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_mutex 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(!audit_dummy_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)) {
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 #ifdef CONFIG_VSERVER_FILESHARING
248 /* MEF: PlanetLab FS module assumes that any file that can be
249 * named (e.g., via a cross mount) is not hidden from another
250 * context or the admin context.
252 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC))
255 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
258 vxwprintk(1, "xid=%d denied access to %p[#%d,%lu] »%s«.",
259 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
264 int permission(struct inode *inode, int mask, struct nameidata *nd)
266 umode_t mode = inode->i_mode;
269 if (mask & MAY_WRITE) {
272 * Nobody gets write access to a read-only fs.
274 if ((IS_RDONLY(inode) || (nd && MNT_IS_RDONLY(nd->mnt))) &&
275 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
279 * Nobody gets write access to an immutable file.
281 if (IS_IMMUTABLE(inode))
287 * MAY_EXEC on regular files requires special handling: We override
288 * filesystem execute permissions if the mode bits aren't set.
290 if ((mask & MAY_EXEC) && S_ISREG(mode) && !(mode & S_IXUGO))
293 /* Ordinary permission routines do not understand MAY_APPEND. */
294 submask = mask & ~MAY_APPEND;
295 if ((retval = xid_permission(inode, mask, nd)))
297 if (inode->i_op && inode->i_op->permission)
298 retval = inode->i_op->permission(inode, submask, nd);
300 retval = generic_permission(inode, submask, NULL);
304 return security_inode_permission(inode, mask, nd);
308 * vfs_permission - check for access rights to a given path
309 * @nd: lookup result that describes the path
310 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
312 * Used to check for read/write/execute permissions on a path.
313 * We use "fsuid" for this, letting us set arbitrary permissions
314 * for filesystem access without changing the "normal" uids which
315 * are used for other things.
317 int vfs_permission(struct nameidata *nd, int mask)
319 return permission(nd->dentry->d_inode, mask, nd);
323 * file_permission - check for additional access rights to a given file
324 * @file: file to check access rights for
325 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
327 * Used to check for read/write/execute permissions on an already opened
331 * Do not use this function in new code. All access checks should
332 * be done using vfs_permission().
334 int file_permission(struct file *file, int mask)
336 return permission(file->f_dentry->d_inode, mask, NULL);
340 * get_write_access() gets write permission for a file.
341 * put_write_access() releases this write permission.
342 * This is used for regular files.
343 * We cannot support write (and maybe mmap read-write shared) accesses and
344 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
345 * can have the following values:
346 * 0: no writers, no VM_DENYWRITE mappings
347 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
348 * > 0: (i_writecount) users are writing to the file.
350 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
351 * except for the cases where we don't hold i_writecount yet. Then we need to
352 * use {get,deny}_write_access() - these functions check the sign and refuse
353 * to do the change if sign is wrong. Exclusion between them is provided by
354 * the inode->i_lock spinlock.
357 int get_write_access(struct inode * inode)
359 spin_lock(&inode->i_lock);
360 if (atomic_read(&inode->i_writecount) < 0) {
361 spin_unlock(&inode->i_lock);
364 atomic_inc(&inode->i_writecount);
365 spin_unlock(&inode->i_lock);
370 int deny_write_access(struct file * file)
372 struct inode *inode = file->f_dentry->d_inode;
374 spin_lock(&inode->i_lock);
375 if (atomic_read(&inode->i_writecount) > 0) {
376 spin_unlock(&inode->i_lock);
379 atomic_dec(&inode->i_writecount);
380 spin_unlock(&inode->i_lock);
385 void path_release(struct nameidata *nd)
392 * umount() mustn't call path_release()/mntput() as that would clear
395 void path_release_on_umount(struct nameidata *nd)
398 mntput_no_expire(nd->mnt);
402 * release_open_intent - free up open intent resources
403 * @nd: pointer to nameidata
405 void release_open_intent(struct nameidata *nd)
407 if (nd->intent.open.file->f_dentry == NULL)
408 put_filp(nd->intent.open.file);
410 fput(nd->intent.open.file);
413 static inline struct dentry *do_revalidate(struct dentry *dentry, struct nameidata *nd)
415 int status = dentry->d_op->d_revalidate(dentry, nd);
416 if (unlikely(status <= 0)) {
418 * The dentry failed validation.
419 * If d_revalidate returned 0 attempt to invalidate
420 * the dentry otherwise d_revalidate is asking us
421 * to return a fail status.
424 if (!d_invalidate(dentry)) {
430 dentry = ERR_PTR(status);
437 * Internal lookup() using the new generic dcache.
440 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
442 struct dentry * dentry = __d_lookup(parent, name);
444 /* lockess __d_lookup may fail due to concurrent d_move()
445 * in some unrelated directory, so try with d_lookup
448 dentry = d_lookup(parent, name);
450 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
451 dentry = do_revalidate(dentry, nd);
457 * Short-cut version of permission(), for calling by
458 * path_walk(), when dcache lock is held. Combines parts
459 * of permission() and generic_permission(), and tests ONLY for
460 * MAY_EXEC permission.
462 * If appropriate, check DAC only. If not appropriate, or
463 * short-cut DAC fails, then call permission() to do more
464 * complete permission check.
466 static int exec_permission_lite(struct inode *inode,
467 struct nameidata *nd)
469 umode_t mode = inode->i_mode;
471 if (vx_barrier(inode))
473 if (inode->i_op && inode->i_op->permission)
476 if (current->fsuid == inode->i_uid)
478 else if (in_group_p(inode->i_gid))
484 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
487 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
490 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
495 return security_inode_permission(inode, MAY_EXEC, nd);
499 * This is called when everything else fails, and we actually have
500 * to go to the low-level filesystem to find out what we should do..
502 * We get the directory semaphore, and after getting that we also
503 * make sure that nobody added the entry to the dcache in the meantime..
506 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
508 struct dentry * result;
509 struct inode *dir = parent->d_inode;
511 mutex_lock(&dir->i_mutex);
513 * First re-do the cached lookup just in case it was created
514 * while we waited for the directory semaphore..
516 * FIXME! This could use version numbering or similar to
517 * avoid unnecessary cache lookups.
519 * The "dcache_lock" is purely to protect the RCU list walker
520 * from concurrent renames at this point (we mustn't get false
521 * negatives from the RCU list walk here, unlike the optimistic
524 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
526 result = d_lookup(parent, name);
528 struct dentry * dentry = d_alloc(parent, name);
529 result = ERR_PTR(-ENOMEM);
531 result = dir->i_op->lookup(dir, dentry, nd);
537 mutex_unlock(&dir->i_mutex);
542 * Uhhuh! Nasty case: the cache was re-populated while
543 * we waited on the semaphore. Need to revalidate.
