4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/quotaops.h>
23 #include <linux/pagemap.h>
24 #include <linux/dnotify.h>
25 #include <linux/smp_lock.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <asm/namei.h>
32 #include <asm/uaccess.h>
34 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
36 /* [Feb-1997 T. Schoebel-Theuer]
37 * Fundamental changes in the pathname lookup mechanisms (namei)
38 * were necessary because of omirr. The reason is that omirr needs
39 * to know the _real_ pathname, not the user-supplied one, in case
40 * of symlinks (and also when transname replacements occur).
42 * The new code replaces the old recursive symlink resolution with
43 * an iterative one (in case of non-nested symlink chains). It does
44 * this with calls to <fs>_follow_link().
45 * As a side effect, dir_namei(), _namei() and follow_link() are now
46 * replaced with a single function lookup_dentry() that can handle all
47 * the special cases of the former code.
49 * With the new dcache, the pathname is stored at each inode, at least as
50 * long as the refcount of the inode is positive. As a side effect, the
51 * size of the dcache depends on the inode cache and thus is dynamic.
53 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
54 * resolution to correspond with current state of the code.
56 * Note that the symlink resolution is not *completely* iterative.
57 * There is still a significant amount of tail- and mid- recursion in
58 * the algorithm. Also, note that <fs>_readlink() is not used in
59 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
60 * may return different results than <fs>_follow_link(). Many virtual
61 * filesystems (including /proc) exhibit this behavior.
64 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
65 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
66 * and the name already exists in form of a symlink, try to create the new
67 * name indicated by the symlink. The old code always complained that the
68 * name already exists, due to not following the symlink even if its target
69 * is nonexistent. The new semantics affects also mknod() and link() when
70 * the name is a symlink pointing to a non-existant name.
72 * I don't know which semantics is the right one, since I have no access
73 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
74 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
75 * "old" one. Personally, I think the new semantics is much more logical.
76 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
77 * file does succeed in both HP-UX and SunOs, but not in Solaris
78 * and in the old Linux semantics.
81 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
82 * semantics. See the comments in "open_namei" and "do_link" below.
84 * [10-Sep-98 Alan Modra] Another symlink change.
87 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
88 * inside the path - always follow.
89 * in the last component in creation/removal/renaming - never follow.
90 * if LOOKUP_FOLLOW passed - follow.
91 * if the pathname has trailing slashes - follow.
92 * otherwise - don't follow.
93 * (applied in that order).
95 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
96 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
97 * During the 2.4 we need to fix the userland stuff depending on it -
98 * hopefully we will be able to get rid of that wart in 2.5. So far only
99 * XEmacs seems to be relying on it...
102 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
103 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
104 * any extra contention...
107 /* In order to reduce some races, while at the same time doing additional
108 * checking and hopefully speeding things up, we copy filenames to the
109 * kernel data space before using them..
111 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
112 * PATH_MAX includes the nul terminator --RR.
114 static inline int do_getname(const char __user *filename, char *page)
117 unsigned long len = PATH_MAX;
119 if (!segment_eq(get_fs(), KERNEL_DS)) {
120 if ((unsigned long) filename >= TASK_SIZE)
122 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
123 len = TASK_SIZE - (unsigned long) filename;
126 retval = strncpy_from_user(page, filename, len);
130 return -ENAMETOOLONG;
136 char * getname(const char __user * filename)
140 result = ERR_PTR(-ENOMEM);
143 int retval = do_getname(filename, tmp);
148 result = ERR_PTR(retval);
151 audit_getname(result);
155 #ifdef CONFIG_AUDITSYSCALL
156 void putname(const char *name)
158 if (unlikely(current->audit_context))
163 EXPORT_SYMBOL(putname);
168 * generic_permission - check for access rights on a Posix-like filesystem
169 * @inode: inode to check access rights for
170 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
171 * @check_acl: optional callback to check for Posix ACLs
173 * Used to check for read/write/execute permissions on a file.
174 * We use "fsuid" for this, letting us set arbitrary permissions
175 * for filesystem access without changing the "normal" uids which
176 * are used for other things..
178 int generic_permission(struct inode *inode, int mask,
179 int (*check_acl)(struct inode *inode, int mask))
181 umode_t mode = inode->i_mode;
183 if (current->fsuid == inode->i_uid)
186 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
187 int error = check_acl(inode, mask);
188 if (error == -EACCES)
189 goto check_capabilities;
190 else if (error != -EAGAIN)
194 if (in_group_p(inode->i_gid))
199 * If the DACs are ok we don't need any capability check.
201 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 * Read/write DACs are always overridable.
207 * Executable DACs are overridable if at least one exec bit is set.
209 if (!(mask & MAY_EXEC) ||
210 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
211 if (capable(CAP_DAC_OVERRIDE))
215 * Searching includes executable on directories, else just read.
217 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
218 if (capable(CAP_DAC_READ_SEARCH))
224 int permission(struct inode *inode, int mask, struct nameidata *nd)
228 if (mask & MAY_WRITE) {
229 umode_t mode = inode->i_mode;
232 * Nobody gets write access to a read-only fs.
234 if (IS_RDONLY(inode) &&
235 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
239 * Nobody gets write access to an immutable file.
241 if (IS_IMMUTABLE(inode))
246 /* Ordinary permission routines do not understand MAY_APPEND. */
247 submask = mask & ~MAY_APPEND;
248 if (inode->i_op && inode->i_op->permission)
249 retval = inode->i_op->permission(inode, submask, nd);
251 retval = generic_permission(inode, submask, NULL);
255 return security_inode_permission(inode, mask, nd);
259 * get_write_access() gets write permission for a file.
260 * put_write_access() releases this write permission.
261 * This is used for regular files.
262 * We cannot support write (and maybe mmap read-write shared) accesses and
263 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
264 * can have the following values:
265 * 0: no writers, no VM_DENYWRITE mappings
266 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
267 * > 0: (i_writecount) users are writing to the file.
