 struct dentry *dentry;
1180 file = fget_light(dfd, &fput_needed);
1185 dentry = file->f_dentry;
1188 if (!S_ISDIR(dentry->d_inode->i_mode))
1191 retval = file_permission(file, MAY_EXEC);
1195 nd->mnt = mntget(file->f_vfsmnt);
1196 nd->dentry = dget(dentry);
1198 fput_light(file, fput_needed);
1200 current->total_link_count = 0;
1201 retval = link_path_walk(name, nd);
1203 if (likely(retval == 0)) {
1204 if (unlikely(current->audit_context && nd && nd->dentry &&
1205 nd->dentry->d_inode))
1206 audit_inode(name, nd->dentry->d_inode, flags);
1212 fput_light(file, fput_needed);
1216 int fastcall path_lookup(const char *name, unsigned int flags,
1217 struct nameidata *nd)
1219 return do_path_lookup(AT_FDCWD, name, flags, nd);
1222 static int __path_lookup_intent_open(int dfd, const char *name,
1223 unsigned int lookup_flags, struct nameidata *nd,
1224 int open_flags, int create_mode)
1226 struct file *filp = get_empty_filp();
1231 nd->intent.open.file = filp;
1232 nd->intent.open.flags = open_flags;
1233 nd->intent.open.create_mode = create_mode;
1234 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1235 if (IS_ERR(nd->intent.open.file)) {
1237 err = PTR_ERR(nd->intent.open.file);
1240 } else if (err != 0)
1241 release_open_intent(nd);
1246 * path_lookup_open - lookup a file path with open intent
1247 * @dfd: the directory to use as base, or AT_FDCWD
1248 * @name: pointer to file name
1249 * @lookup_flags: lookup intent flags
1250 * @nd: pointer to nameidata
1251 * @open_flags: open intent flags
1253 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1254 struct nameidata *nd, int open_flags)
1256 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1261 * path_lookup_create - lookup a file path with open + create intent
1262 * @dfd: the directory to use as base, or AT_FDCWD
1263 * @name: pointer to file name
1264 * @lookup_flags: lookup intent flags
1265 * @nd: pointer to nameidata
1266 * @open_flags: open intent flags
1267 * @create_mode: create intent flags
1269 static int path_lookup_create(int dfd, const char *name,
1270 unsigned int lookup_flags, struct nameidata *nd,
1271 int open_flags, int create_mode)
1273 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1274 nd, open_flags, create_mode);
1277 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1278 struct nameidata *nd, int open_flags)
1280 char *tmp = getname(name);
1281 int err = PTR_ERR(tmp);
1284 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1291 * Restricted form of lookup. Doesn't follow links, single-component only,
1292 * needs parent already locked. Doesn't follow mounts.
1295 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1297 struct dentry * dentry;
1298 struct inode *inode;
1301 inode = base->d_inode;
1302 err = permission(inode, MAY_EXEC, nd);
1303 dentry = ERR_PTR(err);
1308 * See if the low-level filesystem might want
1309 * to use its own hash..
1311 if (base->d_op && base->d_op->d_hash) {
1312 err = base->d_op->d_hash(base, name);
1313 dentry = ERR_PTR(err);
1318 dentry = cached_lookup(base, name, nd);
1320 struct dentry *new = d_alloc(base, name);
1321 dentry = ERR_PTR(-ENOMEM);
1324 dentry = inode->i_op->lookup(inode, new, nd);
1334 static struct dentry *lookup_hash(struct nameidata *nd)
1336 return __lookup_hash(&nd->last, nd->dentry, nd);
1340 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1351 hash = init_name_hash();
1353 c = *(const unsigned char *)name++;
1354 if (c == '/' || c == '\0')
1356 hash = partial_name_hash(c, hash);
1358 this.hash = end_name_hash(hash);
1360 return __lookup_hash(&this, base, NULL);
1362 return ERR_PTR(-EACCES);
1368 * is used by most simple commands to get the inode of a specified name.
1369 * Open, link etc use their own routines, but this is enough for things
1372 * namei exists in two versions: namei/lnamei. The only difference is
1373 * that namei follows links, while lnamei does not.