545 mutex_unlock(&dir->i_mutex);
546 if (result->d_op && result->d_op->d_revalidate) {
547 result = do_revalidate(result, nd);
549 result = ERR_PTR(-ENOENT);
554 static int __emul_lookup_dentry(const char *, struct nameidata *);
557 static __always_inline int
558 walk_init_root(const char *name, struct nameidata *nd)
560 read_lock(¤t->fs->lock);
561 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
562 nd->mnt = mntget(current->fs->altrootmnt);
563 nd->dentry = dget(current->fs->altroot);
564 read_unlock(¤t->fs->lock);
565 if (__emul_lookup_dentry(name,nd))
567 read_lock(¤t->fs->lock);
569 nd->mnt = mntget(current->fs->rootmnt);
570 nd->dentry = dget(current->fs->root);
571 read_unlock(¤t->fs->lock);
575 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
584 if (!walk_init_root(link, nd))
585 /* weird __emul_prefix() stuff did it */
588 res = link_path_walk(link, nd);
590 if (nd->depth || res || nd->last_type!=LAST_NORM)
593 * If it is an iterative symlinks resolution in open_namei() we
594 * have to copy the last component. And all that crap because of
595 * bloody create() on broken symlinks. Furrfu...
598 if (unlikely(!name)) {
602 strcpy(name, nd->last.name);
603 nd->last.name = name;
607 return PTR_ERR(link);
611 struct vfsmount *mnt;
612 struct dentry *dentry;
615 static inline void dput_path(struct path *path, struct nameidata *nd)
618 if (path->mnt != nd->mnt)
622 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
625 if (nd->mnt != path->mnt)
628 nd->dentry = path->dentry;
631 static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
635 struct dentry *dentry = path->dentry;
637 touch_atime(path->mnt, dentry);
638 nd_set_link(nd, NULL);
640 if (path->mnt != nd->mnt) {
641 path_to_nameidata(path, nd);
645 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
646 error = PTR_ERR(cookie);
647 if (!IS_ERR(cookie)) {
648 char *s = nd_get_link(nd);
651 error = __vfs_follow_link(nd, s);
652 if (dentry->d_inode->i_op->put_link)
653 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
662 * This limits recursive symlink follows to 8, while
663 * limiting consecutive symlinks to 40.
665 * Without that kind of total limit, nasty chains of consecutive
666 * symlinks can cause almost arbitrarily long lookups.
668 static inline int do_follow_link(struct path *path, struct nameidata *nd)
671 if (current->link_count >= MAX_NESTED_LINKS)
673 if (current->total_link_count >= 40)
675 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
677 err = security_inode_follow_link(path->dentry, nd);
680 current->link_count++;
681 current->total_link_count++;
683 err = __do_follow_link(path, nd);
684 current->link_count--;
693 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
695 struct vfsmount *parent;
696 struct dentry *mountpoint;
697 spin_lock(&vfsmount_lock);
698 parent=(*mnt)->mnt_parent;
699 if (parent == *mnt) {
700 spin_unlock(&vfsmount_lock);
704 mountpoint=dget((*mnt)->mnt_mountpoint);
705 spin_unlock(&vfsmount_lock);
707 *dentry = mountpoint;
713 /* no need for dcache_lock, as serialization is taken care in
716 static int __follow_mount(struct path *path)
719 while (d_mountpoint(path->dentry)) {
720 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
727 path->dentry = dget(mounted->mnt_root);
733 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
735 while (d_mountpoint(*dentry)) {
736 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
742 *dentry = dget(mounted->mnt_root);
746 /* no need for dcache_lock, as serialization is taken care in
749 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
751 struct vfsmount *mounted;
753 mounted = lookup_mnt(*mnt, *dentry);
758 *dentry = dget(mounted->mnt_root);
764 static __always_inline void follow_dotdot(struct nameidata *nd)
767 struct vfsmount *parent;
768 struct dentry *old = nd->dentry;
770 read_lock(¤t->fs->lock);
771 if (nd->dentry == current->fs->root &&
772 nd->mnt == current->fs->rootmnt) {
773 read_unlock(¤t->fs->lock);
774 /* for sane '/' avoid follow_mount() */
777 read_unlock(¤t->fs->lock);
778 spin_lock(&dcache_lock);
779 if (nd->dentry != nd->mnt->mnt_root) {
780 nd->dentry = dget(nd->dentry->d_parent);
781 spin_unlock(&dcache_lock);
785 spin_unlock(&dcache_lock);
786 spin_lock(&vfsmount_lock);
787 parent = nd->mnt->mnt_parent;
788 if (parent == nd->mnt) {
789 spin_unlock(&vfsmount_lock);
793 nd->dentry = dget(nd->mnt->mnt_mountpoint);
794 spin_unlock(&vfsmount_lock);
799 follow_mount(&nd->mnt, &nd->dentry);
803 * It's more convoluted than I'd like it to be, but... it's still fairly
804 * small and for now I'd prefer to have fast path as straight as possible.
805 * It _is_ time-critical.
807 static int do_lookup(struct nameidata *nd, struct qstr *name,
808 struct path *path, int atomic)
810 struct vfsmount *mnt = nd->mnt;
811 struct dentry *dentry = __d_lookup(nd->dentry, name);
816 if (dentry->d_op && dentry->d_op->d_revalidate)
817 goto need_revalidate;
818 inode = dentry->d_inode;
821 #ifdef CONFIG_VSERVER_FILESHARING
822 /* MEF: PlanetLab FS module assumes that any file that can be
823 * named (e.g., via a cross mount) is not hidden from another
824 * context or the admin context.
826 if (vx_check(inode->i_xid,VX_STATIC|VX_DYNAMIC|VX_ADMIN)) {
829 else /* do the following check */
831 if (!vx_check(inode->i_xid, VX_WATCH|VX_ADMIN|VX_HOSTID|VX_IDENT))
833 if (inode->i_sb->s_magic == PROC_SUPER_MAGIC) {
834 struct proc_dir_entry *de = PDE(inode);
836 if (de && !vx_hide_check(0, de->vx_flags))
841 path->dentry = dentry;
842 __follow_mount(path);
845 vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
846 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
847 vxd_path(dentry, mnt));
853 return -EWOULDBLOCKIO;
854 dentry = real_lookup(nd->dentry, name, nd);
861 return -EWOULDBLOCKIO;
862 dentry = do_revalidate(dentry, nd);
870 return PTR_ERR(dentry);
875 * This is the basic name resolution function, turning a pathname into
876 * the final dentry. We expect 'base' to be positive and a directory.
878 * Returns 0 and nd will have valid dentry and mnt on success.
879 * Returns error and drops reference to input namei data on failure.