269 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
270 * except for the cases where we don't hold i_writecount yet. Then we need to
271 * use {get,deny}_write_access() - these functions check the sign and refuse
272 * to do the change if sign is wrong. Exclusion between them is provided by
273 * the inode->i_lock spinlock.
276 int get_write_access(struct inode * inode)
278 spin_lock(&inode->i_lock);
279 if (atomic_read(&inode->i_writecount) < 0) {
280 spin_unlock(&inode->i_lock);
283 atomic_inc(&inode->i_writecount);
284 spin_unlock(&inode->i_lock);
289 int deny_write_access(struct file * file)
291 struct inode *inode = file->f_dentry->d_inode;
293 spin_lock(&inode->i_lock);
294 if (atomic_read(&inode->i_writecount) > 0) {
295 spin_unlock(&inode->i_lock);
298 atomic_dec(&inode->i_writecount);
299 spin_unlock(&inode->i_lock);
304 void path_release(struct nameidata *nd)
311 * umount() mustn't call path_release()/mntput() as that would clear
314 void path_release_on_umount(struct nameidata *nd)
321 * Internal lookup() using the new generic dcache.
324 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
326 struct dentry * dentry = __d_lookup(parent, name);
328 /* lockess __d_lookup may fail due to concurrent d_move()
329 * in some unrelated directory, so try with d_lookup
332 dentry = d_lookup(parent, name);
334 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
335 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
344 * Short-cut version of permission(), for calling by
345 * path_walk(), when dcache lock is held. Combines parts
346 * of permission() and generic_permission(), and tests ONLY for
347 * MAY_EXEC permission.
349 * If appropriate, check DAC only. If not appropriate, or
350 * short-cut DAC fails, then call permission() to do more
351 * complete permission check.
353 static inline int exec_permission_lite(struct inode *inode,
354 struct nameidata *nd)
356 umode_t mode = inode->i_mode;
358 if (inode->i_op && inode->i_op->permission)
361 if (current->fsuid == inode->i_uid)
363 else if (in_group_p(inode->i_gid))
369 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
372 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
375 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
380 return security_inode_permission(inode, MAY_EXEC, nd);
384 * This is called when everything else fails, and we actually have
385 * to go to the low-level filesystem to find out what we should do..
387 * We get the directory semaphore, and after getting that we also
388 * make sure that nobody added the entry to the dcache in the meantime..
391 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
393 struct dentry * result;
394 struct inode *dir = parent->d_inode;
398 * First re-do the cached lookup just in case it was created
399 * while we waited for the directory semaphore..
401 * FIXME! This could use version numbering or similar to
402 * avoid unnecessary cache lookups.
404 * The "dcache_lock" is purely to protect the RCU list walker
405 * from concurrent renames at this point (we mustn't get false
406 * negatives from the RCU list walk here, unlike the optimistic
409 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
411 result = d_lookup(parent, name);
413 struct dentry * dentry = d_alloc(parent, name);
414 result = ERR_PTR(-ENOMEM);
416 result = dir->i_op->lookup(dir, dentry, nd);
427 * Uhhuh! Nasty case: the cache was re-populated while
428 * we waited on the semaphore. Need to revalidate.
431 if (result->d_op && result->d_op->d_revalidate) {
432 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
434 result = ERR_PTR(-ENOENT);
440 static int __emul_lookup_dentry(const char *, struct nameidata *);
444 walk_init_root(const char *name, struct nameidata *nd)
446 read_lock(¤t->fs->lock);
447 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
448 nd->mnt = mntget(current->fs->altrootmnt);
449 nd->dentry = dget(current->fs->altroot);
450 read_unlock(¤t->fs->lock);
451 if (__emul_lookup_dentry(name,nd))
453 read_lock(¤t->fs->lock);
455 nd->mnt = mntget(current->fs->rootmnt);
456 nd->dentry = dget(current->fs->root);
457 read_unlock(¤t->fs->lock);
461 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
470 if (!walk_init_root(link, nd))
471 /* weird __emul_prefix() stuff did it */
474 res = link_path_walk(link, nd);
476 if (nd->depth || res || nd->last_type!=LAST_NORM)
479 * If it is an iterative symlinks resolution in open_namei() we
480 * have to copy the last component. And all that crap because of
481 * bloody create() on broken symlinks. Furrfu...
484 if (unlikely(!name)) {
488 strcpy(name, nd->last.name);
489 nd->last.name = name;
493 return PTR_ERR(link);
497 struct vfsmount *mnt;
498 struct dentry *dentry;
501 static inline int __do_follow_link(struct path *path, struct nameidata *nd)
504 struct dentry *dentry = path->dentry;
506 touch_atime(nd->mnt, dentry);
507 nd_set_link(nd, NULL);
510 error = dentry->d_inode->i_op->follow_link(dentry, nd);
512 char *s = nd_get_link(nd);
514 error = __vfs_follow_link(nd, s);
515 if (dentry->d_inode->i_op->put_link)
516 dentry->d_inode->i_op->put_link(dentry, nd);
525 * This limits recursive symlink follows to 8, while
526 * limiting consecutive symlinks to 40.
528 * Without that kind of total limit, nasty chains of consecutive
529 * symlinks can cause almost arbitrarily long lookups.
