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/vserver/inode.h>
37 #include <linux/vserver/debug.h>
38 #include <asm/namei.h>
39 #include <asm/uaccess.h>
41 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existant name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
121 static int do_getname(const char __user *filename, char *page)
124 unsigned long len = PATH_MAX;
126 if (!segment_eq(get_fs(), KERNEL_DS)) {
127 if ((unsigned long) filename >= TASK_SIZE)
129 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
130 len = TASK_SIZE - (unsigned long) filename;
133 retval = strncpy_from_user(page, filename, len);
137 return -ENAMETOOLONG;
143 char * getname(const char __user * filename)
147 result = ERR_PTR(-ENOMEM);
150 int retval = do_getname(filename, tmp);
155 result = ERR_PTR(retval);
158 audit_getname(result);
162 #ifdef CONFIG_AUDITSYSCALL
163 void putname(const char *name)
165 if (unlikely(current->audit_context))
170 EXPORT_SYMBOL(putname);
175 * generic_permission - check for access rights on a Posix-like filesystem
176 * @inode: inode to check access rights for
177 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
178 * @check_acl: optional callback to check for Posix ACLs
180 * Used to check for read/write/execute permissions on a file.
181 * We use "fsuid" for this, letting us set arbitrary permissions
182 * for filesystem access without changing the "normal" uids which
183 * are used for other things..
185 int generic_permission(struct inode *inode, int mask,
186 int (*check_acl)(struct inode *inode, int mask))
188 umode_t mode = inode->i_mode;
190 if (current->fsuid == inode->i_uid)
193 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
194 int error = check_acl(inode, mask);
195 if (error == -EACCES)
196 goto check_capabilities;
197 else if (error != -EAGAIN)
201 if (in_group_p(inode->i_gid))
206 * If the DACs are ok we don't need any capability check.
208 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
213 * Read/write DACs are always overridable.
214 * Executable DACs are overridable if at least one exec bit is set.
216 if (!(mask & MAY_EXEC) ||
217 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
218 if (capable(CAP_DAC_OVERRIDE))
222 * Searching includes executable on directories, else just read.
224 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
225 if (capable(CAP_DAC_READ_SEARCH))
231 static inline int vx_barrier(struct inode *inode)
233 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN|VX_IDENT)) {
234 vxwprintk(1, "xid=%d did hit the barrier.",
241 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
243 if (vx_barrier(inode))
245 if (inode->i_xid == 0)
247 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
250 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
251 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
256 int permission(struct inode *inode, int mask, struct nameidata *nd)
260 if (mask & MAY_WRITE) {
261 umode_t mode = inode->i_mode;
264 * Nobody gets write access to a read-only fs.
266 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
267 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
271 * Nobody gets write access to an immutable file.
273 if (IS_IMMUTABLE(inode))
278 /* Ordinary permission routines do not understand MAY_APPEND. */
279 submask = mask & ~MAY_APPEND;
280 if ((retval = xid_permission(inode, mask, nd)))
282 if (inode->i_op && inode->i_op->permission)
283 retval = inode->i_op->permission(inode, submask, nd);
285 retval = generic_permission(inode, submask, NULL);
289 return security_inode_permission(inode, mask, nd);
293 * vfs_permission - check for access rights to a given path
294 * @nd: lookup result that describes the path
295 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
297 * Used to check for read/write/execute permissions on a path.
298 * We use "fsuid" for this, letting us set arbitrary permissions
299 * for filesystem access without changing the "normal" uids which
300 * are used for other things.
302 int vfs_permission(struct nameidata *nd, int mask)
304 return permission(nd->dentry->d_inode, mask, nd);
308 * file_permission - check for additional access rights to a given file
309 * @file: file to check access rights for
310 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
312 * Used to check for read/write/execute permissions on an already opened
316 * Do not use this function in new code. All access checks should
317 * be done using vfs_permission().
319 int file_permission(struct file *file, int mask)
321 return permission(file->f_dentry->d_inode, mask, NULL);
325 * get_write_access() gets write permission for a file.
326 * put_write_access() releases this write permission.
327 * This is used for regular files.
328 * We cannot support write (and maybe mmap read-write shared) accesses and
329 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
330 * can have the following values:
331 * 0: no writers, no VM_DENYWRITE mappings
332 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
333 * > 0: (i_writecount) users are writing to the file.
335 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
336 * except for the cases where we don't hold i_writecount yet. Then we need to
337 * use {get,deny}_write_access() - these functions check the sign and refuse
338 * to do the change if sign is wrong. Exclusion between them is provided by
339 * the inode->i_lock spinlock.
342 int get_write_access(struct inode * inode)
344 spin_lock(&inode->i_lock);
345 if (atomic_read(&inode->i_writecount) < 0) {
346 spin_unlock(&inode->i_lock);
349 atomic_inc(&inode->i_writecount);
350 spin_unlock(&inode->i_lock);
355 int deny_write_access(struct file * file)
357 struct inode *inode = file->f_dentry->d_inode;
359 spin_lock(&inode->i_lock);
360 if (atomic_read(&inode->i_writecount) > 0) {
361 spin_unlock(&inode->i_lock);
364 atomic_dec(&inode->i_writecount);
365 spin_unlock(&inode->i_lock);
370 void path_release(struct nameidata *nd)
377 * umount() mustn't call path_release()/mntput() as that would clear
380 void path_release_on_umount(struct nameidata *nd)
383 mntput_no_expire(nd->mnt);
387 * release_open_intent - free up open intent resources
388 * @nd: pointer to nameidata
390 void release_open_intent(struct nameidata *nd)
392 if (nd->intent.open.file->f_dentry == NULL)
393 put_filp(nd->intent.open.file);
395 fput(nd->intent.open.file);
399 * Internal lookup() using the new generic dcache.
402 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
404 struct dentry * dentry = __d_lookup(parent, name);
406 /* lockess __d_lookup may fail due to concurrent d_move()
407 * in some unrelated directory, so try with d_lookup
410 dentry = d_lookup(parent, name);
412 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
413 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
431 static int exec_permission_lite(struct inode *inode,
432 struct nameidata *nd)
434 umode_t mode = inode->i_mode;
436 if (vx_barrier(inode))
438 if (inode->i_op && inode->i_op->permission)
441 if (current->fsuid == inode->i_uid)
443 else if (in_group_p(inode->i_gid))
449 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
452 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
455 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
460 return security_inode_permission(inode, MAY_EXEC, nd);
464 * This is called when everything else fails, and we actually have
465 * to go to the low-level filesystem to find out what we should do..