881 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
886 unsigned int lookup_flags = nd->flags;
888 atomic = (lookup_flags & LOOKUP_ATOMIC);
895 inode = nd->dentry->d_inode;
897 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
899 /* At this point we know we have a real path component. */
905 nd->flags |= LOOKUP_CONTINUE;
906 err = exec_permission_lite(inode, nd);
908 err = vfs_permission(nd, MAY_EXEC);
913 c = *(const unsigned char *)name;
915 hash = init_name_hash();
918 hash = partial_name_hash(c, hash);
919 c = *(const unsigned char *)name;
920 } while (c && (c != '/'));
921 this.len = name - (const char *) this.name;
922 this.hash = end_name_hash(hash);
924 /* remove trailing slashes? */
927 while (*++name == '/');
929 goto last_with_slashes;
932 * "." and ".." are special - ".." especially so because it has
933 * to be able to know about the current root directory and
934 * parent relationships.
936 if (this.name[0] == '.') switch (this.len) {
940 if (this.name[1] != '.')
943 inode = nd->dentry->d_inode;
949 * See if the low-level filesystem might want
950 * to use its own hash..
952 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
953 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
957 /* This does the actual lookups.. */
958 err = do_lookup(nd, &this, &next, atomic);
963 inode = next.dentry->d_inode;
970 if (inode->i_op->follow_link) {
971 err = do_follow_link(&next, nd);
975 inode = nd->dentry->d_inode;
982 path_to_nameidata(&next, nd);
984 if (!inode->i_op->lookup)
987 /* here ends the main loop */
990 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
992 /* Clear LOOKUP_CONTINUE iff it was previously unset */
993 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
994 if (lookup_flags & LOOKUP_PARENT)
996 if (this.name[0] == '.') switch (this.len) {
1000 if (this.name[1] != '.')
1003 inode = nd->dentry->d_inode;
1008 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
1009 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
1013 err = do_lookup(nd, &this, &next, atomic);
1016 inode = next.dentry->d_inode;
1017 if ((lookup_flags & LOOKUP_FOLLOW)
1018 && inode && inode->i_op && inode->i_op->follow_link) {
1019 err = do_follow_link(&next, nd);
1022 inode = nd->dentry->d_inode;
1024 path_to_nameidata(&next, nd);
1028 if (lookup_flags & LOOKUP_DIRECTORY) {
1030 if (!inode->i_op || !inode->i_op->lookup)
1036 nd->last_type = LAST_NORM;
1037 if (this.name[0] != '.')
1040 nd->last_type = LAST_DOT;
1041 else if (this.len == 2 && this.name[1] == '.')
1042 nd->last_type = LAST_DOTDOT;
1047 * We bypassed the ordinary revalidation routines.
1048 * We may need to check the cached dentry for staleness.
1050 if (nd->dentry && nd->dentry->d_sb &&
1051 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
1053 /* Note: we do not d_invalidate() */
1054 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
1060 dput_path(&next, nd);
1069 * Wrapper to retry pathname resolution whenever the underlying
1070 * file system returns an ESTALE.
1072 * Retry the whole path once, forcing real lookup requests
1073 * instead of relying on the dcache.
1075 int fastcall link_path_walk(const char *name, struct nameidata *nd)
1077 struct nameidata save = *nd;
1080 /* make sure the stuff we saved doesn't go away */
1084 result = __link_path_walk(name, nd);
1085 if (result == -ESTALE) {
1089 nd->flags |= LOOKUP_REVAL;
1090 result = __link_path_walk(name, nd);
1099 int fastcall path_walk(const char * name, struct nameidata *nd)
1101 current->total_link_count = 0;
1102 return link_path_walk(name, nd);
1106 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1107 * everything is done. Returns 0 and drops input nd, if lookup failed;
1109 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1111 if (path_walk(name, nd))
1112 return 0; /* something went wrong... */
1114 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1115 struct dentry *old_dentry = nd->dentry;
1116 struct vfsmount *old_mnt = nd->mnt;
1117 struct qstr last = nd->last;
1118 int last_type = nd->last_type;
1120 * NAME was not found in alternate root or it's a directory. Try to find
1121 * it in the normal root:
1123 nd->last_type = LAST_ROOT;
1124 read_lock(¤t->fs->lock);
1125 nd->mnt = mntget(current->fs->rootmnt);
1126 nd->dentry = dget(current->fs->root);
1127 read_unlock(¤t->fs->lock);
1128 if (path_walk(name, nd) == 0) {
1129 if (nd->dentry->d_inode) {
1136 nd->dentry = old_dentry;
1139 nd->last_type = last_type;
1144 void set_fs_altroot(void)
1146 char *emul = __emul_prefix();
1147 struct nameidata nd;
1148 struct vfsmount *mnt = NULL, *oldmnt;
1149 struct dentry *dentry = NULL, *olddentry;
1154 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1160 write_lock(¤t->fs->lock);
1161 oldmnt = current->fs->altrootmnt;
1162 olddentry = current->fs->altroot;
1163 current->fs->altrootmnt = mnt;
1164 current->fs->altroot = dentry;
1165 write_unlock(¤t->fs->lock);
1172 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1173 static int fastcall do_path_lookup(int dfd, const char *name,
1174 unsigned int flags, struct nameidata *nd)
1180 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1185 read_lock(¤t->fs->lock);
1186 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1187 nd->mnt = mntget(current->fs->altrootmnt);
1188 nd->dentry = dget(current->fs->altroot);
1189 read_unlock(¤t->fs->lock);
1190 if (__emul_lookup_dentry(name,nd))
1191 goto out; /* found in altroot */
1192 read_lock(¤t->fs->lock);
1194 nd->mnt = mntget(current->fs->rootmnt);
1195 nd->dentry = dget(current->fs->root);
1196 read_unlock(¤t->fs->lock);
1197 } else if (dfd == AT_FDCWD) {
1198 read_lock(¤t->fs->lock);
1199 nd->mnt = mntget(current->fs->pwdmnt);
1200 nd->dentry = dget(current->fs->pwd);
1201 read_unlock(¤t->fs->lock);
1203 struct dentry *dentry;
1205 file = fget_light(dfd, &fput_needed);
1210 dentry = file->f_dentry;
1213 if (!S_ISDIR(dentry->d_inode->i_mode))
1216 retval = file_permission(file, MAY_EXEC);
1220 nd->mnt = mntget(file->f_vfsmnt);
1221 nd->dentry = dget(dentry);
1223 fput_light(file, fput_needed);
1225 current->total_link_count = 0;
1226 retval = link_path_walk(name, nd);
1228 if (likely(retval == 0)) {
1229 if (unlikely(!audit_dummy_context() && nd && nd->dentry &&
1230 nd->dentry->d_inode))
1231 audit_inode(name, nd->dentry->d_inode);
1237 fput_light(file, fput_needed);