1376 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1377 struct nameidata *nd)
1379 char *tmp = getname(name);
1380 int err = PTR_ERR(tmp);
1383 err = do_path_lookup(dfd, tmp, flags, nd);
1389 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1391 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1395 * It's inline, so penalty for filesystems that don't use sticky bit is
1398 static inline int check_sticky(struct inode *dir, struct inode *inode)
1400 if (!(dir->i_mode & S_ISVTX))
1402 if (inode->i_uid == current->fsuid)
1404 if (dir->i_uid == current->fsuid)
1406 return !capable(CAP_FOWNER);
1410 * Check whether we can remove a link victim from directory dir, check
1411 * whether the type of victim is right.
1412 * 1. We can't do it if dir is read-only (done in permission())
1413 * 2. We should have write and exec permissions on dir
1414 * 3. We can't remove anything from append-only dir
1415 * 4. We can't do anything with immutable dir (done in permission())
1416 * 5. If the sticky bit on dir is set we should either
1417 * a. be owner of dir, or
1418 * b. be owner of victim, or
1419 * c. have CAP_FOWNER capability
1420 * 6. If the victim is append-only or immutable we can't do antyhing with
1421 * links pointing to it.
1422 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1423 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1424 * 9. We can't remove a root or mountpoint.
1425 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1426 * nfs_async_unlink().
1428 static int may_delete(struct inode *dir, struct dentry *victim,
1429 int isdir, struct nameidata *nd)
1433 if (!victim->d_inode)
1436 BUG_ON(victim->d_parent->d_inode != dir);
1437 audit_inode_child(victim->d_name.name, victim->d_inode, dir->i_ino);
1439 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1444 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1445 IS_IXORUNLINK(victim->d_inode))
1448 if (!S_ISDIR(victim->d_inode->i_mode))
1450 if (IS_ROOT(victim))
1452 } else if (S_ISDIR(victim->d_inode->i_mode))
1454 if (IS_DEADDIR(dir))
1456 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1461 /* Check whether we can create an object with dentry child in directory
1463 * 1. We can't do it if child already exists (open has special treatment for
1464 * this case, but since we are inlined it's OK)
1465 * 2. We can't do it if dir is read-only (done in permission())
1466 * 3. We should have write and exec permissions on dir
1467 * 4. We can't do it if dir is immutable (done in permission())
1469 static inline int may_create(struct inode *dir, struct dentry *child,
1470 struct nameidata *nd)
1474 if (IS_DEADDIR(dir))
1476 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1480 * O_DIRECTORY translates into forcing a directory lookup.
1482 static inline int lookup_flags(unsigned int f)
1484 unsigned long retval = LOOKUP_FOLLOW;
1487 retval &= ~LOOKUP_FOLLOW;
1489 if (f & O_DIRECTORY)
1490 retval |= LOOKUP_DIRECTORY;
1491 if (f & O_ATOMICLOOKUP)
1492 retval |= LOOKUP_ATOMIC;
1498 * p1 and p2 should be directories on the same fs.
1500 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1505 mutex_lock(&p1->d_inode->i_mutex);
1509 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1511 for (p = p1; p->d_parent != p; p = p->d_parent) {
1512 if (p->d_parent == p2) {
1513 mutex_lock(&p2->d_inode->i_mutex);
1514 mutex_lock(&p1->d_inode->i_mutex);
1519 for (p = p2; p->d_parent != p; p = p->d_parent) {
1520 if (p->d_parent == p1) {
1521 mutex_lock(&p1->d_inode->i_mutex);
1522 mutex_lock(&p2->d_inode->i_mutex);
1527 mutex_lock(&p1->d_inode->i_mutex);
1528 mutex_lock(&p2->d_inode->i_mutex);
1532 void unlock_rename(struct dentry *p1, struct dentry *p2)
1534 mutex_unlock(&p1->d_inode->i_mutex);
1536 mutex_unlock(&p2->d_inode->i_mutex);
1537 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1541 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1542 struct nameidata *nd)
1544 int error = may_create(dir, dentry, nd);
1549 if (!dir->i_op || !dir->i_op->create)
1550 return -EACCES; /* shouldn't it be ENOSYS? */
1553 error = security_inode_create(dir, dentry, mode);
1557 error = dir->i_op->create(dir, dentry, mode, nd);
1559 fsnotify_create(dir, dentry);
1563 int may_open(struct nameidata *nd, int acc_mode, int flag)
1565 struct dentry *dentry = nd->dentry;
1566 struct inode *inode = dentry->d_inode;
1572 if (S_ISLNK(inode->i_mode))
1575 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1578 error = vfs_permission(nd, acc_mode);
1583 * FIFO's, sockets and device files are special: they don't
1584 * actually live on the filesystem itself, and as such you
1585 * can write to them even if the filesystem is read-only.