467 * We get the directory semaphore, and after getting that we also
468 * make sure that nobody added the entry to the dcache in the meantime..
471 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
473 struct dentry * result;
474 struct inode *dir = parent->d_inode;
476 mutex_lock(&dir->i_mutex);
478 * First re-do the cached lookup just in case it was created
479 * while we waited for the directory semaphore..
481 * FIXME! This could use version numbering or similar to
482 * avoid unnecessary cache lookups.
484 * The "dcache_lock" is purely to protect the RCU list walker
485 * from concurrent renames at this point (we mustn't get false
486 * negatives from the RCU list walk here, unlike the optimistic
489 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
491 result = d_lookup(parent, name);
493 struct dentry * dentry = d_alloc(parent, name);
494 result = ERR_PTR(-ENOMEM);
496 result = dir->i_op->lookup(dir, dentry, nd);
502 mutex_unlock(&dir->i_mutex);
507 * Uhhuh! Nasty case: the cache was re-populated while
508 * we waited on the semaphore. Need to revalidate.
510 mutex_unlock(&dir->i_mutex);
511 if (result->d_op && result->d_op->d_revalidate) {
512 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
514 result = ERR_PTR(-ENOENT);
520 static int __emul_lookup_dentry(const char *, struct nameidata *);
523 static __always_inline int
524 walk_init_root(const char *name, struct nameidata *nd)
526 read_lock(¤t->fs->lock);
527 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
528 nd->mnt = mntget(current->fs->altrootmnt);
529 nd->dentry = dget(current->fs->altroot);
530 read_unlock(¤t->fs->lock);
531 if (__emul_lookup_dentry(name,nd))
533 read_lock(¤t->fs->lock);
535 nd->mnt = mntget(current->fs->rootmnt);
536 nd->dentry = dget(current->fs->root);
537 read_unlock(¤t->fs->lock);
541 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
550 if (!walk_init_root(link, nd))
551 /* weird __emul_prefix() stuff did it */
554 res = link_path_walk(link, nd);
556 if (nd->depth || res || nd->last_type!=LAST_NORM)
559 * If it is an iterative symlinks resolution in open_namei() we
560 * have to copy the last component. And all that crap because of
561 * bloody create() on broken symlinks. Furrfu...
564 if (unlikely(!name)) {
568 strcpy(name, nd->last.name);
569 nd->last.name = name;
573 return PTR_ERR(link);
577 struct vfsmount *mnt;
578 struct dentry *dentry;
581 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
585 struct dentry *dentry = path->dentry;
587 touch_atime(path->mnt, dentry);
588 nd_set_link(nd, NULL);
590 if (path->mnt == nd->mnt)
592 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
593 error = PTR_ERR(cookie);
594 if (!IS_ERR(cookie)) {
595 char *s = nd_get_link(nd);
598 error = __vfs_follow_link(nd, s);
599 if (dentry->d_inode->i_op->put_link)
600 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
608 static inline void dput_path(struct path *path, struct nameidata *nd)
611 if (path->mnt != nd->mnt)
615 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
618 if (nd->mnt != path->mnt)
621 nd->dentry = path->dentry;
625 * This limits recursive symlink follows to 8, while
626 * limiting consecutive symlinks to 40.
628 * Without that kind of total limit, nasty chains of consecutive
629 * symlinks can cause almost arbitrarily long lookups.
631 static inline int do_follow_link(struct path *path, struct nameidata *nd)
634 if (current->link_count >= MAX_NESTED_LINKS)
636 if (current->total_link_count >= 40)
638 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
640 err = security_inode_follow_link(path->dentry, nd);
643 current->link_count++;
644 current->total_link_count++;
646 err = __do_follow_link(path, nd);
647 current->link_count--;
656 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
658 struct vfsmount *parent;
659 struct dentry *mountpoint;
660 spin_lock(&vfsmount_lock);
661 parent=(*mnt)->mnt_parent;
662 if (parent == *mnt) {
663 spin_unlock(&vfsmount_lock);
667 mountpoint=dget((*mnt)->mnt_mountpoint);
668 spin_unlock(&vfsmount_lock);
670 *dentry = mountpoint;
676 /* no need for dcache_lock, as serialization is taken care in
679 static int __follow_mount(struct path *path)
682 while (d_mountpoint(path->dentry)) {
683 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
690 path->dentry = dget(mounted->mnt_root);
696 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
698 while (d_mountpoint(*dentry)) {
699 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
705 *dentry = dget(mounted->mnt_root);
709 /* no need for dcache_lock, as serialization is taken care in
712 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
714 struct vfsmount *mounted;
716 mounted = lookup_mnt(*mnt, *dentry);
721 *dentry = dget(mounted->mnt_root);
727 static __always_inline void follow_dotdot(struct nameidata *nd)
730 struct vfsmount *parent;
731 struct dentry *old = nd->dentry;
733 read_lock(¤t->fs->lock);
734 if (nd->dentry == current->fs->root &&
735 nd->mnt == current->fs->rootmnt) {
736 read_unlock(¤t->fs->lock);
737 /* for sane '/' avoid follow_mount() */
740 read_unlock(¤t->fs->lock);
741 spin_lock(&dcache_lock);
742 if (nd->dentry != nd->mnt->mnt_root) {
743 nd->dentry = dget(nd->dentry->d_parent);
744 spin_unlock(&dcache_lock);
748 spin_unlock(&dcache_lock);
749 spin_lock(&vfsmount_lock);
750 parent = nd->mnt->mnt_parent;
751 if (parent == nd->mnt) {
752 spin_unlock(&vfsmount_lock);
756 nd->dentry = dget(nd->mnt->mnt_mountpoint);
757 spin_unlock(&vfsmount_lock);
762 follow_mount(&nd->mnt, &nd->dentry);
766 * It's more convoluted than I'd like it to be, but... it's still fairly
767 * small and for now I'd prefer to have fast path as straight as possible.
768 * It _is_ time-critical.
770 static int do_lookup(struct nameidata *nd, struct qstr *name,
773 struct vfsmount *mnt = nd->mnt;
774 struct dentry *dentry = __d_lookup(nd->dentry, name);
779 if (dentry->d_op && dentry->d_op->d_revalidate)
780 goto need_revalidate;
781 inode = dentry->d_inode;
784 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
786 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
787 struct proc_dir_entry *de = PDE(inode);
789 if (de && !vx_hide_check(0, de->vx_flags))
794 path->dentry = dentry;
795 __follow_mount(path);
798 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
799 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
800 vxd_path(dentry, mnt));
805 dentry = real_lookup(nd->dentry, name, nd);
811 if (dentry->d_op->d_revalidate(dentry, nd))
813 if (d_invalidate(dentry))
819 return PTR_ERR(dentry);
824 * This is the basic name resolution function, turning a pathname into
825 * the final dentry. We expect 'base' to be positive and a directory.