1241 int fastcall path_lookup(const char *name, unsigned int flags,
1242 struct nameidata *nd)
1244 return do_path_lookup(AT_FDCWD, name, flags, nd);
1247 static int __path_lookup_intent_open(int dfd, const char *name,
1248 unsigned int lookup_flags, struct nameidata *nd,
1249 int open_flags, int create_mode)
1251 struct file *filp = get_empty_filp();
1256 nd->intent.open.file = filp;
1257 nd->intent.open.flags = open_flags;
1258 nd->intent.open.create_mode = create_mode;
1259 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
1260 if (IS_ERR(nd->intent.open.file)) {
1262 err = PTR_ERR(nd->intent.open.file);
1265 } else if (err != 0)
1266 release_open_intent(nd);
1271 * path_lookup_open - lookup a file path with open intent
1272 * @dfd: the directory to use as base, or AT_FDCWD
1273 * @name: pointer to file name
1274 * @lookup_flags: lookup intent flags
1275 * @nd: pointer to nameidata
1276 * @open_flags: open intent flags
1278 int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
1279 struct nameidata *nd, int open_flags)
1281 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
1286 * path_lookup_create - lookup a file path with open + create intent
1287 * @dfd: the directory to use as base, or AT_FDCWD
1288 * @name: pointer to file name
1289 * @lookup_flags: lookup intent flags
1290 * @nd: pointer to nameidata
1291 * @open_flags: open intent flags
1292 * @create_mode: create intent flags
1294 static int path_lookup_create(int dfd, const char *name,
1295 unsigned int lookup_flags, struct nameidata *nd,
1296 int open_flags, int create_mode)
1298 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1299 nd, open_flags, create_mode);
1302 int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1303 struct nameidata *nd, int open_flags)
1305 char *tmp = getname(name);
1306 int err = PTR_ERR(tmp);
1309 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
1316 * Restricted form of lookup. Doesn't follow links, single-component only,
1317 * needs parent already locked. Doesn't follow mounts.
1320 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1322 struct dentry * dentry;
1323 struct inode *inode;
1326 inode = base->d_inode;
1327 err = permission(inode, MAY_EXEC, nd);
1328 dentry = ERR_PTR(err);
1333 * See if the low-level filesystem might want
1334 * to use its own hash..
1336 if (base->d_op && base->d_op->d_hash) {
1337 err = base->d_op->d_hash(base, name);
1338 dentry = ERR_PTR(err);
1343 dentry = cached_lookup(base, name, nd);
1345 struct dentry *new = d_alloc(base, name);
1346 dentry = ERR_PTR(-ENOMEM);
1349 dentry = inode->i_op->lookup(inode, new, nd);
1359 static struct dentry *lookup_hash(struct nameidata *nd)
1361 return __lookup_hash(&nd->last, nd->dentry, nd);
1365 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1376 hash = init_name_hash();
1378 c = *(const unsigned char *)name++;
1379 if (c == '/' || c == '\0')
1381 hash = partial_name_hash(c, hash);
1383 this.hash = end_name_hash(hash);
1385 return __lookup_hash(&this, base, NULL);
1387 return ERR_PTR(-EACCES);
1393 * is used by most simple commands to get the inode of a specified name.
1394 * Open, link etc use their own routines, but this is enough for things
1397 * namei exists in two versions: namei/lnamei. The only difference is
1398 * that namei follows links, while lnamei does not.
1401 int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1402 struct nameidata *nd)
1404 char *tmp = getname(name);
1405 int err = PTR_ERR(tmp);
1408 err = do_path_lookup(dfd, tmp, flags, nd);
1414 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1416 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1420 * It's inline, so penalty for filesystems that don't use sticky bit is
1423 static inline int check_sticky(struct inode *dir, struct inode *inode)
1425 if (!(dir->i_mode & S_ISVTX))
1427 if (inode->i_uid == current->fsuid)
1429 if (dir->i_uid == current->fsuid)
1431 return !capable(CAP_FOWNER);
1435 * Check whether we can remove a link victim from directory dir, check
1436 * whether the type of victim is right.
1437 * 1. We can't do it if dir is read-only (done in permission())
1438 * 2. We should have write and exec permissions on dir
1439 * 3. We can't remove anything from append-only dir
1440 * 4. We can't do anything with immutable dir (done in permission())
1441 * 5. If the sticky bit on dir is set we should either
1442 * a. be owner of dir, or
1443 * b. be owner of victim, or
1444 * c. have CAP_FOWNER capability
1445 * 6. If the victim is append-only or immutable we can't do antyhing with
1446 * links pointing to it.
1447 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1448 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1449 * 9. We can't remove a root or mountpoint.
1450 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1451 * nfs_async_unlink().
1453 static int may_delete(struct inode *dir, struct dentry *victim,
1454 int isdir, struct nameidata *nd)
1458 if (!victim->d_inode)
1461 BUG_ON(victim->d_parent->d_inode != dir);
1462 audit_inode_child(victim->d_name.name, victim->d_inode, dir);
1464 error = permission(dir,MAY_WRITE | MAY_EXEC, nd);
1469 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1470 IS_IXORUNLINK(victim->d_inode))
1473 if (!S_ISDIR(victim->d_inode->i_mode))
1475 if (IS_ROOT(victim))
1477 } else if (S_ISDIR(victim->d_inode->i_mode))
1479 if (IS_DEADDIR(dir))
1481 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1486 /* Check whether we can create an object with dentry child in directory
1488 * 1. We can't do it if child already exists (open has special treatment for
1489 * this case, but since we are inlined it's OK)
1490 * 2. We can't do it if dir is read-only (done in permission())
1491 * 3. We should have write and exec permissions on dir
1492 * 4. We can't do it if dir is immutable (done in permission())
1494 static inline int may_create(struct inode *dir, struct dentry *child,
1495 struct nameidata *nd)
1499 if (IS_DEADDIR(dir))
1501 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1505 * O_DIRECTORY translates into forcing a directory lookup.
1507 static inline int lookup_flags(unsigned int f)
1509 unsigned long retval = LOOKUP_FOLLOW;
1512 retval &= ~LOOKUP_FOLLOW;
1514 if (f & O_DIRECTORY)
1515 retval |= LOOKUP_DIRECTORY;
1516 if (f & O_ATOMICLOOKUP)
1517 retval |= LOOKUP_ATOMIC;
1523 * p1 and p2 should be directories on the same fs.