531 static inline int do_follow_link(struct path *path, struct nameidata *nd)
534 if (current->link_count >= MAX_NESTED_LINKS)
536 if (current->total_link_count >= 40)
538 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
540 err = security_inode_follow_link(path->dentry, nd);
543 current->link_count++;
544 current->total_link_count++;
546 err = __do_follow_link(path, nd);
547 current->link_count--;
556 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
558 struct vfsmount *parent;
559 struct dentry *mountpoint;
560 spin_lock(&vfsmount_lock);
561 parent=(*mnt)->mnt_parent;
562 if (parent == *mnt) {
563 spin_unlock(&vfsmount_lock);
567 mountpoint=dget((*mnt)->mnt_mountpoint);
568 spin_unlock(&vfsmount_lock);
570 *dentry = mountpoint;
576 /* no need for dcache_lock, as serialization is taken care in
579 static int __follow_mount(struct path *path)
582 while (d_mountpoint(path->dentry)) {
583 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
590 path->dentry = dget(mounted->mnt_root);
596 static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
599 while (d_mountpoint(*dentry)) {
600 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
606 *dentry = dget(mounted->mnt_root);
612 /* no need for dcache_lock, as serialization is taken care in
615 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
617 struct vfsmount *mounted;
619 mounted = lookup_mnt(*mnt, *dentry);
624 *dentry = dget(mounted->mnt_root);
630 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
632 return __follow_down(mnt,dentry);
635 static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
638 struct vfsmount *parent;
639 struct dentry *old = *dentry;
641 read_lock(¤t->fs->lock);
642 if (*dentry == current->fs->root &&
643 *mnt == current->fs->rootmnt) {
644 read_unlock(¤t->fs->lock);
647 read_unlock(¤t->fs->lock);
648 spin_lock(&dcache_lock);
649 if (*dentry != (*mnt)->mnt_root) {
650 *dentry = dget((*dentry)->d_parent);
651 spin_unlock(&dcache_lock);
655 spin_unlock(&dcache_lock);
656 spin_lock(&vfsmount_lock);
657 parent = (*mnt)->mnt_parent;
658 if (parent == *mnt) {
659 spin_unlock(&vfsmount_lock);
663 *dentry = dget((*mnt)->mnt_mountpoint);
664 spin_unlock(&vfsmount_lock);
669 follow_mount(mnt, dentry);
673 * It's more convoluted than I'd like it to be, but... it's still fairly
674 * small and for now I'd prefer to have fast path as straight as possible.
675 * It _is_ time-critical.
677 static int do_lookup(struct nameidata *nd, struct qstr *name,
680 struct vfsmount *mnt = nd->mnt;
681 struct dentry *dentry = __d_lookup(nd->dentry, name);
685 if (dentry->d_op && dentry->d_op->d_revalidate)
686 goto need_revalidate;
689 path->dentry = dentry;
693 dentry = real_lookup(nd->dentry, name, nd);
699 if (dentry->d_op->d_revalidate(dentry, nd))
701 if (d_invalidate(dentry))
707 return PTR_ERR(dentry);
712 * This is the basic name resolution function, turning a pathname into
713 * the final dentry. We expect 'base' to be positive and a directory.
715 * Returns 0 and nd will have valid dentry and mnt on success.
716 * Returns error and drops reference to input namei data on failure.
718 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
723 unsigned int lookup_flags = nd->flags;
730 inode = nd->dentry->d_inode;
732 lookup_flags = LOOKUP_FOLLOW;
734 /* At this point we know we have a real path component. */
740 err = exec_permission_lite(inode, nd);
741 if (err == -EAGAIN) {
742 err = permission(inode, MAY_EXEC, nd);
748 c = *(const unsigned char *)name;
750 hash = init_name_hash();
753 hash = partial_name_hash(c, hash);
754 c = *(const unsigned char *)name;
755 } while (c && (c != '/'));
756 this.len = name - (const char *) this.name;
757 this.hash = end_name_hash(hash);
759 /* remove trailing slashes? */
762 while (*++name == '/');
764 goto last_with_slashes;
767 * "." and ".." are special - ".." especially so because it has
768 * to be able to know about the current root directory and
769 * parent relationships.
771 if (this.name[0] == '.') switch (this.len) {
775 if (this.name[1] != '.')
777 follow_dotdot(&nd->mnt, &nd->dentry);
778 inode = nd->dentry->d_inode;
784 * See if the low-level filesystem might want
785 * to use its own hash..
787 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
788 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
792 nd->flags |= LOOKUP_CONTINUE;
793 /* This does the actual lookups.. */
794 err = do_lookup(nd, &this, &next);
797 /* Check mountpoints.. */
798 __follow_mount(&next);
799 if (nd->mnt != next.mnt)
803 inode = next.dentry->d_inode;
810 if (inode->i_op->follow_link) {
811 err = do_follow_link(&next, nd);
815 inode = nd->dentry->d_inode;
824 nd->dentry = next.dentry;
827 if (!inode->i_op->lookup)
830 /* here ends the main loop */
833 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
835 nd->flags &= ~LOOKUP_CONTINUE;
836 if (lookup_flags & LOOKUP_PARENT)
838 if (this.name[0] == '.') switch (this.len) {
842 if (this.name[1] != '.')
844 follow_dotdot(&nd->mnt, &nd->dentry);
845 inode = nd->dentry->d_inode;
850 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
851 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
855 err = do_lookup(nd, &this, &next);
858 __follow_mount(&next);
859 if (nd->mnt != next.mnt)
861 inode = next.dentry->d_inode;
862 if ((lookup_flags & LOOKUP_FOLLOW)
863 && inode && inode->i_op && inode->i_op->follow_link) {
864 err = do_follow_link(&next, nd);
867 inode = nd->dentry->d_inode;
871 nd->dentry = next.dentry;
876 if (lookup_flags & LOOKUP_DIRECTORY) {
878 if (!inode->i_op || !inode->i_op->lookup)
884 nd->last_type = LAST_NORM;
885 if (this.name[0] != '.')
888 nd->last_type = LAST_DOT;
889 else if (this.len == 2 && this.name[1] == '.')
890 nd->last_type = LAST_DOTDOT;
895 * We bypassed the ordinary revalidation routines.
896 * We may need to check the cached dentry for staleness.
898 if (nd->dentry && nd->dentry->d_sb &&
899 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
901 /* Note: we do not d_invalidate() */
902 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
917 * Wrapper to retry pathname resolution whenever the underlying
918 * file system returns an ESTALE.
920 * Retry the whole path once, forcing real lookup requests
921 * instead of relying on the dcache.