827 * Returns 0 and nd will have valid dentry and mnt on success.
828 * Returns error and drops reference to input namei data on failure.
830 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
835 unsigned int lookup_flags = nd->flags;
842 inode = nd->dentry->d_inode;
844 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
846 /* At this point we know we have a real path component. */
852 nd->flags |= LOOKUP_CONTINUE;
853 err = exec_permission_lite(inode, nd);
855 err = vfs_permission(nd, MAY_EXEC);
860 c = *(const unsigned char *)name;
862 hash = init_name_hash();
865 hash = partial_name_hash(c, hash);
866 c = *(const unsigned char *)name;
867 } while (c && (c != '/'));
868 this.len = name - (const char *) this.name;
869 this.hash = end_name_hash(hash);
871 /* remove trailing slashes? */
874 while (*++name == '/');
876 goto last_with_slashes;
879 * "." and ".." are special - ".." especially so because it has
880 * to be able to know about the current root directory and
881 * parent relationships.
883 if (this.name[0] == '.') switch (this.len) {
887 if (this.name[1] != '.')
890 inode = nd->dentry->d_inode;
896 * See if the low-level filesystem might want
897 * to use its own hash..
899 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
900 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
904 /* This does the actual lookups.. */
905 err = do_lookup(nd, &this, &next);
910 inode = next.dentry->d_inode;
917 if (inode->i_op->follow_link) {
918 err = do_follow_link(&next, nd);
922 inode = nd->dentry->d_inode;
929 path_to_nameidata(&next, nd);
931 if (!inode->i_op->lookup)
934 /* here ends the main loop */
937 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
939 /* Clear LOOKUP_CONTINUE iff it was previously unset */
940 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
941 if (lookup_flags & LOOKUP_PARENT)
943 if (this.name[0] == '.') switch (this.len) {
947 if (this.name[1] != '.')
950 inode = nd->dentry->d_inode;
955 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
956 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
960 err = do_lookup(nd, &this, &next);
963 inode = next.dentry->d_inode;
964 if ((lookup_flags & LOOKUP_FOLLOW)
965 && inode && inode->i_op && inode->i_op->follow_link) {
966 err = do_follow_link(&next, nd);
969 inode = nd->dentry->d_inode;
971 path_to_nameidata(&next, nd);
975 if (lookup_flags & LOOKUP_DIRECTORY) {
977 if (!inode->i_op || !inode->i_op->lookup)
983 nd->last_type = LAST_NORM;
984 if (this.name[0] != '.')
987 nd->last_type = LAST_DOT;
988 else if (this.len == 2 && this.name[1] == '.')
989 nd->last_type = LAST_DOTDOT;
994 * We bypassed the ordinary revalidation routines.
995 * We may need to check the cached dentry for staleness.
997 if (nd->dentry && nd->dentry->d_sb &&
998 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1000 /* Note: we do not d_invalidate() */
1001 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
1007 dput_path(&next, nd);
1016 * Wrapper to retry pathname resolution whenever the underlying
1017 * file system returns an ESTALE.
1019 * Retry the whole path once, forcing real lookup requests
1020 * instead of relying on the dcache.
1022 int fastcall link_path_walk(const char *name, struct nameidata *nd)
1024 struct nameidata save = *nd;
1027 /* make sure the stuff we saved doesn't go away */
1031 result = __link_path_walk(name, nd);
1032 if (result == -ESTALE) {
1036 nd->flags |= LOOKUP_REVAL;
1037 result = __link_path_walk(name, nd);
1046 int fastcall path_walk(const char * name, struct nameidata *nd)
1048 current->total_link_count = 0;
1049 return link_path_walk(name, nd);
1053 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1054 * everything is done. Returns 0 and drops input nd, if lookup failed;
1056 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1058 if (path_walk(name, nd))
1059 return 0; /* something went wrong... */
1061 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1062 struct dentry *old_dentry = nd->dentry;
1063 struct vfsmount *old_mnt = nd->mnt;
1064 struct qstr last = nd->last;
1065 int last_type = nd->last_type;
1067 * NAME was not found in alternate root or it's a directory. Try to find
1068 * it in the normal root:
1070 nd->last_type = LAST_ROOT;
1071 read_lock(¤t->fs->lock);
1072 nd->mnt = mntget(current->fs->rootmnt);
1073 nd->dentry = dget(current->fs->root);
1074 read_unlock(¤t->fs->lock);
1075 if (path_walk(name, nd) == 0) {
1076 if (nd->dentry->d_inode) {
1083 nd->dentry = old_dentry;
1086 nd->last_type = last_type;
1091 void set_fs_altroot(void)
1093 char *emul = __emul_prefix();
1094 struct nameidata nd;
1095 struct vfsmount *mnt = NULL, *oldmnt;
1096 struct dentry *dentry = NULL, *olddentry;
1101 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1107 write_lock(¤t->fs->lock);
1108 oldmnt = current->fs->altrootmnt;
1109 olddentry = current->fs->altroot;
1110 current->fs->altrootmnt = mnt;
1111 current->fs->altroot = dentry;
1112 write_unlock(¤t->fs->lock);
1119 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1120 static int fastcall do_path_lookup(int dfd, const char *name,
1121 unsigned int flags, struct nameidata *nd)
1127 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1132 read_lock(¤t->fs->lock);
1133 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1134 nd->mnt = mntget(current->fs->altrootmnt);
1135 nd->dentry = dget(current->fs->altroot);
1136 read_unlock(¤t->fs->lock);
1137 if (__emul_lookup_dentry(name,nd))
1138 goto out; /* found in altroot */
1139 read_lock(¤t->fs->lock);
1141 nd->mnt = mntget(current->fs->rootmnt);
1142 nd->dentry = dget(current->fs->root);
1143 read_unlock(¤t->fs->lock);
1144 } else if (dfd == AT_FDCWD) {
1145 read_lock(¤t->fs->lock);
1146 nd->mnt = mntget(current->fs->pwdmnt);
1147 nd->dentry = dget(current->fs->pwd);
1148 read_unlock(¤t->fs->lock);
1150 struct dentry *dentry;
1152 file = fget_light(dfd, &fput_needed);
1157 dentry = file->f_dentry;
1160 if (!S_ISDIR(dentry->d_inode->i_mode))
1163 retval = file_permission(file, MAY_EXEC);
1167 nd->mnt = mntget(file->f_vfsmnt);
1168 nd->dentry = dget(dentry);
1170 fput_light(file, fput_needed);
1172 current->total_link_count = 0;
1173 retval = link_path_walk(name, nd);
1175 if (likely(retval == 0)) {
1176 if (unlikely(current->audit_context && nd && nd->dentry &&
1177 nd->dentry->d_inode))