1525 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1530 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1534 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1536 for (p = p1; p->d_parent != p; p = p->d_parent) {
1537 if (p->d_parent == p2) {
1538 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1539 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1544 for (p = p2; p->d_parent != p; p = p->d_parent) {
1545 if (p->d_parent == p1) {
1546 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1547 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1552 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1553 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1557 void unlock_rename(struct dentry *p1, struct dentry *p2)
1559 mutex_unlock(&p1->d_inode->i_mutex);
1561 mutex_unlock(&p2->d_inode->i_mutex);
1562 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1566 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1567 struct nameidata *nd)
1569 int error = may_create(dir, dentry, nd);
1574 if (!dir->i_op || !dir->i_op->create)
1575 return -EACCES; /* shouldn't it be ENOSYS? */
1578 error = security_inode_create(dir, dentry, mode);
1582 error = dir->i_op->create(dir, dentry, mode, nd);
1584 fsnotify_create(dir, dentry);
1588 int may_open(struct nameidata *nd, int acc_mode, int flag)
1590 struct dentry *dentry = nd->dentry;
1591 struct inode *inode = dentry->d_inode;
1597 if (S_ISLNK(inode->i_mode))
1600 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1603 error = vfs_permission(nd, acc_mode);
1608 * FIFO's, sockets and device files are special: they don't
1609 * actually live on the filesystem itself, and as such you
1610 * can write to them even if the filesystem is read-only.
1612 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1614 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1615 if (nd->mnt->mnt_flags & MNT_NODEV)
1619 } else if ((IS_RDONLY(inode) || MNT_IS_RDONLY(nd->mnt))
1620 && (flag & FMODE_WRITE))
1623 * An append-only file must be opened in append mode for writing.
1625 if (IS_APPEND(inode)) {
1626 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1632 /* O_NOATIME can only be set by the owner or superuser */
1633 if (flag & O_NOATIME)
1634 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1638 * Ensure there are no outstanding leases on the file.
1640 error = break_lease(inode, flag);
1644 if (flag & O_TRUNC) {
1645 error = get_write_access(inode);
1650 * Refuse to truncate files with mandatory locks held on them.
1652 error = locks_verify_locked(inode);
1656 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
1658 put_write_access(inode);
1662 if (flag & FMODE_WRITE)
1671 * namei for open - this is in fact almost the whole open-routine.
1673 * Note that the low bits of "flag" aren't the same as in the open
1674 * system call - they are 00 - no permissions needed
1675 * 01 - read permission needed
1676 * 10 - write permission needed
1677 * 11 - read/write permissions needed
1678 * which is a lot more logical, and also allows the "no perm" needed
1679 * for symlinks (where the permissions are checked later).
1682 int open_namei(int dfd, const char *pathname, int flag,
1683 int mode, struct nameidata *nd)
1685 int acc_mode, error;
1690 acc_mode = ACC_MODE(flag);
1692 /* O_TRUNC implies we need access checks for write permissions */
1694 acc_mode |= MAY_WRITE;
1696 /* Allow the LSM permission hook to distinguish append
1697 access from general write access. */
1698 if (flag & O_APPEND)
1699 acc_mode |= MAY_APPEND;
1702 * The simplest case - just a plain lookup.
1704 if (!(flag & O_CREAT)) {
1705 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1713 * Create - we need to know the parent.
1715 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
1720 * We have the parent and last component. First of all, check
1721 * that we are not asked to creat(2) an obvious directory - that
1725 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1729 nd->flags &= ~LOOKUP_PARENT;
1730 mutex_lock(&dir->d_inode->i_mutex);
1731 path.dentry = lookup_hash(nd);
1735 error = PTR_ERR(path.dentry);
1736 if (IS_ERR(path.dentry)) {
1737 mutex_unlock(&dir->d_inode->i_mutex);
1741 if (IS_ERR(nd->intent.open.file)) {
1742 mutex_unlock(&dir->d_inode->i_mutex);
1743 error = PTR_ERR(nd->intent.open.file);
1747 /* Negative dentry, just create the file */
1748 if (!path.dentry->d_inode) {
1749 if (!IS_POSIXACL(dir->d_inode))
1750 mode &= ~current->fs->umask;
1751 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1752 mutex_unlock(&dir->d_inode->i_mutex);
1754 nd->dentry = path.dentry;
1757 /* Don't check for write permission, don't truncate */
1764 * It already exists.
1766 mutex_unlock(&dir->d_inode->i_mutex);
1767 audit_inode_update(path.dentry->d_inode);
1773 if (__follow_mount(&path)) {
1775 if (flag & O_NOFOLLOW)
1780 if (!path.dentry->d_inode)
1782 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1785 path_to_nameidata(&path, nd);
1787 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1790 error = may_open(nd, acc_mode, flag);
1796 dput_path(&path, nd);
1798 if (!IS_ERR(nd->intent.open.file))
1799 release_open_intent(nd);
1805 if (flag & O_NOFOLLOW)
1808 * This is subtle. Instead of calling do_follow_link() we do the
1809 * thing by hands. The reason is that this way we have zero link_count
1810 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1811 * After that we have the parent and last component, i.e.
1812 * we are in the same situation as after the first path_walk().
1813 * Well, almost - if the last component is normal we get its copy
1814 * stored in nd->last.name and we will have to putname() it when we
1815 * are done. Procfs-like symlinks just set LAST_BIND.
1817 nd->flags |= LOOKUP_PARENT;
1818 error = security_inode_follow_link(path.dentry, nd);
1821 error = __do_follow_link(&path, nd);
1823 /* Does someone understand code flow here? Or it is only
1824 * me so stupid? Anathema to whoever designed this non-sense
1825 * with "intent.open".
1827 release_open_intent(nd);
1830 nd->flags &= ~LOOKUP_PARENT;
1831 if (nd->last_type == LAST_BIND)
1834 if (nd->last_type != LAST_NORM)
1836 if (nd->last.name[nd->last.len]) {
1837 __putname(nd->last.name);
1842 __putname(nd->last.name);
1846 mutex_lock(&dir->d_inode->i_mutex);
1847 path.dentry = lookup_hash(nd);
1849 __putname(nd->last.name);
1854 * lookup_create - lookup a dentry, creating it if it doesn't exist
1855 * @nd: nameidata info
1856 * @is_dir: directory flag
1858 * Simple function to lookup and return a dentry and create it
1859 * if it doesn't exist. Is SMP-safe.
1861 * Returns with nd->dentry->d_inode->i_mutex locked.
1863 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1865 struct dentry *dentry = ERR_PTR(-EEXIST);
1867 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
1869 * Yucky last component or no last component at all?
1870 * (foo/., foo/.., /////)
1872 if (nd->last_type != LAST_NORM)
1874 nd->flags &= ~LOOKUP_PARENT;
1875 nd->flags |= LOOKUP_CREATE;
1876 nd->intent.open.flags = O_EXCL;
1879 * Do the final lookup.