923 int fastcall link_path_walk(const char *name, struct nameidata *nd)
925 struct nameidata save = *nd;
928 /* make sure the stuff we saved doesn't go away */
932 result = __link_path_walk(name, nd);
933 if (result == -ESTALE) {
937 nd->flags |= LOOKUP_REVAL;
938 result = __link_path_walk(name, nd);
947 int fastcall path_walk(const char * name, struct nameidata *nd)
949 current->total_link_count = 0;
950 return link_path_walk(name, nd);
954 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
955 * everything is done. Returns 0 and drops input nd, if lookup failed;
957 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
959 if (path_walk(name, nd))
960 return 0; /* something went wrong... */
962 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
963 struct dentry *old_dentry = nd->dentry;
964 struct vfsmount *old_mnt = nd->mnt;
965 struct qstr last = nd->last;
966 int last_type = nd->last_type;
968 * NAME was not found in alternate root or it's a directory. Try to find
969 * it in the normal root:
971 nd->last_type = LAST_ROOT;
972 read_lock(¤t->fs->lock);
973 nd->mnt = mntget(current->fs->rootmnt);
974 nd->dentry = dget(current->fs->root);
975 read_unlock(¤t->fs->lock);
976 if (path_walk(name, nd) == 0) {
977 if (nd->dentry->d_inode) {
984 nd->dentry = old_dentry;
987 nd->last_type = last_type;
992 void set_fs_altroot(void)
994 char *emul = __emul_prefix();
996 struct vfsmount *mnt = NULL, *oldmnt;
997 struct dentry *dentry = NULL, *olddentry;
1002 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1008 write_lock(¤t->fs->lock);
1009 oldmnt = current->fs->altrootmnt;
1010 olddentry = current->fs->altroot;
1011 current->fs->altrootmnt = mnt;
1012 current->fs->altroot = dentry;
1013 write_unlock(¤t->fs->lock);
1020 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1021 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1025 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1029 read_lock(¤t->fs->lock);
1031 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1032 nd->mnt = mntget(current->fs->altrootmnt);
1033 nd->dentry = dget(current->fs->altroot);
1034 read_unlock(¤t->fs->lock);
1035 if (__emul_lookup_dentry(name,nd))
1036 goto out; /* found in altroot */
1037 read_lock(¤t->fs->lock);
1039 nd->mnt = mntget(current->fs->rootmnt);
1040 nd->dentry = dget(current->fs->root);
1042 nd->mnt = mntget(current->fs->pwdmnt);
1043 nd->dentry = dget(current->fs->pwd);
1045 read_unlock(¤t->fs->lock);
1046 current->total_link_count = 0;
1047 retval = link_path_walk(name, nd);
1049 if (unlikely(current->audit_context
1050 && nd && nd->dentry && nd->dentry->d_inode))
1051 audit_inode(name, nd->dentry->d_inode);
1056 * Restricted form of lookup. Doesn't follow links, single-component only,
1057 * needs parent already locked. Doesn't follow mounts.
1060 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1062 struct dentry * dentry;
1063 struct inode *inode;
1066 inode = base->d_inode;
1067 err = permission(inode, MAY_EXEC, nd);
1068 dentry = ERR_PTR(err);
1073 * See if the low-level filesystem might want
1074 * to use its own hash..
1076 if (base->d_op && base->d_op->d_hash) {
1077 err = base->d_op->d_hash(base, name);
1078 dentry = ERR_PTR(err);
1083 dentry = cached_lookup(base, name, nd);
1085 struct dentry *new = d_alloc(base, name);
1086 dentry = ERR_PTR(-ENOMEM);
1089 dentry = inode->i_op->lookup(inode, new, nd);
1099 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1101 return __lookup_hash(name, base, NULL);
1105 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1116 hash = init_name_hash();
1118 c = *(const unsigned char *)name++;
1119 if (c == '/' || c == '\0')
1121 hash = partial_name_hash(c, hash);
1123 this.hash = end_name_hash(hash);
1125 return lookup_hash(&this, base);
1127 return ERR_PTR(-EACCES);
1133 * is used by most simple commands to get the inode of a specified name.
1134 * Open, link etc use their own routines, but this is enough for things
1137 * namei exists in two versions: namei/lnamei. The only difference is
1138 * that namei follows links, while lnamei does not.
1141 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1143 char *tmp = getname(name);
1144 int err = PTR_ERR(tmp);
1147 err = path_lookup(tmp, flags, nd);
1154 * It's inline, so penalty for filesystems that don't use sticky bit is
1157 static inline int check_sticky(struct inode *dir, struct inode *inode)
1159 if (!(dir->i_mode & S_ISVTX))
1161 if (inode->i_uid == current->fsuid)
1163 if (dir->i_uid == current->fsuid)
1165 return !capable(CAP_FOWNER);
1169 * Check whether we can remove a link victim from directory dir, check
1170 * whether the type of victim is right.
1171 * 1. We can't do it if dir is read-only (done in permission())
1172 * 2. We should have write and exec permissions on dir
1173 * 3. We can't remove anything from append-only dir
1174 * 4. We can't do anything with immutable dir (done in permission())
1175 * 5. If the sticky bit on dir is set we should either
1176 * a. be owner of dir, or
1177 * b. be owner of victim, or
1178 * c. have CAP_FOWNER capability
1179 * 6. If the victim is append-only or immutable we can't do antyhing with
1180 * links pointing to it.
1181 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1182 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1183 * 9. We can't remove a root or mountpoint.
1184 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1185 * nfs_async_unlink().