1587 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1589 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1590 if (nd->mnt->mnt_flags & MNT_NODEV)
1594 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1595 && (flag & FMODE_WRITE))
1598 * An append-only file must be opened in append mode for writing.
1600 if (IS_APPEND(inode)) {
1601 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1607 /* O_NOATIME can only be set by the owner or superuser */
1608 if (flag & O_NOATIME)
1609 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1613 * Ensure there are no outstanding leases on the file.
1615 error = break_lease(inode, flag);
1619 if (flag & O_TRUNC) {
1620 error = get_write_access(inode);
1625 * Refuse to truncate files with mandatory locks held on them.
1627 error = locks_verify_locked(inode);
1631 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1633 put_write_access(inode);
1637 if (flag & FMODE_WRITE)
1646 * namei for open - this is in fact almost the whole open-routine.
1648 * Note that the low bits of "flag" aren't the same as in the open
1649 * system call - they are 00 - no permissions needed
1650 * 01 - read permission needed
1651 * 10 - write permission needed
1652 * 11 - read/write permissions needed
1653 * which is a lot more logical, and also allows the "no perm" needed
1654 * for symlinks (where the permissions are checked later).
1657 int open_namei(int dfd, const char *pathname, int flag,
1658 int mode, struct nameidata *nd)
1660 int acc_mode, error;
1665 acc_mode = ACC_MODE(flag);
1667 /* O_TRUNC implies we need access checks for write permissions */
1669 acc_mode |= MAY_WRITE;
1671 /* Allow the LSM permission hook to distinguish append
1672 access from general write access. */
1673 if (flag & O_APPEND)
1674 acc_mode |= MAY_APPEND;
1677 * The simplest case - just a plain lookup.
1679 if (!(flag & O_CREAT)) {
1680 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1688 * Create - we need to know the parent.
1690 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1695 * We have the parent and last component. First of all, check
1696 * that we are not asked to creat(2) an obvious directory - that
1700 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1704 nd->flags &= ~LOOKUP_PARENT;
1705 mutex_lock(&dir->d_inode->i_mutex);
1706 path.dentry = lookup_hash(nd);
1710 error = PTR_ERR(path.dentry);
1711 if (IS_ERR(path.dentry)) {
1712 mutex_unlock(&dir->d_inode->i_mutex);
1716 if (IS_ERR(nd->intent.open.file)) {
1717 mutex_unlock(&dir->d_inode->i_mutex);
1718 error = PTR_ERR(nd->intent.open.file);
1722 /* Negative dentry, just create the file */
1723 if (!path.dentry->d_inode) {
1724 if (!IS_POSIXACL(dir->d_inode))
1725 mode &= ~current->fs->umask;
1726 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1727 mutex_unlock(&dir->d_inode->i_mutex);
1729 nd->dentry = path.dentry;
1732 /* Don't check for write permission, don't truncate */
1739 * It already exists.
1741 mutex_unlock(&dir->d_inode->i_mutex);
1747 if (__follow_mount(&path)) {
1749 if (flag & O_NOFOLLOW)
1753 if (!path.dentry->d_inode)
1755 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1758 path_to_nameidata(&path, nd);
1760 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1763 error = may_open(nd, acc_mode, flag);
1769 dput_path(&path, nd);
1771 if (!IS_ERR(nd->intent.open.file))
1772 release_open_intent(nd);
1778 if (flag & O_NOFOLLOW)
1781 * This is subtle. Instead of calling do_follow_link() we do the
1782 * thing by hands. The reason is that this way we have zero link_count
1783 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1784 * After that we have the parent and last component, i.e.
1785 * we are in the same situation as after the first path_walk().
1786 * Well, almost - if the last component is normal we get its copy
1787 * stored in nd->last.name and we will have to putname() it when we
1788 * are done. Procfs-like symlinks just set LAST_BIND.