1178 audit_inode(name, nd->dentry->d_inode, flags);
1184 fput_light(file, fput_needed);
1188 int fastcall path_lookup(const char *name, unsigned int flags,
1189 struct nameidata *nd)
1191 return do_path_lookup(AT_FDCWD, name, flags, nd);
1194 static int __path_lookup_intent_open(int dfd, const char *name,
1195 unsigned int lookup_flags, struct nameidata *nd,
1196 int open_flags, int create_mode)
1198 struct file *filp = get_empty_filp();
1203 nd->intent.open.file = filp;
1204 nd->intent.open.flags = open_flags;
1205 nd->intent.open.create_mode = create_mode;
1206 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1207 if (IS_ERR(nd->intent.open.file)) {
1209 err = PTR_ERR(nd->intent.open.file);
1212 } else if (err != 0)
1213 release_open_intent(nd);
1218 * path_lookup_open - lookup a file path with open intent
1219 * @dfd: the directory to use as base, or AT_FDCWD
1220 * @name: pointer to file name
1221 * @lookup_flags: lookup intent flags
1222 * @nd: pointer to nameidata
1223 * @open_flags: open intent flags
1225 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1226 struct nameidata *nd, int open_flags)
1228 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1233 * path_lookup_create - lookup a file path with open + create intent
1234 * @dfd: the directory to use as base, or AT_FDCWD
1235 * @name: pointer to file name
1236 * @lookup_flags: lookup intent flags
1237 * @nd: pointer to nameidata
1238 * @open_flags: open intent flags
1239 * @create_mode: create intent flags
1241 static int path_lookup_create(int dfd, const char *name,
1242 unsigned int lookup_flags, struct nameidata *nd,
1243 int open_flags, int create_mode)
1245 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1246 nd, open_flags, create_mode);
1249 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1250 struct nameidata *nd, int open_flags)
1252 char *tmp = getname(name);
1253 int err = PTR_ERR(tmp);
1256 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1263 * Restricted form of lookup. Doesn't follow links, single-component only,
1264 * needs parent already locked. Doesn't follow mounts.
1267 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1269 struct dentry * dentry;
1270 struct inode *inode;
1273 inode = base->d_inode;
1274 err = permission(inode, MAY_EXEC, nd);
1275 dentry = ERR_PTR(err);
1280 * See if the low-level filesystem might want
1281 * to use its own hash..
1283 if (base->d_op && base->d_op->d_hash) {
1284 err = base->d_op->d_hash(base, name);
1285 dentry = ERR_PTR(err);
1290 dentry = cached_lookup(base, name, nd);
1292 struct dentry *new = d_alloc(base, name);
1293 dentry = ERR_PTR(-ENOMEM);
1296 dentry = inode->i_op->lookup(inode, new, nd);
1306 struct dentry * lookup_hash(struct nameidata *nd)
1308 return __lookup_hash(&nd->last, nd->dentry, nd);
1312 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1323 hash = init_name_hash();
1325 c = *(const unsigned char *)name++;
1326 if (c == '/' || c == '\0')
1328 hash = partial_name_hash(c, hash);
1330 this.hash = end_name_hash(hash);
1332 return __lookup_hash(&this, base, NULL);
1334 return ERR_PTR(-EACCES);
1340 * is used by most simple commands to get the inode of a specified name.
1341 * Open, link etc use their own routines, but this is enough for things
1344 * namei exists in two versions: namei/lnamei. The only difference is
1345 * that namei follows links, while lnamei does not.
1348 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1349 struct nameidata *nd)
1351 char *tmp = getname(name);
1352 int err = PTR_ERR(tmp);
1355 err = do_path_lookup(dfd, tmp, flags, nd);
1361 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1363 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1367 * It's inline, so penalty for filesystems that don't use sticky bit is
1370 static inline int check_sticky(struct inode *dir, struct inode *inode)
1372 if (!(dir->i_mode & S_ISVTX))
1374 if (inode->i_uid == current->fsuid)
1376 if (dir->i_uid == current->fsuid)
1378 return !capable(CAP_FOWNER);
1382 * Check whether we can remove a link victim from directory dir, check
1383 * whether the type of victim is right.
1384 * 1. We can't do it if dir is read-only (done in permission())
1385 * 2. We should have write and exec permissions on dir
1386 * 3. We can't remove anything from append-only dir
1387 * 4. We can't do anything with immutable dir (done in permission())
1388 * 5. If the sticky bit on dir is set we should either
1389 * a. be owner of dir, or
1390 * b. be owner of victim, or
1391 * c. have CAP_FOWNER capability
1392 * 6. If the victim is append-only or immutable we can't do antyhing with
1393 * links pointing to it.
1394 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1395 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1396 * 9. We can't remove a root or mountpoint.
1397 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1398 * nfs_async_unlink().
1400 static int may_delete(struct inode *dir, struct dentry *victim,
1401 int isdir, struct nameidata *nd)
1405 if (!victim->d_inode)
1408 BUG_ON(victim->d_parent->d_inode != dir);
1410 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1415 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1416 IS_IXORUNLINK(victim->d_inode))
1419 if (!S_ISDIR(victim->d_inode->i_mode))
1421 if (IS_ROOT(victim))
1423 } else if (S_ISDIR(victim->d_inode->i_mode))
1425 if (IS_DEADDIR(dir))
1427 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1432 /* Check whether we can create an object with dentry child in directory
1434 * 1. We can't do it if child already exists (open has special treatment for
1435 * this case, but since we are inlined it's OK)
1436 * 2. We can't do it if dir is read-only (done in permission())
1437 * 3. We should have write and exec permissions on dir
1438 * 4. We can't do it if dir is immutable (done in permission())
1440 static inline int may_create(struct inode *dir, struct dentry *child,
1441 struct nameidata *nd)
1445 if (IS_DEADDIR(dir))
1447 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1451 * O_DIRECTORY translates into forcing a directory lookup.
1453 static inline int lookup_flags(unsigned int f)
1455 unsigned long retval = LOOKUP_FOLLOW;
1458 retval &= ~LOOKUP_FOLLOW;
1460 if (f & O_DIRECTORY)
1461 retval |= LOOKUP_DIRECTORY;
1467 * p1 and p2 should be directories on the same fs.