1187 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1191 if (!victim->d_inode)
1194 BUG_ON(victim->d_parent->d_inode != dir);
1196 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1201 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1202 IS_IMMUTABLE(victim->d_inode))
1205 if (!S_ISDIR(victim->d_inode->i_mode))
1207 if (IS_ROOT(victim))
1209 } else if (S_ISDIR(victim->d_inode->i_mode))
1211 if (IS_DEADDIR(dir))
1213 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1218 /* Check whether we can create an object with dentry child in directory
1220 * 1. We can't do it if child already exists (open has special treatment for
1221 * this case, but since we are inlined it's OK)
1222 * 2. We can't do it if dir is read-only (done in permission())
1223 * 3. We should have write and exec permissions on dir
1224 * 4. We can't do it if dir is immutable (done in permission())
1226 static inline int may_create(struct inode *dir, struct dentry *child,
1227 struct nameidata *nd)
1231 if (IS_DEADDIR(dir))
1233 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1237 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1240 * O_DIRECTORY translates into forcing a directory lookup.
1242 static inline int lookup_flags(unsigned int f)
1244 unsigned long retval = LOOKUP_FOLLOW;
1247 retval &= ~LOOKUP_FOLLOW;
1249 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1250 retval &= ~LOOKUP_FOLLOW;
1252 if (f & O_DIRECTORY)
1253 retval |= LOOKUP_DIRECTORY;
1259 * p1 and p2 should be directories on the same fs.
1261 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1266 down(&p1->d_inode->i_sem);
1270 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1272 for (p = p1; p->d_parent != p; p = p->d_parent) {
1273 if (p->d_parent == p2) {
1274 down(&p2->d_inode->i_sem);
1275 down(&p1->d_inode->i_sem);
1280 for (p = p2; p->d_parent != p; p = p->d_parent) {
1281 if (p->d_parent == p1) {
1282 down(&p1->d_inode->i_sem);
1283 down(&p2->d_inode->i_sem);
1288 down(&p1->d_inode->i_sem);
1289 down(&p2->d_inode->i_sem);
1293 void unlock_rename(struct dentry *p1, struct dentry *p2)
1295 up(&p1->d_inode->i_sem);
1297 up(&p2->d_inode->i_sem);
1298 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1302 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1303 struct nameidata *nd)
1305 int error = may_create(dir, dentry, nd);
1310 if (!dir->i_op || !dir->i_op->create)
1311 return -EACCES; /* shouldn't it be ENOSYS? */
1314 error = security_inode_create(dir, dentry, mode);
1318 error = dir->i_op->create(dir, dentry, mode, nd);
1320 inode_dir_notify(dir, DN_CREATE);
1321 security_inode_post_create(dir, dentry, mode);
1326 int may_open(struct nameidata *nd, int acc_mode, int flag)
1328 struct dentry *dentry = nd->dentry;
1329 struct inode *inode = dentry->d_inode;
1335 if (S_ISLNK(inode->i_mode))
1338 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1341 error = permission(inode, acc_mode, nd);
1346 * FIFO's, sockets and device files are special: they don't
1347 * actually live on the filesystem itself, and as such you
1348 * can write to them even if the filesystem is read-only.
1350 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1352 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1353 if (nd->mnt->mnt_flags & MNT_NODEV)
1357 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1360 * An append-only file must be opened in append mode for writing.
1362 if (IS_APPEND(inode)) {
1363 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1369 /* O_NOATIME can only be set by the owner or superuser */
1370 if (flag & O_NOATIME)
1371 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1375 * Ensure there are no outstanding leases on the file.
1377 error = break_lease(inode, flag);
1381 if (flag & O_TRUNC) {
1382 error = get_write_access(inode);
1387 * Refuse to truncate files with mandatory locks held on them.
1389 error = locks_verify_locked(inode);
1393 error = do_truncate(dentry, 0);
1395 put_write_access(inode);
1399 if (flag & FMODE_WRITE)
1408 * namei for open - this is in fact almost the whole open-routine.
1410 * Note that the low bits of "flag" aren't the same as in the open
1411 * system call - they are 00 - no permissions needed
1412 * 01 - read permission needed
1413 * 10 - write permission needed
1414 * 11 - read/write permissions needed
1415 * which is a lot more logical, and also allows the "no perm" needed
1416 * for symlinks (where the permissions are checked later).
1419 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1421 int acc_mode, error = 0;
1426 acc_mode = ACC_MODE(flag);
1428 /* Allow the LSM permission hook to distinguish append
1429 access from general write access. */
1430 if (flag & O_APPEND)
1431 acc_mode |= MAY_APPEND;
1433 /* Fill in the open() intent data */
1434 nd->intent.open.flags = flag;
1435 nd->intent.open.create_mode = mode;
1438 * The simplest case - just a plain lookup.
1440 if (!(flag & O_CREAT)) {
1441 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1448 * Create - we need to know the parent.
1450 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1455 * We have the parent and last component. First of all, check
1456 * that we are not asked to creat(2) an obvious directory - that
1460 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1464 nd->flags &= ~LOOKUP_PARENT;
1465 down(&dir->d_inode->i_sem);
1466 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1470 error = PTR_ERR(path.dentry);
1471 if (IS_ERR(path.dentry)) {
1472 up(&dir->d_inode->i_sem);
1476 /* Negative dentry, just create the file */
1477 if (!path.dentry->d_inode) {
1478 if (!IS_POSIXACL(dir->d_inode))
1479 mode &= ~current->fs->umask;
1480 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1481 up(&dir->d_inode->i_sem);
1483 nd->dentry = path.dentry;
1486 /* Don't check for write permission, don't truncate */
1493 * It already exists.
1495 up(&dir->d_inode->i_sem);
1501 if (d_mountpoint(path.dentry)) {
1503 if (flag & O_NOFOLLOW)
1505 while (__follow_down(&path.mnt,&path.dentry) && d_mountpoint(path.dentry));
1509 if (!path.dentry->d_inode)
1511 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1515 nd->dentry = path.dentry;
1517 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1520 error = may_open(nd, acc_mode, flag);
1533 if (flag & O_NOFOLLOW)
1536 * This is subtle. Instead of calling do_follow_link() we do the
1537 * thing by hands. The reason is that this way we have zero link_count
1538 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1539 * After that we have the parent and last component, i.e.
1540 * we are in the same situation as after the first path_walk().