1790 nd->flags |= LOOKUP_PARENT;
1791 error = security_inode_follow_link(path.dentry, nd);
1794 error = __do_follow_link(&path, nd);
1796 /* Does someone understand code flow here? Or it is only
1797 * me so stupid? Anathema to whoever designed this non-sense
1798 * with "intent.open".
1800 release_open_intent(nd);
1803 nd->flags &= ~LOOKUP_PARENT;
1804 if (nd->last_type == LAST_BIND)
1807 if (nd->last_type != LAST_NORM)
1809 if (nd->last.name[nd->last.len]) {
1810 __putname(nd->last.name);
1815 __putname(nd->last.name);
1819 mutex_lock(&dir->d_inode->i_mutex);
1820 path.dentry = lookup_hash(nd);
1822 __putname(nd->last.name);
1827 * lookup_create - lookup a dentry, creating it if it doesn't exist
1828 * @nd: nameidata info
1829 * @is_dir: directory flag
1831 * Simple function to lookup and return a dentry and create it
1832 * if it doesn't exist. Is SMP-safe.
1834 * Returns with nd->dentry->d_inode->i_mutex locked.
1836 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1838 struct dentry *dentry = ERR_PTR(-EEXIST);
1840 mutex_lock(&nd->dentry->d_inode->i_mutex);
1842 * Yucky last component or no last component at all?
1843 * (foo/., foo/.., /////)
1845 if (nd->last_type != LAST_NORM)
1847 nd->flags &= ~LOOKUP_PARENT;
1850 * Do the final lookup.
1852 dentry = lookup_hash(nd);
1857 * Special case - lookup gave negative, but... we had foo/bar/
1858 * From the vfs_mknod() POV we just have a negative dentry -
1859 * all is fine. Let's be bastards - you had / on the end, you've
1860 * been asking for (non-existent) directory. -ENOENT for you.
1862 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1867 dentry = ERR_PTR(-ENOENT);
1871 EXPORT_SYMBOL_GPL(lookup_create);
1873 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1874 int mode, dev_t dev, struct nameidata *nd)
1876 int error = may_create(dir, dentry, nd);
1881 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1884 if (!dir->i_op || !dir->i_op->mknod)
1887 error = security_inode_mknod(dir, dentry, mode, dev);
1892 error = dir->i_op->mknod(dir, dentry, mode, dev);
1894 fsnotify_create(dir, dentry);
1898 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1903 struct dentry * dentry;
1904 struct nameidata nd;
1908 tmp = getname(filename);
1910 return PTR_ERR(tmp);
1912 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1915 dentry = lookup_create(&nd, 0);
1916 error = PTR_ERR(dentry);
1918 if (!IS_POSIXACL(nd.dentry->d_inode))
1919 mode &= ~current->fs->umask;
1920 if (!IS_ERR(dentry)) {
1921 switch (mode & S_IFMT) {
1922 case 0: case S_IFREG:
1923 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1925 case S_IFCHR: case S_IFBLK:
1926 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1927 new_decode_dev(dev), &nd);
1929 case S_IFIFO: case S_IFSOCK:
1930 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1941 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1949 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1951 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1954 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1955 int mode, struct nameidata *nd)
1957 int error = may_create(dir, dentry, nd);
1962 if (!dir->i_op || !dir->i_op->mkdir)
1965 mode &= (S_IRWXUGO|S_ISVTX);
1966 error = security_inode_mkdir(dir, dentry, mode);
1971 error = dir->i_op->mkdir(dir, dentry, mode);
1973 fsnotify_mkdir(dir, dentry);
1977 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1982 tmp = getname(pathname);
1983 error = PTR_ERR(tmp);
1985 struct dentry *dentry;
1986 struct nameidata nd;
1988 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1991 dentry = lookup_create(&nd, 1);
1992 error = PTR_ERR(dentry);
1993 if (!IS_ERR(dentry)) {
1994 if (!IS_POSIXACL(nd.dentry->d_inode))
1995 mode &= ~current->fs->umask;
1996 error = vfs_mkdir(nd.dentry->d_inode, dentry,
2000 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2009 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2011 return sys_mkdirat(AT_FDCWD, pathname, mode);
2015 * We try to drop the dentry early: we should have
2016 * a usage count of 2 if we're the only user of this
2017 * dentry, and if that is true (possibly after pruning
2018 * the dcache), then we drop the dentry now.