1469 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1474 mutex_lock(&p1->d_inode->i_mutex);
1478 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1480 for (p = p1; p->d_parent != p; p = p->d_parent) {
1481 if (p->d_parent == p2) {
1482 mutex_lock(&p2->d_inode->i_mutex);
1483 mutex_lock(&p1->d_inode->i_mutex);
1488 for (p = p2; p->d_parent != p; p = p->d_parent) {
1489 if (p->d_parent == p1) {
1490 mutex_lock(&p1->d_inode->i_mutex);
1491 mutex_lock(&p2->d_inode->i_mutex);
1496 mutex_lock(&p1->d_inode->i_mutex);
1497 mutex_lock(&p2->d_inode->i_mutex);
1501 void unlock_rename(struct dentry *p1, struct dentry *p2)
1503 mutex_unlock(&p1->d_inode->i_mutex);
1505 mutex_unlock(&p2->d_inode->i_mutex);
1506 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1510 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1511 struct nameidata *nd)
1513 int error = may_create(dir, dentry, nd);
1518 if (!dir->i_op || !dir->i_op->create)
1519 return -EACCES; /* shouldn't it be ENOSYS? */
1522 error = security_inode_create(dir, dentry, mode);
1526 error = dir->i_op->create(dir, dentry, mode, nd);
1528 fsnotify_create(dir, dentry->d_name.name);
1532 int may_open(struct nameidata *nd, int acc_mode, int flag)
1534 struct dentry *dentry = nd->dentry;
1535 struct inode *inode = dentry->d_inode;
1541 if (S_ISLNK(inode->i_mode))
1544 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1547 error = vfs_permission(nd, acc_mode);
1552 * FIFO's, sockets and device files are special: they don't
1553 * actually live on the filesystem itself, and as such you
1554 * can write to them even if the filesystem is read-only.
1556 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1558 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1559 if (nd->mnt->mnt_flags & MNT_NODEV)
1563 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1564 && (flag & FMODE_WRITE))
1567 * An append-only file must be opened in append mode for writing.
1569 if (IS_APPEND(inode)) {
1570 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1576 /* O_NOATIME can only be set by the owner or superuser */
1577 if (flag & O_NOATIME)
1578 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1582 * Ensure there are no outstanding leases on the file.
1584 error = break_lease(inode, flag);
1588 if (flag & O_TRUNC) {
1589 error = get_write_access(inode);
1594 * Refuse to truncate files with mandatory locks held on them.
1596 error = locks_verify_locked(inode);
1600 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1602 put_write_access(inode);
1606 if (flag & FMODE_WRITE)
1615 * namei for open - this is in fact almost the whole open-routine.
1617 * Note that the low bits of "flag" aren't the same as in the open
1618 * system call - they are 00 - no permissions needed
1619 * 01 - read permission needed
1620 * 10 - write permission needed
1621 * 11 - read/write permissions needed
1622 * which is a lot more logical, and also allows the "no perm" needed
1623 * for symlinks (where the permissions are checked later).
1626 int open_namei(int dfd, const char *pathname, int flag,
1627 int mode, struct nameidata *nd)
1629 int acc_mode, error;
1634 acc_mode = ACC_MODE(flag);
1636 /* O_TRUNC implies we need access checks for write permissions */
1638 acc_mode |= MAY_WRITE;
1640 /* Allow the LSM permission hook to distinguish append
1641 access from general write access. */
1642 if (flag & O_APPEND)
1643 acc_mode |= MAY_APPEND;
1646 * The simplest case - just a plain lookup.
1648 if (!(flag & O_CREAT)) {
1649 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1657 * Create - we need to know the parent.
1659 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1664 * We have the parent and last component. First of all, check
1665 * that we are not asked to creat(2) an obvious directory - that
1669 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1673 nd->flags &= ~LOOKUP_PARENT;
1674 mutex_lock(&dir->d_inode->i_mutex);
1675 path.dentry = lookup_hash(nd);
1679 error = PTR_ERR(path.dentry);
1680 if (IS_ERR(path.dentry)) {
1681 mutex_unlock(&dir->d_inode->i_mutex);
1685 if (IS_ERR(nd->intent.open.file)) {
1686 mutex_unlock(&dir->d_inode->i_mutex);
1687 error = PTR_ERR(nd->intent.open.file);
1691 /* Negative dentry, just create the file */
1692 if (!path.dentry->d_inode) {
1693 if (!IS_POSIXACL(dir->d_inode))
1694 mode &= ~current->fs->umask;
1695 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1696 mutex_unlock(&dir->d_inode->i_mutex);
1698 nd->dentry = path.dentry;
1701 /* Don't check for write permission, don't truncate */
1708 * It already exists.
1710 mutex_unlock(&dir->d_inode->i_mutex);
1716 if (__follow_mount(&path)) {
1718 if (flag & O_NOFOLLOW)
1722 if (!path.dentry->d_inode)
1724 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1727 path_to_nameidata(&path, nd);
1729 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1732 error = may_open(nd, acc_mode, flag);
1738 dput_path(&path, nd);
1740 if (!IS_ERR(nd->intent.open.file))
1741 release_open_intent(nd);
1747 if (flag & O_NOFOLLOW)
1750 * This is subtle. Instead of calling do_follow_link() we do the
1751 * thing by hands. The reason is that this way we have zero link_count
1752 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1753 * After that we have the parent and last component, i.e.
1754 * we are in the same situation as after the first path_walk().
1755 * Well, almost - if the last component is normal we get its copy
1756 * stored in nd->last.name and we will have to putname() it when we
1757 * are done. Procfs-like symlinks just set LAST_BIND.
1759 nd->flags |= LOOKUP_PARENT;
1760 error = security_inode_follow_link(path.dentry, nd);
1763 error = __do_follow_link(&path, nd);
1765 /* Does someone understand code flow here? Or it is only
1766 * me so stupid? Anathema to whoever designed this non-sense
1767 * with "intent.open".
1769 release_open_intent(nd);
1772 nd->flags &= ~LOOKUP_PARENT;
1773 if (nd->last_type == LAST_BIND)
1776 if (nd->last_type != LAST_NORM)
1778 if (nd->last.name[nd->last.len]) {
1779 __putname(nd->last.name);
1784 __putname(nd->last.name);
1788 mutex_lock(&dir->d_inode->i_mutex);
1789 path.dentry = lookup_hash(nd);
1791 __putname(nd->last.name);
1796 * lookup_create - lookup a dentry, creating it if it doesn't exist
1797 * @nd: nameidata info
1798 * @is_dir: directory flag
1800 * Simple function to lookup and return a dentry and create it
1801 * if it doesn't exist. Is SMP-safe.