1541 * Well, almost - if the last component is normal we get its copy
1542 * stored in nd->last.name and we will have to putname() it when we
1543 * are done. Procfs-like symlinks just set LAST_BIND.
1545 nd->flags |= LOOKUP_PARENT;
1546 error = security_inode_follow_link(path.dentry, nd);
1549 error = __do_follow_link(&path, nd);
1552 nd->flags &= ~LOOKUP_PARENT;
1553 if (nd->last_type == LAST_BIND)
1556 if (nd->last_type != LAST_NORM)
1558 if (nd->last.name[nd->last.len]) {
1559 putname(nd->last.name);
1564 putname(nd->last.name);
1568 down(&dir->d_inode->i_sem);
1569 path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1571 putname(nd->last.name);
1576 * lookup_create - lookup a dentry, creating it if it doesn't exist
1577 * @nd: nameidata info
1578 * @is_dir: directory flag
1580 * Simple function to lookup and return a dentry and create it
1581 * if it doesn't exist. Is SMP-safe.
1583 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1585 struct dentry *dentry;
1587 down(&nd->dentry->d_inode->i_sem);
1588 dentry = ERR_PTR(-EEXIST);
1589 if (nd->last_type != LAST_NORM)
1591 nd->flags &= ~LOOKUP_PARENT;
1592 dentry = lookup_hash(&nd->last, nd->dentry);
1595 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1600 dentry = ERR_PTR(-ENOENT);
1604 EXPORT_SYMBOL_GPL(lookup_create);
1606 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1608 int error = may_create(dir, dentry, NULL);
1613 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1616 if (!dir->i_op || !dir->i_op->mknod)
1619 error = security_inode_mknod(dir, dentry, mode, dev);
1624 error = dir->i_op->mknod(dir, dentry, mode, dev);
1626 inode_dir_notify(dir, DN_CREATE);
1627 security_inode_post_mknod(dir, dentry, mode, dev);
1632 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1636 struct dentry * dentry;
1637 struct nameidata nd;
1641 tmp = getname(filename);
1643 return PTR_ERR(tmp);
1645 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1648 dentry = lookup_create(&nd, 0);
1649 error = PTR_ERR(dentry);
1651 if (!IS_POSIXACL(nd.dentry->d_inode))
1652 mode &= ~current->fs->umask;
1653 if (!IS_ERR(dentry)) {
1654 switch (mode & S_IFMT) {
1655 case 0: case S_IFREG:
1656 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1658 case S_IFCHR: case S_IFBLK:
1659 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1660 new_decode_dev(dev));
1662 case S_IFIFO: case S_IFSOCK:
1663 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1673 up(&nd.dentry->d_inode->i_sem);
1681 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1683 int error = may_create(dir, dentry, NULL);
1688 if (!dir->i_op || !dir->i_op->mkdir)
1691 mode &= (S_IRWXUGO|S_ISVTX);
1692 error = security_inode_mkdir(dir, dentry, mode);
1697 error = dir->i_op->mkdir(dir, dentry, mode);
1699 inode_dir_notify(dir, DN_CREATE);
1700 security_inode_post_mkdir(dir,dentry, mode);
1705 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1710 tmp = getname(pathname);
1711 error = PTR_ERR(tmp);
1713 struct dentry *dentry;
1714 struct nameidata nd;
1716 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1719 dentry = lookup_create(&nd, 1);
1720 error = PTR_ERR(dentry);
1721 if (!IS_ERR(dentry)) {
1722 if (!IS_POSIXACL(nd.dentry->d_inode))
1723 mode &= ~current->fs->umask;
1724 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1727 up(&nd.dentry->d_inode->i_sem);
1737 * We try to drop the dentry early: we should have
1738 * a usage count of 2 if we're the only user of this
1739 * dentry, and if that is true (possibly after pruning
1740 * the dcache), then we drop the dentry now.
1742 * A low-level filesystem can, if it choses, legally
1745 * if (!d_unhashed(dentry))
1748 * if it cannot handle the case of removing a directory
1749 * that is still in use by something else..
1751 void dentry_unhash(struct dentry *dentry)
1754 if (atomic_read(&dentry->d_count))
1755 shrink_dcache_parent(dentry);
1756 spin_lock(&dcache_lock);
1757 spin_lock(&dentry->d_lock);
1758 if (atomic_read(&dentry->d_count) == 2)
1760 spin_unlock(&dentry->d_lock);
1761 spin_unlock(&dcache_lock);
1764 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1766 int error = may_delete(dir, dentry, 1);
1771 if (!dir->i_op || !dir->i_op->rmdir)
1776 down(&dentry->d_inode->i_sem);
1777 dentry_unhash(dentry);
1778 if (d_mountpoint(dentry))
1781 error = security_inode_rmdir(dir, dentry);
1783 error = dir->i_op->rmdir(dir, dentry);
1785 dentry->d_inode->i_flags |= S_DEAD;
1788 up(&dentry->d_inode->i_sem);
1790 inode_dir_notify(dir, DN_DELETE);
1798 asmlinkage long sys_rmdir(const char __user * pathname)
1802 struct dentry *dentry;
1803 struct nameidata nd;
1805 name = getname(pathname);
1807 return PTR_ERR(name);
1809 error = path_lookup(name, LOOKUP_PARENT, &nd);
1813 switch(nd.last_type) {
1824 down(&nd.dentry->d_inode->i_sem);
1825 dentry = lookup_hash(&nd.last, nd.dentry);
1826 error = PTR_ERR(dentry);
1827 if (!IS_ERR(dentry)) {
1828 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1831 up(&nd.dentry->d_inode->i_sem);
1839 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1841 int error = may_delete(dir, dentry, 0);
1846 if (!dir->i_op || !dir->i_op->unlink)
1851 down(&dentry->d_inode->i_sem);
1852 if (d_mountpoint(dentry))
1855 error = security_inode_unlink(dir, dentry);
1857 error = dir->i_op->unlink(dir, dentry);
1859 up(&dentry->d_inode->i_sem);
1861 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1862 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1864 inode_dir_notify(dir, DN_DELETE);
1870 * Make sure that the actual truncation of the file will occur outside its
1871 * directory's i_sem. Truncate can take a long time if there is a lot of
1872 * writeout happening, and we don't want to prevent access to the directory
1873 * while waiting on the I/O.