2020 * A low-level filesystem can, if it choses, legally
2023 * if (!d_unhashed(dentry))
2026 * if it cannot handle the case of removing a directory
2027 * that is still in use by something else..
2029 void dentry_unhash(struct dentry *dentry)
2032 if (atomic_read(&dentry->d_count))
2033 shrink_dcache_parent(dentry);
2034 spin_lock(&dcache_lock);
2035 spin_lock(&dentry->d_lock);
2036 if (atomic_read(&dentry->d_count) == 2)
2038 spin_unlock(&dentry->d_lock);
2039 spin_unlock(&dcache_lock);
2042 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
2043 struct nameidata *nd)
2045 int error = may_delete(dir, dentry, 1, nd);
2050 if (!dir->i_op || !dir->i_op->rmdir)
2055 mutex_lock(&dentry->d_inode->i_mutex);
2056 dentry_unhash(dentry);
2057 if (d_mountpoint(dentry))
2060 error = security_inode_rmdir(dir, dentry);
2062 error = dir->i_op->rmdir(dir, dentry);
2064 dentry->d_inode->i_flags |= S_DEAD;
2067 mutex_unlock(&dentry->d_inode->i_mutex);
2076 static long do_rmdir(int dfd, const char __user *pathname)
2080 struct dentry *dentry;
2081 struct nameidata nd;
2083 name = getname(pathname);
2085 return PTR_ERR(name);
2087 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2091 switch(nd.last_type) {
2102 mutex_lock(&nd.dentry->d_inode->i_mutex);
2103 dentry = lookup_hash(&nd);
2104 error = PTR_ERR(dentry);
2105 if (!IS_ERR(dentry)) {
2106 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
2109 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2117 asmlinkage long sys_rmdir(const char __user *pathname)
2119 return do_rmdir(AT_FDCWD, pathname);
2122 int vfs_unlink(struct inode *dir, struct dentry *dentry,
2123 struct nameidata *nd)
2125 int error = may_delete(dir, dentry, 0, nd);
2130 if (!dir->i_op || !dir->i_op->unlink)
2135 mutex_lock(&dentry->d_inode->i_mutex);
2136 if (d_mountpoint(dentry))
2139 error = security_inode_unlink(dir, dentry);
2141 error = dir->i_op->unlink(dir, dentry);
2143 mutex_unlock(&dentry->d_inode->i_mutex);
2145 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2146 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2154 * Make sure that the actual truncation of the file will occur outside its
2155 * directory's i_mutex. Truncate can take a long time if there is a lot of
2156 * writeout happening, and we don't want to prevent access to the directory
2157 * while waiting on the I/O.
2159 static long do_unlinkat(int dfd, const char __user *pathname)
2163 struct dentry *dentry;
2164 struct nameidata nd;
2165 struct inode *inode = NULL;
2167 name = getname(pathname);
2169 return PTR_ERR(name);
2171 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2175 if (nd.last_type != LAST_NORM)
2177 mutex_lock(&nd.dentry->d_inode->i_mutex);
2178 dentry = lookup_hash(&nd);
2179 error = PTR_ERR(dentry);
2180 if (!IS_ERR(dentry)) {
2181 /* Why not before? Because we want correct error value */
2182 if (nd.last.name[nd.last.len])
2184 inode = dentry->d_inode;
2186 atomic_inc(&inode->i_count);
2187 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
2191 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2193 iput(inode); /* truncate the inode here */
2201 error = !dentry->d_inode ? -ENOENT :
2202 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2206 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2208 if ((flag & ~AT_REMOVEDIR) != 0)
2211 if (flag & AT_REMOVEDIR)
2212 return do_rmdir(dfd, pathname);
2214 return do_unlinkat(dfd, pathname);
2217 asmlinkage long sys_unlink(const char __user *pathname)
2219 return do_unlinkat(AT_FDCWD, pathname);
2222 int vfs_symlink(struct inode *dir, struct dentry *dentry,
2223 const char *oldname, int mode, struct nameidata *nd)
2225 int error = may_create(dir, dentry, nd);
2230 if (!dir->i_op || !dir->i_op->symlink)
2233 error = security_inode_symlink(dir, dentry, oldname);
2238 error = dir->i_op->symlink(dir, dentry, oldname);
2240 fsnotify_create(dir, dentry);
2244 asmlinkage long sys_symlinkat(const char __user *oldname,
2245 int newdfd, const char __user *newname)
2251 from = getname(oldname);
2253 return PTR_ERR(from);
2254 to = getname(newname);
2255 error = PTR_ERR(to);
2257 struct dentry *dentry;
2258 struct nameidata nd;
2260 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2263 dentry = lookup_create(&nd, 0);
2264 error = PTR_ERR(dentry);
2265 if (!IS_ERR(dentry)) {
2266 error = vfs_symlink(nd.dentry->d_inode, dentry,
2267 from, S_IALLUGO, &nd);
2270 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2279 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2281 return sys_symlinkat(oldname, AT_FDCWD, newname);
2284 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2285 struct dentry *new_dentry, struct nameidata *nd)
2287 struct inode *inode = old_dentry->d_inode;
2293 error = may_create(dir, new_dentry, nd);
2297 if (dir->i_sb != inode->i_sb)
2301 * A link to an append-only or immutable file cannot be created.