1881 dentry = lookup_hash(nd);
1886 * Special case - lookup gave negative, but... we had foo/bar/
1887 * From the vfs_mknod() POV we just have a negative dentry -
1888 * all is fine. Let's be bastards - you had / on the end, you've
1889 * been asking for (non-existent) directory. -ENOENT for you.
1891 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1896 dentry = ERR_PTR(-ENOENT);
1900 EXPORT_SYMBOL_GPL(lookup_create);
1902 int vfs_mknod(struct inode *dir, struct dentry *dentry,
1903 int mode, dev_t dev, struct nameidata *nd)
1905 int error = may_create(dir, dentry, nd);
1910 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1913 if (!dir->i_op || !dir->i_op->mknod)
1916 error = security_inode_mknod(dir, dentry, mode, dev);
1921 error = dir->i_op->mknod(dir, dentry, mode, dev);
1923 fsnotify_create(dir, dentry);
1927 asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1932 struct dentry * dentry;
1933 struct nameidata nd;
1937 tmp = getname(filename);
1939 return PTR_ERR(tmp);
1941 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1944 dentry = lookup_create(&nd, 0);
1945 error = PTR_ERR(dentry);
1947 if (!IS_POSIXACL(nd.dentry->d_inode))
1948 mode &= ~current->fs->umask;
1949 if (!IS_ERR(dentry)) {
1950 switch (mode & S_IFMT) {
1951 case 0: case S_IFREG:
1952 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1954 case S_IFCHR: case S_IFBLK:
1955 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1956 new_decode_dev(dev), &nd);
1958 case S_IFIFO: case S_IFSOCK:
1959 error = vfs_mknod(nd.dentry->d_inode, dentry, mode,
1970 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1978 asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1980 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1983 int vfs_mkdir(struct inode *dir, struct dentry *dentry,
1984 int mode, struct nameidata *nd)
1986 int error = may_create(dir, dentry, nd);
1991 if (!dir->i_op || !dir->i_op->mkdir)
1994 mode &= (S_IRWXUGO|S_ISVTX);
1995 error = security_inode_mkdir(dir, dentry, mode);
2000 error = dir->i_op->mkdir(dir, dentry, mode);
2002 fsnotify_mkdir(dir, dentry);
2006 asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
2011 tmp = getname(pathname);
2012 error = PTR_ERR(tmp);
2014 struct dentry *dentry;
2015 struct nameidata nd;
2017 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2020 dentry = lookup_create(&nd, 1);
2021 error = PTR_ERR(dentry);
2022 if (!IS_ERR(dentry)) {
2023 if (!IS_POSIXACL(nd.dentry->d_inode))
2024 mode &= ~current->fs->umask;
2025 error = vfs_mkdir(nd.dentry->d_inode, dentry,
2029 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2038 asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2040 return sys_mkdirat(AT_FDCWD, pathname, mode);
2044 * We try to drop the dentry early: we should have
2045 * a usage count of 2 if we're the only user of this
2046 * dentry, and if that is true (possibly after pruning
2047 * the dcache), then we drop the dentry now.
2049 * A low-level filesystem can, if it choses, legally
2052 * if (!d_unhashed(dentry))
2055 * if it cannot handle the case of removing a directory
2056 * that is still in use by something else..
2058 void dentry_unhash(struct dentry *dentry)
2061 if (atomic_read(&dentry->d_count))
2062 shrink_dcache_parent(dentry);
2063 spin_lock(&dcache_lock);
2064 spin_lock(&dentry->d_lock);
2065 if (atomic_read(&dentry->d_count) == 2)
2067 spin_unlock(&dentry->d_lock);
2068 spin_unlock(&dcache_lock);
2071 int vfs_rmdir(struct inode *dir, struct dentry *dentry,
2072 struct nameidata *nd)
2074 int error = may_delete(dir, dentry, 1, nd);
2079 if (!dir->i_op || !dir->i_op->rmdir)
2084 mutex_lock(&dentry->d_inode->i_mutex);
2085 dentry_unhash(dentry);
2086 if (d_mountpoint(dentry))
2089 error = security_inode_rmdir(dir, dentry);
2091 error = dir->i_op->rmdir(dir, dentry);
2093 dentry->d_inode->i_flags |= S_DEAD;
2096 mutex_unlock(&dentry->d_inode->i_mutex);
2105 static long do_rmdir(int dfd, const char __user *pathname)
2109 struct dentry *dentry;
2110 struct nameidata nd;
2112 name = getname(pathname);
2114 return PTR_ERR(name);
2116 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2120 switch(nd.last_type) {
2131 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2132 dentry = lookup_hash(&nd);
2133 error = PTR_ERR(dentry);
2134 if (!IS_ERR(dentry)) {
2135 error = vfs_rmdir(nd.dentry->d_inode, dentry, &nd);
2138 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2146 asmlinkage long sys_rmdir(const char __user *pathname)
2148 return do_rmdir(AT_FDCWD, pathname);
2151 int vfs_unlink(struct inode *dir, struct dentry *dentry,
2152 struct nameidata *nd)
2154 int error = may_delete(dir, dentry, 0, nd);
2159 if (!dir->i_op || !dir->i_op->unlink)
2164 mutex_lock(&dentry->d_inode->i_mutex);
2165 if (d_mountpoint(dentry))
2168 error = security_inode_unlink(dir, dentry);
2170 error = dir->i_op->unlink(dir, dentry);
2172 mutex_unlock(&dentry->d_inode->i_mutex);
2174 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2175 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2183 * Make sure that the actual truncation of the file will occur outside its
2184 * directory's i_mutex. Truncate can take a long time if there is a lot of
2185 * writeout happening, and we don't want to prevent access to the directory
2186 * while waiting on the I/O.