1803 * Returns with nd->dentry->d_inode->i_mutex locked.
1805 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1807 struct dentry *dentry = ERR_PTR(-EEXIST);
1809 mutex_lock(&nd->dentry->d_inode->i_mutex);
1811 * Yucky last component or no last component at all?
1812 * (foo/., foo/.., /////)
1814 if (nd->last_type != LAST_NORM)
1816 nd->flags &= ~LOOKUP_PARENT;
1819 * Do the final lookup.
1821 dentry = lookup_hash(nd);
1826 * Special case - lookup gave negative, but... we had foo/bar/
1827 * From the vfs_mknod() POV we just have a negative dentry -
1828 * all is fine. Let's be bastards - you had / on the end, you've
1829 * been asking for (non-existent) directory. -ENOENT for you.
1831 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1836 dentry = ERR_PTR(-ENOENT);
1840 EXPORT_SYMBOL_GPL(lookup_create);
1842 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1843 int mode, dev_t dev, struct nameidata *nd)
1845 int error = may_create(dir, dentry, nd);
1850 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1853 if (!dir->i_op || !dir->i_op->mknod)
1856 error = security_inode_mknod(dir, dentry, mode, dev);
1861 error = dir->i_op->mknod(dir, dentry, mode, dev);
1863 fsnotify_create(dir, dentry->d_name.name);
1867 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1872 struct dentry * dentry;
1873 struct nameidata nd;
1877 tmp = getname(filename);
1879 return PTR_ERR(tmp);
1881 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1884 dentry = lookup_create(&nd, 0);
1885 error = PTR_ERR(dentry);
1887 if (!IS_POSIXACL(nd.dentry->d_inode))
1888 mode &= ~current->fs->umask;
1889 if (!IS_ERR(dentry)) {
1890 switch (mode & S_IFMT) {
1891 case 0: case S_IFREG:
1892 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1894 case S_IFCHR: case S_IFBLK:
1895 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1896 new_decode_dev(dev), &nd);
1898 case S_IFIFO: case S_IFSOCK:
1899 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1910 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1918 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1920 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1923 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1924 int mode, struct nameidata *nd)
1926 int error = may_create(dir, dentry, nd);
1931 if (!dir->i_op || !dir->i_op->mkdir)
1934 mode &= (S_IRWXUGO|S_ISVTX);
1935 error = security_inode_mkdir(dir, dentry, mode);
1940 error = dir->i_op->mkdir(dir, dentry, mode);
1942 fsnotify_mkdir(dir, dentry->d_name.name);
1946 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
1951 tmp = getname(pathname);
1952 error = PTR_ERR(tmp);
1954 struct dentry *dentry;
1955 struct nameidata nd;
1957 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1960 dentry = lookup_create(&nd, 1);
1961 error = PTR_ERR(dentry);
1962 if (!IS_ERR(dentry)) {
1963 if (!IS_POSIXACL(nd.dentry->d_inode))
1964 mode &= ~current->fs->umask;
1965 error = vfs_mkdir(nd.dentry->d_inode, dentry,
1969 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1978 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1980 return sys_mkdirat(AT_FDCWD, pathname, mode);
1984 * We try to drop the dentry early: we should have
1985 * a usage count of 2 if we're the only user of this
1986 * dentry, and if that is true (possibly after pruning
1987 * the dcache), then we drop the dentry now.
1989 * A low-level filesystem can, if it choses, legally
1992 * if (!d_unhashed(dentry))
1995 * if it cannot handle the case of removing a directory
1996 * that is still in use by something else..
1998 void dentry_unhash(struct dentry *dentry)
2001 if (atomic_read(&dentry->d_count))
2002 shrink_dcache_parent(dentry);
2003 spin_lock(&dcache_lock);
2004 spin_lock(&dentry->d_lock);
2005 if (atomic_read(&dentry->d_count) == 2)
2007 spin_unlock(&dentry->d_lock);
2008 spin_unlock(&dcache_lock);
2011 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
2012 struct nameidata *nd)
2014 int error = may_delete(dir, dentry, 1, nd);
2019 if (!dir->i_op || !dir->i_op->rmdir)
2024 mutex_lock(&dentry->d_inode->i_mutex);
2025 dentry_unhash(dentry);
2026 if (d_mountpoint(dentry))
2029 error = security_inode_rmdir(dir, dentry);
2031 error = dir->i_op->rmdir(dir, dentry);
2033 dentry->d_inode->i_flags |= S_DEAD;
2036 mutex_unlock(&dentry->d_inode->i_mutex);
2045 static long do_rmdir(int dfd, const char __user *pathname)
2049 struct dentry *dentry;
2050 struct nameidata nd;
2052 name = getname(pathname);
2054 return PTR_ERR(name);
2056 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2060 switch(nd.last_type) {
2071 mutex_lock(&nd.dentry->d_inode->i_mutex);
2072 dentry = lookup_hash(&nd);
2073 error = PTR_ERR(dentry);
2074 if (!IS_ERR(dentry)) {
2075 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
2078 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2086 asmlinkage long sys_rmdir(const char __user *pathname)
2088 return do_rmdir(AT_FDCWD, pathname);
2091 int vfs_unlink(struct inode *dir, struct dentry *dentry,
2092 struct nameidata *nd)
2094 int error = may_delete(dir, dentry, 0, nd);
2099 if (!dir->i_op || !dir->i_op->unlink)
2104 mutex_lock(&dentry->d_inode->i_mutex);
2105 if (d_mountpoint(dentry))
2108 error = security_inode_unlink(dir, dentry);
2110 error = dir->i_op->unlink(dir, dentry);
2112 mutex_unlock(&dentry->d_inode->i_mutex);
2114 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2115 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2123 * Make sure that the actual truncation of the file will occur outside its
2124 * directory's i_mutex. Truncate can take a long time if there is a lot of
2125 * writeout happening, and we don't want to prevent access to the directory
2126 * while waiting on the I/O.