2303 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2305 if (!dir->i_op || !dir->i_op->link)
2307 if (S_ISDIR(old_dentry->d_inode->i_mode))
2310 error = security_inode_link(old_dentry, dir, new_dentry);
2314 mutex_lock(&old_dentry->d_inode->i_mutex);
2316 error = dir->i_op->link(old_dentry, dir, new_dentry);
2317 mutex_unlock(&old_dentry->d_inode->i_mutex);
2319 fsnotify_create(dir, new_dentry);
2324 * Hardlinks are often used in delicate situations. We avoid
2325 * security-related surprises by not following symlinks on the
2328 * We don't follow them on the oldname either to be compatible
2329 * with linux 2.0, and to avoid hard-linking to directories
2330 * and other special files. --ADM
2332 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2333 int newdfd, const char __user *newname,
2336 struct dentry *new_dentry;
2337 struct nameidata nd, old_nd;
2344 to = getname(newname);
2348 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
2351 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2355 * We allow hard-links to be created to a bind-mount as long
2356 * as the bind-mount is not read-only. Checking for cross-dev
2357 * links is subsumed by the superblock check in vfs_link().
2360 if (MNT_IS_RDONLY(old_nd.mnt))
2362 new_dentry = lookup_create(&nd, 0);
2363 error = PTR_ERR(new_dentry);
2364 if (!IS_ERR(new_dentry)) {
2365 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2369 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2373 path_release(&old_nd);
2380 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2382 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2386 * The worst of all namespace operations - renaming directory. "Perverted"
2387 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2389 * a) we can get into loop creation. Check is done in is_subdir().
2390 * b) race potential - two innocent renames can create a loop together.
2391 * That's where 4.4 screws up. Current fix: serialization on
2392 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2394 * c) we have to lock _three_ objects - parents and victim (if it exists).
2395 * And that - after we got ->i_mutex on parents (until then we don't know
2396 * whether the target exists). Solution: try to be smart with locking
2397 * order for inodes. We rely on the fact that tree topology may change
2398 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2399 * move will be locked. Thus we can rank directories by the tree
2400 * (ancestors first) and rank all non-directories after them.
2401 * That works since everybody except rename does "lock parent, lookup,
2402 * lock child" and rename is under ->s_vfs_rename_mutex.
2403 * HOWEVER, it relies on the assumption that any object with ->lookup()
2404 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2405 * we'd better make sure that there's no link(2) for them.
2406 * d) some filesystems don't support opened-but-unlinked directories,
2407 * either because of layout or because they are not ready to deal with
2408 * all cases correctly. The latter will be fixed (taking this sort of
2409 * stuff into VFS), but the former is not going away. Solution: the same
2410 * trick as in rmdir().
2411 * e) conversion from fhandle to dentry may come in the wrong moment - when
2412 * we are removing the target. Solution: we will have to grab ->i_mutex
2413 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2414 * ->i_mutex on parents, which works but leads to some truely excessive
2417 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2418 struct inode *new_dir, struct dentry *new_dentry)
2421 struct inode *target;
2424 * If we are going to change the parent - check write permissions,
2425 * we'll need to flip '..'.