2188 static long do_unlinkat(int dfd, const char __user *pathname)
2192 struct dentry *dentry;
2193 struct nameidata nd;
2194 struct inode *inode = NULL;
2196 name = getname(pathname);
2198 return PTR_ERR(name);
2200 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
2204 if (nd.last_type != LAST_NORM)
2206 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2207 dentry = lookup_hash(&nd);
2208 error = PTR_ERR(dentry);
2209 if (!IS_ERR(dentry)) {
2210 /* Why not before? Because we want correct error value */
2211 if (nd.last.name[nd.last.len])
2213 inode = dentry->d_inode;
2215 atomic_inc(&inode->i_count);
2216 error = vfs_unlink(nd.dentry->d_inode, dentry, &nd);
2220 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2222 iput(inode); /* truncate the inode here */
2230 error = !dentry->d_inode ? -ENOENT :
2231 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2235 asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2237 if ((flag & ~AT_REMOVEDIR) != 0)
2240 if (flag & AT_REMOVEDIR)
2241 return do_rmdir(dfd, pathname);
2243 return do_unlinkat(dfd, pathname);
2246 asmlinkage long sys_unlink(const char __user *pathname)
2248 return do_unlinkat(AT_FDCWD, pathname);
2251 int vfs_symlink(struct inode *dir, struct dentry *dentry,
2252 const char *oldname, int mode, struct nameidata *nd)
2254 int error = may_create(dir, dentry, nd);
2259 if (!dir->i_op || !dir->i_op->symlink)
2262 error = security_inode_symlink(dir, dentry, oldname);
2267 error = dir->i_op->symlink(dir, dentry, oldname);
2269 fsnotify_create(dir, dentry);
2273 asmlinkage long sys_symlinkat(const char __user *oldname,
2274 int newdfd, const char __user *newname)
2280 from = getname(oldname);
2282 return PTR_ERR(from);
2283 to = getname(newname);
2284 error = PTR_ERR(to);
2286 struct dentry *dentry;
2287 struct nameidata nd;
2289 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2292 dentry = lookup_create(&nd, 0);
2293 error = PTR_ERR(dentry);
2294 if (!IS_ERR(dentry)) {
2295 error = vfs_symlink(nd.dentry->d_inode, dentry,
2296 from, S_IALLUGO, &nd);
2299 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2308 asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2310 return sys_symlinkat(oldname, AT_FDCWD, newname);
2313 int vfs_link(struct dentry *old_dentry, struct inode *dir,
2314 struct dentry *new_dentry, struct nameidata *nd)
2316 struct inode *inode = old_dentry->d_inode;
2322 error = may_create(dir, new_dentry, nd);
2326 if (dir->i_sb != inode->i_sb)
2330 * A link to an append-only or immutable file cannot be created.
2332 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
2334 if (!dir->i_op || !dir->i_op->link)
2336 if (S_ISDIR(old_dentry->d_inode->i_mode))
2339 error = security_inode_link(old_dentry, dir, new_dentry);
2343 mutex_lock(&old_dentry->d_inode->i_mutex);
2345 error = dir->i_op->link(old_dentry, dir, new_dentry);
2346 mutex_unlock(&old_dentry->d_inode->i_mutex);
2348 fsnotify_create(dir, new_dentry);
2353 * Hardlinks are often used in delicate situations. We avoid
2354 * security-related surprises by not following symlinks on the
2357 * We don't follow them on the oldname either to be compatible
2358 * with linux 2.0, and to avoid hard-linking to directories
2359 * and other special files. --ADM
2361 asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
2362 int newdfd, const char __user *newname,
2365 struct dentry *new_dentry;
2366 struct nameidata nd, old_nd;
2370 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
2373 to = getname(newname);
2377 error = __user_walk_fd(olddfd, oldname,
2378 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2382 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2386 * We allow hard-links to be created to a bind-mount as long
2387 * as the bind-mount is not read-only. Checking for cross-dev
2388 * links is subsumed by the superblock check in vfs_link().
2391 if (MNT_IS_RDONLY(old_nd.mnt))
2393 new_dentry = lookup_create(&nd, 0);
2394 error = PTR_ERR(new_dentry);
2395 if (!IS_ERR(new_dentry)) {
2396 error = vfs_link(old_nd.dentry, nd.dentry->d_inode,
2400 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2404 path_release(&old_nd);
2411 asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2413 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2417 * The worst of all namespace operations - renaming directory. "Perverted"
2418 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2420 * a) we can get into loop creation. Check is done in is_subdir().
2421 * b) race potential - two innocent renames can create a loop together.
2422 * That's where 4.4 screws up. Current fix: serialization on
2423 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2425 * c) we have to lock _three_ objects - parents and victim (if it exists).
2426 * And that - after we got ->i_mutex on parents (until then we don't know
2427 * whether the target exists). Solution: try to be smart with locking
2428 * order for inodes. We rely on the fact that tree topology may change
2429 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
2430 * move will be locked. Thus we can rank directories by the tree
2431 * (ancestors first) and rank all non-directories after them.
2432 * That works since everybody except rename does "lock parent, lookup,
2433 * lock child" and rename is under ->s_vfs_rename_mutex.
2434 * HOWEVER, it relies on the assumption that any object with ->lookup()
2435 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2436 * we'd better make sure that there's no link(2) for them.
2437 * d) some filesystems don't support opened-but-unlinked directories,
2438 * either because of layout or because they are not ready to deal with
2439 * all cases correctly. The latter will be fixed (taking this sort of
2440 * stuff into VFS), but the former is not going away. Solution: the same
2441 * trick as in rmdir().
2442 * e) conversion from fhandle to dentry may come in the wrong moment - when
2443 * we are removing the target. Solution: we will have to grab ->i_mutex
2444 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2445 * ->i_mutex on parents, which works but leads to some truely excessive
2448 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2449 struct inode *new_dir, struct dentry *new_dentry)
2452 struct inode *target;
2455 * If we are going to change the parent - check write permissions,
2456 * we'll need to flip '..'.