2128 static long do_unlinkat(int dfd, const char __user *pathname)
2132 struct dentry *dentry;
2133 struct nameidata nd;
2134 struct inode *inode = NULL;
2136 name = getname(pathname);
2138 return PTR_ERR(name);
2140 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2144 if (nd.last_type != LAST_NORM)
2146 mutex_lock(&nd.dentry->d_inode->i_mutex);
2147 dentry = lookup_hash(&nd);
2148 error = PTR_ERR(dentry);
2149 if (!IS_ERR(dentry)) {
2150 /* Why not before? Because we want correct error value */
2151 if (nd.last.name[nd.last.len])
2153 inode = dentry->d_inode;
2155 atomic_inc(&inode->i_count);
2156 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
2160 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2162 iput(inode); /* truncate the inode here */
2170 error = !dentry->d_inode ? -ENOENT :
2171 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2175 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2177 if ((flag & ~AT_REMOVEDIR) != 0)
2180 if (flag & AT_REMOVEDIR)
2181 return do_rmdir(dfd, pathname);
2183 return do_unlinkat(dfd, pathname);
2186 asmlinkage long sys_unlink(const char __user *pathname)
2188 return do_unlinkat(AT_FDCWD, pathname);
2191 int vfs_symlink(struct inode *dir, struct dentry *dentry,
2192 const char *oldname, int mode, struct nameidata *nd)
2194 int error = may_create(dir, dentry, nd);
2199 if (!dir->i_op || !dir->i_op->symlink)
2202 error = security_inode_symlink(dir, dentry, oldname);
2207 error = dir->i_op->symlink(dir, dentry, oldname);
2209 fsnotify_create(dir, dentry->d_name.name);
2213 asmlinkage long sys_symlinkat(const char __user *oldname,
2214 int newdfd, const char __user *newname)
2220 from = getname(oldname);
2222 return PTR_ERR(from);
2223 to = getname(newname);
2224 error = PTR_ERR(to);
2226 struct dentry *dentry;
2227 struct nameidata nd;
2229 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2232 dentry = lookup_create(&nd, 0);
2233 error = PTR_ERR(dentry);
2234 if (!IS_ERR(dentry)) {
2235 error = vfs_symlink(nd.dentry->d_inode, dentry,
2236 from, S_IALLUGO, &nd);
2239 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2248 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2250 return sys_symlinkat(oldname, AT_FDCWD, newname);
2253 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2254 struct dentry *new_dentry, struct nameidata *nd)
2256 struct inode *inode = old_dentry->d_inode;
2262 error = may_create(dir, new_dentry, nd);
2266 if (dir->i_sb != inode->i_sb)
2270 * A link to an append-only or immutable file cannot be created.
2272 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2274 if (!dir->i_op || !dir->i_op->link)
2276 if (S_ISDIR(old_dentry->d_inode->i_mode))
2279 error = security_inode_link(old_dentry, dir, new_dentry);
2283 mutex_lock(&old_dentry->d_inode->i_mutex);
2285 error = dir->i_op->link(old_dentry, dir, new_dentry);
2286 mutex_unlock(&old_dentry->d_inode->i_mutex);
2288 fsnotify_create(dir, new_dentry->d_name.name);
2293 * Hardlinks are often used in delicate situations. We avoid
2294 * security-related surprises by not following symlinks on the
2297 * We don't follow them on the oldname either to be compatible
2298 * with linux 2.0, and to avoid hard-linking to directories
2299 * and other special files. --ADM
2301 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2302 int newdfd, const char __user *newname,
2305 struct dentry *new_dentry;
2306 struct nameidata nd, old_nd;
2313 to = getname(newname);
2317 error = __user_walk_fd(olddfd, oldname, 0, &old_nd);
2320 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2324 if (old_nd.mnt != nd.mnt)
2326 new_dentry = lookup_create(&nd, 0);
2327 error = PTR_ERR(new_dentry);
2328 if (!IS_ERR(new_dentry)) {
2329 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2333 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2337 path_release(&old_nd);
2344 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2346 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2350 * The worst of all namespace operations - renaming directory. "Perverted"
2351 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2353 * a) we can get into loop creation. Check is done in is_subdir().
2354 * b) race potential - two innocent renames can create a loop together.
2355 * That's where 4.4 screws up. Current fix: serialization on
2356 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2358 * c) we have to lock _three_ objects - parents and victim (if it exists).
2359 * And that - after we got ->i_mutex on parents (until then we don't know
2360 * whether the target exists). Solution: try to be smart with locking
2361 * order for inodes. We rely on the fact that tree topology may change
2362 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2363 * move will be locked. Thus we can rank directories by the tree
2364 * (ancestors first) and rank all non-directories after them.
2365 * That works since everybody except rename does "lock parent, lookup,
2366 * lock child" and rename is under ->s_vfs_rename_sem.
2367 * HOWEVER, it relies on the assumption that any object with ->lookup()
2368 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2369 * we'd better make sure that there's no link(2) for them.
2370 * d) some filesystems don't support opened-but-unlinked directories,
2371 * either because of layout or because they are not ready to deal with
2372 * all cases correctly. The latter will be fixed (taking this sort of
2373 * stuff into VFS), but the former is not going away. Solution: the same
2374 * trick as in rmdir().
2375 * e) conversion from fhandle to dentry may come in the wrong moment - when
2376 * we are removing the target. Solution: we will have to grab ->i_mutex
2377 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2378 * ->i_mutex on parents, which works but leads to some truely excessive
2381 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2382 struct inode *new_dir, struct dentry *new_dentry)
2385 struct inode *target;
2388 * If we are going to change the parent - check write permissions,
2389 * we'll need to flip '..'.