2427 if (new_dir != old_dir) {
2428 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2433 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2437 target = new_dentry->d_inode;
2439 mutex_lock(&target->i_mutex);
2440 dentry_unhash(new_dentry);
2442 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2445 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2448 target->i_flags |= S_DEAD;
2449 mutex_unlock(&target->i_mutex);
2450 if (d_unhashed(new_dentry))
2451 d_rehash(new_dentry);
2455 d_move(old_dentry,new_dentry);
2459 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2460 struct inode *new_dir, struct dentry *new_dentry)
2462 struct inode *target;
2465 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2470 target = new_dentry->d_inode;
2472 mutex_lock(&target->i_mutex);
2473 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2476 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2478 /* The following d_move() should become unconditional */
2479 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2480 d_move(old_dentry, new_dentry);
2483 mutex_unlock(&target->i_mutex);
2488 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2489 struct inode *new_dir, struct dentry *new_dentry)
2492 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2493 const char *old_name;
2495 if (old_dentry->d_inode == new_dentry->d_inode)
2498 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2502 if (!new_dentry->d_inode)
2503 error = may_create(new_dir, new_dentry, NULL);
2505 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2509 if (!old_dir->i_op || !old_dir->i_op->rename)
2512 DQUOT_INIT(old_dir);
2513 DQUOT_INIT(new_dir);
2515 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2518 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2520 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2522 const char *new_name = old_dentry->d_name.name;
2523 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2524 new_dentry->d_inode, old_dentry->d_inode);
2526 fsnotify_oldname_free(old_name);
2531 static int do_rename(int olddfd, const char *oldname,
2532 int newdfd, const char *newname)
2535 struct dentry * old_dir, * new_dir;
2536 struct dentry * old_dentry, *new_dentry;
2537 struct dentry * trap;
2538 struct nameidata oldnd, newnd;
2540 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2544 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2549 if (oldnd.mnt != newnd.mnt)
2552 old_dir = oldnd.dentry;
2554 if (oldnd.last_type != LAST_NORM)
2557 new_dir = newnd.dentry;
2558 if (newnd.last_type != LAST_NORM)
2561 trap = lock_rename(new_dir, old_dir);
2563 old_dentry = lookup_hash(&oldnd);
2564 error = PTR_ERR(old_dentry);
2565 if (IS_ERR(old_dentry))
2567 /* source must exist */
2569 if (!old_dentry->d_inode)
2571 /* unless the source is a directory trailing slashes give -ENOTDIR */
2572 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2574 if (oldnd.last.name[oldnd.last.len])
2576 if (newnd.last.name[newnd.last.len])
2579 /* source should not be ancestor of target */
2581 if (old_dentry == trap)
2584 if (MNT_IS_RDONLY(newnd.mnt))
2586 new_dentry = lookup_hash(&newnd);
2587 error = PTR_ERR(new_dentry);
2588 if (IS_ERR(new_dentry))
2590 /* target should not be an ancestor of source */
2592 if (new_dentry == trap)
2595 error = vfs_rename(old_dir->d_inode, old_dentry,
2596 new_dir->d_inode, new_dentry);
2602 unlock_rename(new_dir, old_dir);
2604 path_release(&newnd);
2606 path_release(&oldnd);
2611 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2612 int newdfd, const char __user *newname)
2618 from = getname(oldname);
2620 return PTR_ERR(from);
2621 to = getname(newname);
2622 error = PTR_ERR(to);
2624 error = do_rename(olddfd, from, newdfd, to);
2631 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2633 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2636 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2640 len = PTR_ERR(link);
2645 if (len > (unsigned) buflen)
2647 if (copy_to_user(buffer, link, len))
2654 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2655 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2656 * using) it for any given inode is up to filesystem.