2458 if (new_dir != old_dir) {
2459 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2464 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2468 target = new_dentry->d_inode;
2470 mutex_lock(&target->i_mutex);
2471 dentry_unhash(new_dentry);
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);
2479 target->i_flags |= S_DEAD;
2480 mutex_unlock(&target->i_mutex);
2481 if (d_unhashed(new_dentry))
2482 d_rehash(new_dentry);
2486 d_move(old_dentry,new_dentry);
2490 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2491 struct inode *new_dir, struct dentry *new_dentry)
2493 struct inode *target;
2496 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2501 target = new_dentry->d_inode;
2503 mutex_lock(&target->i_mutex);
2504 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2507 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2509 /* The following d_move() should become unconditional */
2510 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2511 d_move(old_dentry, new_dentry);
2514 mutex_unlock(&target->i_mutex);
2519 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2520 struct inode *new_dir, struct dentry *new_dentry)
2523 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2524 const char *old_name;
2526 if (old_dentry->d_inode == new_dentry->d_inode)
2529 error = may_delete(old_dir, old_dentry, is_dir, NULL);
2533 if (!new_dentry->d_inode)
2534 error = may_create(new_dir, new_dentry, NULL);
2536 error = may_delete(new_dir, new_dentry, is_dir, NULL);
2540 if (!old_dir->i_op || !old_dir->i_op->rename)
2543 DQUOT_INIT(old_dir);
2544 DQUOT_INIT(new_dir);
2546 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2549 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2551 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2553 const char *new_name = old_dentry->d_name.name;
2554 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2555 new_dentry->d_inode, old_dentry->d_inode);
2557 fsnotify_oldname_free(old_name);
2562 static int do_rename(int olddfd, const char *oldname,
2563 int newdfd, const char *newname)
2566 struct dentry * old_dir, * new_dir;
2567 struct dentry * old_dentry, *new_dentry;
2568 struct dentry * trap;
2569 struct nameidata oldnd, newnd;
2571 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
2575 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
2580 if (oldnd.mnt != newnd.mnt)
2583 old_dir = oldnd.dentry;
2585 if (oldnd.last_type != LAST_NORM)
2588 new_dir = newnd.dentry;
2589 if (newnd.last_type != LAST_NORM)
2592 trap = lock_rename(new_dir, old_dir);
2594 old_dentry = lookup_hash(&oldnd);
2595 error = PTR_ERR(old_dentry);
2596 if (IS_ERR(old_dentry))
2598 /* source must exist */
2600 if (!old_dentry->d_inode)
2602 /* unless the source is a directory trailing slashes give -ENOTDIR */
2603 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2605 if (oldnd.last.name[oldnd.last.len])
2607 if (newnd.last.name[newnd.last.len])
2610 /* source should not be ancestor of target */
2612 if (old_dentry == trap)
2615 if (MNT_IS_RDONLY(newnd.mnt))
2617 new_dentry = lookup_hash(&newnd);
2618 error = PTR_ERR(new_dentry);
2619 if (IS_ERR(new_dentry))
2621 /* target should not be an ancestor of source */
2623 if (new_dentry == trap)
2626 error = vfs_rename(old_dir->d_inode, old_dentry,
2627 new_dir->d_inode, new_dentry);
2633 unlock_rename(new_dir, old_dir);
2635 path_release(&newnd);
2637 path_release(&oldnd);
2642 asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2643 int newdfd, const char __user *newname)
2649 from = getname(oldname);
2651 return PTR_ERR(from);
2652 to = getname(newname);
2653 error = PTR_ERR(to);
2655 error = do_rename(olddfd, from, newdfd, to);
2662 asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2664 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2667 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2671 len = PTR_ERR(link);
2676 if (len > (unsigned) buflen)
2678 if (copy_to_user(buffer, link, len))
2685 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2686 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2687 * using) it for any given inode is up to filesystem.
2689 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2691 struct nameidata nd;
2695 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2696 if (!IS_ERR(cookie)) {
2697 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2698 if (dentry->d_inode->i_op->put_link)
2699 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2700 cookie = ERR_PTR(res);
2702 return PTR_ERR(cookie);
2705 int vfs_follow_link(struct nameidata *nd, const char *link)
2707 return __vfs_follow_link(nd, link);
2710 /* get the link contents into pagecache */
2711 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2714 struct address_space *mapping = dentry->d_inode->i_mapping;
2715 page = read_mapping_page(mapping, 0, NULL);
2718 wait_on_page_locked(page);
2719 if (!PageUptodate(page))
2725 page_cache_release(page);
2726 return ERR_PTR(-EIO);
2732 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2734 struct page *page = NULL;
2735 char *s = page_getlink(dentry, &page);
2736 int res = vfs_readlink(dentry,buffer,buflen,s);
2739 page_cache_release(page);
2744 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2746 struct page *page = NULL;
2747 nd_set_link(nd, page_getlink(dentry, &page));
2751 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2753 struct page *page = cookie;
2757 page_cache_release(page);
2761 int __page_symlink(struct inode *inode, const char *symname, int len,
2764 struct address_space *mapping = inode->i_mapping;
2770 page = find_or_create_page(mapping, 0, gfp_mask);
2773 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2774 if (err == AOP_TRUNCATED_PAGE) {
2775 page_cache_release(page);
2780 kaddr = kmap_atomic(page, KM_USER0);
2781 memcpy(kaddr, symname, len-1);
2782 kunmap_atomic(kaddr, KM_USER0);
2783 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2784 if (err == AOP_TRUNCATED_PAGE) {
2785 page_cache_release(page);
2791 * Notice that we are _not_ going to block here - end of page is
2792 * unmapped, so this will only try to map the rest of page, see
2793 * that it is unmapped (typically even will not look into inode -
2794 * ->i_size will be enough for everything) and zero it out.
2795 * OTOH it's obviously correct and should make the page up-to-date.
2797 if (!PageUptodate(page)) {
2798 err = mapping->a_ops->readpage(NULL, page);
2799 if (err != AOP_TRUNCATED_PAGE)
2800 wait_on_page_locked(page);
2804 page_cache_release(page);
2807 mark_inode_dirty(inode);
2811 page_cache_release(page);
2816 int page_symlink(struct inode *inode, const char *symname, int len)
2818 return __page_symlink(inode, symname, len,
2819 mapping_gfp_mask(inode->i_mapping));
2822 struct inode_operations page_symlink_inode_operations = {
2823 .readlink = generic_readlink,
2824 .follow_link = page_follow_link_light,
2825 .put_link = page_put_link,
2828 EXPORT_SYMBOL(__user_walk);
2829 EXPORT_SYMBOL(__user_walk_fd);
2830 EXPORT_SYMBOL(follow_down);
2831 EXPORT_SYMBOL(follow_up);
2832 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2833 EXPORT_SYMBOL(getname);
2834 EXPORT_SYMBOL(lock_rename);
2835 EXPORT_SYMBOL(lookup_one_len);
2836 EXPORT_SYMBOL(page_follow_link_light);
2837 EXPORT_SYMBOL(page_put_link);
2838 EXPORT_SYMBOL(page_readlink);
2839 EXPORT_SYMBOL(__page_symlink);
2840 EXPORT_SYMBOL(page_symlink);
2841 EXPORT_SYMBOL(page_symlink_inode_operations);
2842 EXPORT_SYMBOL(path_lookup);
2843 EXPORT_SYMBOL(path_release);
2844 EXPORT_SYMBOL(path_walk);
2845 EXPORT_SYMBOL(permission);
2846 EXPORT_SYMBOL(vfs_permission);
2847 EXPORT_SYMBOL(file_permission);
2848 EXPORT_SYMBOL(unlock_rename);
2849 EXPORT_SYMBOL(vfs_create);
2850 EXPORT_SYMBOL(vfs_follow_link);
2851 EXPORT_SYMBOL(vfs_link);
2852 EXPORT_SYMBOL(vfs_mkdir);
2853 EXPORT_SYMBOL(vfs_mknod);
2854 EXPORT_SYMBOL(generic_permission);
2855 EXPORT_SYMBOL(vfs_readlink);
2856 EXPORT_SYMBOL(vfs_rename);
2857 EXPORT_SYMBOL(vfs_rmdir);
2858 EXPORT_SYMBOL(vfs_symlink);
2859 EXPORT_SYMBOL(vfs_unlink);
2860 EXPORT_SYMBOL(dentry_unhash);
2861 EXPORT_SYMBOL(generic_readlink);