2391 if (new_dir != old_dir) {
2392 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2397 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2401 target = new_dentry->d_inode;
2403 mutex_lock(&target->i_mutex);
2404 dentry_unhash(new_dentry);
2406 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2409 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2412 target->i_flags |= S_DEAD;
2413 mutex_unlock(&target->i_mutex);
2414 if (d_unhashed(new_dentry))
2415 d_rehash(new_dentry);
2419 d_move(old_dentry,new_dentry);
2423 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2424 struct inode *new_dir, struct dentry *new_dentry)
2426 struct inode *target;
2429 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2434 target = new_dentry->d_inode;
2436 mutex_lock(&target->i_mutex);
2437 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2440 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2442 /* The following d_move() should become unconditional */
2443 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2444 d_move(old_dentry, new_dentry);
2447 mutex_unlock(&target->i_mutex);
2452 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2453 struct inode *new_dir, struct dentry *new_dentry)
2456 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2457 const char *old_name;
2459 if (old_dentry->d_inode == new_dentry->d_inode)
2462 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2466 if (!new_dentry->d_inode)
2467 error = may_create(new_dir, new_dentry, NULL);
2469 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2473 if (!old_dir->i_op || !old_dir->i_op->rename)
2476 DQUOT_INIT(old_dir);
2477 DQUOT_INIT(new_dir);
2479 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2482 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2484 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2486 const char *new_name = old_dentry->d_name.name;
2487 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2488 new_dentry->d_inode, old_dentry->d_inode);
2490 fsnotify_oldname_free(old_name);
2495 static int do_rename(int olddfd, const char *oldname,
2496 int newdfd, const char *newname)
2499 struct dentry * old_dir, * new_dir;
2500 struct dentry * old_dentry, *new_dentry;
2501 struct dentry * trap;
2502 struct nameidata oldnd, newnd;
2504 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2508 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2513 if (oldnd.mnt != newnd.mnt)
2516 old_dir = oldnd.dentry;
2518 if (oldnd.last_type != LAST_NORM)
2521 new_dir = newnd.dentry;
2522 if (newnd.last_type != LAST_NORM)
2525 trap = lock_rename(new_dir, old_dir);
2527 old_dentry = lookup_hash(&oldnd);
2528 error = PTR_ERR(old_dentry);
2529 if (IS_ERR(old_dentry))
2531 /* source must exist */
2533 if (!old_dentry->d_inode)
2535 /* unless the source is a directory trailing slashes give -ENOTDIR */
2536 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2538 if (oldnd.last.name[oldnd.last.len])
2540 if (newnd.last.name[newnd.last.len])
2543 /* source should not be ancestor of target */
2545 if (old_dentry == trap)
2548 if (MNT_IS_RDONLY(newnd.mnt))
2550 new_dentry = lookup_hash(&newnd);
2551 error = PTR_ERR(new_dentry);
2552 if (IS_ERR(new_dentry))
2554 /* target should not be an ancestor of source */
2556 if (new_dentry == trap)
2559 error = vfs_rename(old_dir->d_inode, old_dentry,
2560 new_dir->d_inode, new_dentry);
2566 unlock_rename(new_dir, old_dir);
2568 path_release(&newnd);
2570 path_release(&oldnd);
2575 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2576 int newdfd, const char __user *newname)
2582 from = getname(oldname);
2584 return PTR_ERR(from);
2585 to = getname(newname);
2586 error = PTR_ERR(to);
2588 error = do_rename(olddfd, from, newdfd, to);
2595 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2597 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2600 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2604 len = PTR_ERR(link);
2609 if (len > (unsigned) buflen)
2611 if (copy_to_user(buffer, link, len))
2618 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2619 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2620 * using) it for any given inode is up to filesystem.
2622 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2624 struct nameidata nd;
2628 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2629 if (!IS_ERR(cookie)) {
2630 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2631 if (dentry->d_inode->i_op->put_link)
2632 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2633 cookie = ERR_PTR(res);
2635 return PTR_ERR(cookie);
2638 int vfs_follow_link(struct nameidata *nd, const char *link)
2640 return __vfs_follow_link(nd, link);
2643 /* get the link contents into pagecache */
2644 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2647 struct address_space *mapping = dentry->d_inode->i_mapping;
2648 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2652 wait_on_page_locked(page);
2653 if (!PageUptodate(page))
2659 page_cache_release(page);
2660 return ERR_PTR(-EIO);
2666 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2668 struct page *page = NULL;
2669 char *s = page_getlink(dentry, &page);
2670 int res = vfs_readlink(dentry,buffer,buflen,s);
2673 page_cache_release(page);
2678 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2680 struct page *page = NULL;
2681 nd_set_link(nd, page_getlink(dentry, &page));
2685 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2687 struct page *page = cookie;
2691 page_cache_release(page);
2695 int __page_symlink(struct inode *inode, const char *symname, int len,
2698 struct address_space *mapping = inode->i_mapping;
2703 page = find_or_create_page(mapping, 0, gfp_mask);
2706 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2709 kaddr = kmap_atomic(page, KM_USER0);
2710 memcpy(kaddr, symname, len-1);
2711 kunmap_atomic(kaddr, KM_USER0);
2712 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2714 * Notice that we are _not_ going to block here - end of page is
2715 * unmapped, so this will only try to map the rest of page, see
2716 * that it is unmapped (typically even will not look into inode -
2717 * ->i_size will be enough for everything) and zero it out.
2718 * OTOH it's obviously correct and should make the page up-to-date.
2720 if (!PageUptodate(page)) {
2721 err = mapping->a_ops->readpage(NULL, page);
2722 wait_on_page_locked(page);
2726 page_cache_release(page);
2729 mark_inode_dirty(inode);
2733 page_cache_release(page);
2738 int page_symlink(struct inode *inode, const char *symname, int len)
2740 return __page_symlink(inode, symname, len,
2741 mapping_gfp_mask(inode->i_mapping));
2744 struct inode_operations page_symlink_inode_operations = {
2745 .readlink = generic_readlink,
2746 .follow_link = page_follow_link_light,
2747 .put_link = page_put_link,
2750 EXPORT_SYMBOL(__user_walk);
2751 EXPORT_SYMBOL(__user_walk_fd);
2752 EXPORT_SYMBOL(follow_down);
2753 EXPORT_SYMBOL(follow_up);
2754 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2755 EXPORT_SYMBOL(getname);
2756 EXPORT_SYMBOL(lock_rename);
2757 EXPORT_SYMBOL(lookup_hash);
2758 EXPORT_SYMBOL(lookup_one_len);
2759 EXPORT_SYMBOL(page_follow_link_light);
2760 EXPORT_SYMBOL(page_put_link);
2761 EXPORT_SYMBOL(page_readlink);
2762 EXPORT_SYMBOL(__page_symlink);
2763 EXPORT_SYMBOL(page_symlink);
2764 EXPORT_SYMBOL(page_symlink_inode_operations);
2765 EXPORT_SYMBOL(path_lookup);
2766 EXPORT_SYMBOL(path_release);
2767 EXPORT_SYMBOL(path_walk);
2768 EXPORT_SYMBOL(permission);
2769 EXPORT_SYMBOL(vfs_permission);
2770 EXPORT_SYMBOL(file_permission);
2771 EXPORT_SYMBOL(unlock_rename);
2772 EXPORT_SYMBOL(vfs_create);
2773 EXPORT_SYMBOL(vfs_follow_link);
2774 EXPORT_SYMBOL(vfs_link);
2775 EXPORT_SYMBOL(vfs_mkdir);
2776 EXPORT_SYMBOL(vfs_mknod);
2777 EXPORT_SYMBOL(generic_permission);
2778 EXPORT_SYMBOL(vfs_readlink);
2779 EXPORT_SYMBOL(vfs_rename);
2780 EXPORT_SYMBOL(vfs_rmdir);
2781 EXPORT_SYMBOL(vfs_symlink);
2782 EXPORT_SYMBOL(vfs_unlink);
2783 EXPORT_SYMBOL(dentry_unhash);
2784 EXPORT_SYMBOL(generic_readlink);