2658 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2660 struct nameidata nd;
2664 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2665 if (!IS_ERR(cookie)) {
2666 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2667 if (dentry->d_inode->i_op->put_link)
2668 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2669 cookie = ERR_PTR(res);
2671 return PTR_ERR(cookie);
2674 int vfs_follow_link(struct nameidata *nd, const char *link)
2676 return __vfs_follow_link(nd, link);
2679 /* get the link contents into pagecache */
2680 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2683 struct address_space *mapping = dentry->d_inode->i_mapping;
2684 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2688 wait_on_page_locked(page);
2689 if (!PageUptodate(page))
2695 page_cache_release(page);
2696 return ERR_PTR(-EIO);
2702 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2704 struct page *page = NULL;
2705 char *s = page_getlink(dentry, &page);
2706 int res = vfs_readlink(dentry,buffer,buflen,s);
2709 page_cache_release(page);
2714 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2716 struct page *page = NULL;
2717 nd_set_link(nd, page_getlink(dentry, &page));
2721 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2723 struct page *page = cookie;
2727 page_cache_release(page);
2731 int __page_symlink(struct inode *inode, const char *symname, int len,
2734 struct address_space *mapping = inode->i_mapping;
2740 page = find_or_create_page(mapping, 0, gfp_mask);
2743 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2744 if (err == AOP_TRUNCATED_PAGE) {
2745 page_cache_release(page);
2750 kaddr = kmap_atomic(page, KM_USER0);
2751 memcpy(kaddr, symname, len-1);
2752 kunmap_atomic(kaddr, KM_USER0);
2753 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2754 if (err == AOP_TRUNCATED_PAGE) {
2755 page_cache_release(page);
2761 * Notice that we are _not_ going to block here - end of page is
2762 * unmapped, so this will only try to map the rest of page, see
2763 * that it is unmapped (typically even will not look into inode -
2764 * ->i_size will be enough for everything) and zero it out.
2765 * OTOH it's obviously correct and should make the page up-to-date.
2767 if (!PageUptodate(page)) {
2768 err = mapping->a_ops->readpage(NULL, page);
2769 if (err != AOP_TRUNCATED_PAGE)
2770 wait_on_page_locked(page);
2774 page_cache_release(page);
2777 mark_inode_dirty(inode);
2781 page_cache_release(page);
2786 int page_symlink(struct inode *inode, const char *symname, int len)
2788 return __page_symlink(inode, symname, len,
2789 mapping_gfp_mask(inode->i_mapping));
2792 struct inode_operations page_symlink_inode_operations = {
2793 .readlink = generic_readlink,
2794 .follow_link = page_follow_link_light,
2795 .put_link = page_put_link,
2798 EXPORT_SYMBOL(__user_walk);
2799 EXPORT_SYMBOL(__user_walk_fd);
2800 EXPORT_SYMBOL(follow_down);
2801 EXPORT_SYMBOL(follow_up);
2802 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2803 EXPORT_SYMBOL(getname);
2804 EXPORT_SYMBOL(lock_rename);
2805 EXPORT_SYMBOL(lookup_one_len);
2806 EXPORT_SYMBOL(page_follow_link_light);
2807 EXPORT_SYMBOL(page_put_link);
2808 EXPORT_SYMBOL(page_readlink);
2809 EXPORT_SYMBOL(__page_symlink);
2810 EXPORT_SYMBOL(page_symlink);
2811 EXPORT_SYMBOL(page_symlink_inode_operations);
2812 EXPORT_SYMBOL(path_lookup);
2813 EXPORT_SYMBOL(path_release);
2814 EXPORT_SYMBOL(path_walk);
2815 EXPORT_SYMBOL(permission);
2816 EXPORT_SYMBOL(vfs_permission);
2817 EXPORT_SYMBOL(file_permission);
2818 EXPORT_SYMBOL(unlock_rename);
2819 EXPORT_SYMBOL(vfs_create);
2820 EXPORT_SYMBOL(vfs_follow_link);
2821 EXPORT_SYMBOL(vfs_link);
2822 EXPORT_SYMBOL(vfs_mkdir);
2823 EXPORT_SYMBOL(vfs_mknod);
2824 EXPORT_SYMBOL(generic_permission);
2825 EXPORT_SYMBOL(vfs_readlink);
2826 EXPORT_SYMBOL(vfs_rename);
2827 EXPORT_SYMBOL(vfs_rmdir);
2828 EXPORT_SYMBOL(vfs_symlink);
2829 EXPORT_SYMBOL(vfs_unlink);
2830 EXPORT_SYMBOL(dentry_unhash);
2831 EXPORT_SYMBOL(generic_readlink);