1875 asmlinkage long sys_unlink(const char __user * pathname)
1879 struct dentry *dentry;
1880 struct nameidata nd;
1881 struct inode *inode = NULL;
1883 name = getname(pathname);
1885 return PTR_ERR(name);
1887 error = path_lookup(name, LOOKUP_PARENT, &nd);
1891 if (nd.last_type != LAST_NORM)
1893 down(&nd.dentry->d_inode->i_sem);
1894 dentry = lookup_hash(&nd.last, nd.dentry);
1895 error = PTR_ERR(dentry);
1896 if (!IS_ERR(dentry)) {
1897 /* Why not before? Because we want correct error value */
1898 if (nd.last.name[nd.last.len])
1900 inode = dentry->d_inode;
1902 atomic_inc(&inode->i_count);
1903 error = vfs_unlink(nd.dentry->d_inode, dentry);
1907 up(&nd.dentry->d_inode->i_sem);
1909 iput(inode); /* truncate the inode here */
1917 error = !dentry->d_inode ? -ENOENT :
1918 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1922 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1924 int error = may_create(dir, dentry, NULL);
1929 if (!dir->i_op || !dir->i_op->symlink)
1932 error = security_inode_symlink(dir, dentry, oldname);
1937 error = dir->i_op->symlink(dir, dentry, oldname);
1939 inode_dir_notify(dir, DN_CREATE);
1940 security_inode_post_symlink(dir, dentry, oldname);
1945 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1951 from = getname(oldname);
1953 return PTR_ERR(from);
1954 to = getname(newname);
1955 error = PTR_ERR(to);
1957 struct dentry *dentry;
1958 struct nameidata nd;
1960 error = path_lookup(to, LOOKUP_PARENT, &nd);
1963 dentry = lookup_create(&nd, 0);
1964 error = PTR_ERR(dentry);
1965 if (!IS_ERR(dentry)) {
1966 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1969 up(&nd.dentry->d_inode->i_sem);
1978 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1980 struct inode *inode = old_dentry->d_inode;
1986 error = may_create(dir, new_dentry, NULL);
1990 if (dir->i_sb != inode->i_sb)
1994 * A link to an append-only or immutable file cannot be created.
1996 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1998 if (!dir->i_op || !dir->i_op->link)
2000 if (S_ISDIR(old_dentry->d_inode->i_mode))
2003 error = security_inode_link(old_dentry, dir, new_dentry);
2007 down(&old_dentry->d_inode->i_sem);
2009 error = dir->i_op->link(old_dentry, dir, new_dentry);
2010 up(&old_dentry->d_inode->i_sem);
2012 inode_dir_notify(dir, DN_CREATE);
2013 security_inode_post_link(old_dentry, dir, new_dentry);
2019 * Hardlinks are often used in delicate situations. We avoid
2020 * security-related surprises by not following symlinks on the
2023 * We don't follow them on the oldname either to be compatible
2024 * with linux 2.0, and to avoid hard-linking to directories
2025 * and other special files. --ADM
2027 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2029 struct dentry *new_dentry;
2030 struct nameidata nd, old_nd;
2034 to = getname(newname);
2038 error = __user_walk(oldname, 0, &old_nd);
2041 error = path_lookup(to, LOOKUP_PARENT, &nd);
2045 if (old_nd.mnt != nd.mnt)
2047 new_dentry = lookup_create(&nd, 0);
2048 error = PTR_ERR(new_dentry);
2049 if (!IS_ERR(new_dentry)) {
2050 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2053 up(&nd.dentry->d_inode->i_sem);
2057 path_release(&old_nd);
2065 * The worst of all namespace operations - renaming directory. "Perverted"
2066 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2068 * a) we can get into loop creation. Check is done in is_subdir().
2069 * b) race potential - two innocent renames can create a loop together.
2070 * That's where 4.4 screws up. Current fix: serialization on
2071 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2073 * c) we have to lock _three_ objects - parents and victim (if it exists).
2074 * And that - after we got ->i_sem on parents (until then we don't know
2075 * whether the target exists). Solution: try to be smart with locking
2076 * order for inodes. We rely on the fact that tree topology may change
2077 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2078 * move will be locked. Thus we can rank directories by the tree
2079 * (ancestors first) and rank all non-directories after them.
2080 * That works since everybody except rename does "lock parent, lookup,
2081 * lock child" and rename is under ->s_vfs_rename_sem.
2082 * HOWEVER, it relies on the assumption that any object with ->lookup()
2083 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2084 * we'd better make sure that there's no link(2) for them.
2085 * d) some filesystems don't support opened-but-unlinked directories,
2086 * either because of layout or because they are not ready to deal with
2087 * all cases correctly. The latter will be fixed (taking this sort of
2088 * stuff into VFS), but the former is not going away. Solution: the same
2089 * trick as in rmdir().
2090 * e) conversion from fhandle to dentry may come in the wrong moment - when
2091 * we are removing the target. Solution: we will have to grab ->i_sem
2092 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2093 * ->i_sem on parents, which works but leads to some truely excessive
2096 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2097 struct inode *new_dir, struct dentry *new_dentry)
2100 struct inode *target;
2103 * If we are going to change the parent - check write permissions,
2104 * we'll need to flip '..'.
2106 if (new_dir != old_dir) {
2107 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2112 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2116 target = new_dentry->d_inode;
2118 down(&target->i_sem);
2119 dentry_unhash(new_dentry);
2121 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2124 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2127 target->i_flags |= S_DEAD;
2129 if (d_unhashed(new_dentry))
2130 d_rehash(new_dentry);
2134 d_move(old_dentry,new_dentry);
2135 security_inode_post_rename(old_dir, old_dentry,
2136 new_dir, new_dentry);
2141 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2142 struct inode *new_dir, struct dentry *new_dentry)
2144 struct inode *target;
2147 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2152 target = new_dentry->d_inode;
2154 down(&target->i_sem);
2155 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2158 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2160 /* The following d_move() should become unconditional */
2161 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2162 d_move(old_dentry, new_dentry);
2163 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2171 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2172 struct inode *new_dir, struct dentry *new_dentry)
2175 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2177 if (old_dentry->d_inode == new_dentry->d_inode)
2180 error = may_delete(old_dir, old_dentry, is_dir);
2184 if (!new_dentry->d_inode)
2185 error = may_create(new_dir, new_dentry, NULL);
2187 error = may_delete(new_dir, new_dentry, is_dir);
2191 if (!old_dir->i_op || !old_dir->i_op->rename)
2194 DQUOT_INIT(old_dir);
2195 DQUOT_INIT(new_dir);
2198 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2200 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2202 if (old_dir == new_dir)
2203 inode_dir_notify(old_dir, DN_RENAME);
2205 inode_dir_notify(old_dir, DN_DELETE);
2206 inode_dir_notify(new_dir, DN_CREATE);
2212 static inline int do_rename(const char * oldname, const char * newname)
2215 struct dentry * old_dir, * new_dir;
2216 struct dentry * old_dentry, *new_dentry;
2217 struct dentry * trap;
2218 struct nameidata oldnd, newnd;
2220 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2224 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2229 if (oldnd.mnt != newnd.mnt)
2232 old_dir = oldnd.dentry;
2234 if (oldnd.last_type != LAST_NORM)
2237 new_dir = newnd.dentry;
2238 if (newnd.last_type != LAST_NORM)
2241 trap = lock_rename(new_dir, old_dir);
2243 old_dentry = lookup_hash(&oldnd.last, old_dir);
2244 error = PTR_ERR(old_dentry);
2245 if (IS_ERR(old_dentry))
2247 /* source must exist */
2249 if (!old_dentry->d_inode)
2251 /* unless the source is a directory trailing slashes give -ENOTDIR */
2252 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2254 if (oldnd.last.name[oldnd.last.len])
2256 if (newnd.last.name[newnd.last.len])
2259 /* source should not be ancestor of target */
2261 if (old_dentry == trap)
2263 new_dentry = lookup_hash(&newnd.last, new_dir);
2264 error = PTR_ERR(new_dentry);
2265 if (IS_ERR(new_dentry))
2267 /* target should not be an ancestor of source */
2269 if (new_dentry == trap)
2272 error = vfs_rename(old_dir->d_inode, old_dentry,
2273 new_dir->d_inode, new_dentry);
2279 unlock_rename(new_dir, old_dir);
2281 path_release(&newnd);
2283 path_release(&oldnd);
2288 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2294 from = getname(oldname);
2296 return PTR_ERR(from);
2297 to = getname(newname);
2298 error = PTR_ERR(to);
2300 error = do_rename(from,to);
2307 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2311 len = PTR_ERR(link);
2316 if (len > (unsigned) buflen)
2318 if (copy_to_user(buffer, link, len))
2325 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2326 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2327 * using) it for any given inode is up to filesystem.
2329 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2331 struct nameidata nd;
2334 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2336 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2337 if (dentry->d_inode->i_op->put_link)
2338 dentry->d_inode->i_op->put_link(dentry, &nd);
2343 int vfs_follow_link(struct nameidata *nd, const char *link)
2345 return __vfs_follow_link(nd, link);
2348 /* get the link contents into pagecache */
2349 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2352 struct address_space *mapping = dentry->d_inode->i_mapping;
2353 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2357 wait_on_page_locked(page);
2358 if (!PageUptodate(page))
2364 page_cache_release(page);
2365 return ERR_PTR(-EIO);
2371 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2373 struct page *page = NULL;
2374 char *s = page_getlink(dentry, &page);
2375 int res = vfs_readlink(dentry,buffer,buflen,s);
2378 page_cache_release(page);
2383 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2386 nd_set_link(nd, page_getlink(dentry, &page));
2390 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2392 if (!IS_ERR(nd_get_link(nd))) {
2394 page = find_get_page(dentry->d_inode->i_mapping, 0);
2398 page_cache_release(page);
2399 page_cache_release(page);
2403 int page_symlink(struct inode *inode, const char *symname, int len)
2405 struct address_space *mapping = inode->i_mapping;
2406 struct page *page = grab_cache_page(mapping, 0);
2412 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2415 kaddr = kmap_atomic(page, KM_USER0);
2416 memcpy(kaddr, symname, len-1);
2417 kunmap_atomic(kaddr, KM_USER0);
2418 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2420 * Notice that we are _not_ going to block here - end of page is
2421 * unmapped, so this will only try to map the rest of page, see
2422 * that it is unmapped (typically even will not look into inode -
2423 * ->i_size will be enough for everything) and zero it out.
2424 * OTOH it's obviously correct and should make the page up-to-date.
2426 if (!PageUptodate(page)) {
2427 err = mapping->a_ops->readpage(NULL, page);
2428 wait_on_page_locked(page);
2432 page_cache_release(page);
2435 mark_inode_dirty(inode);
2439 page_cache_release(page);
2444 struct inode_operations page_symlink_inode_operations = {
2445 .readlink = generic_readlink,
2446 .follow_link = page_follow_link_light,
2447 .put_link = page_put_link,
2450 EXPORT_SYMBOL(__user_walk);
2451 EXPORT_SYMBOL(follow_down);
2452 EXPORT_SYMBOL(follow_up);
2453 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2454 EXPORT_SYMBOL(getname);
2455 EXPORT_SYMBOL(lock_rename);
2456 EXPORT_SYMBOL(lookup_hash);
2457 EXPORT_SYMBOL(lookup_one_len);
2458 EXPORT_SYMBOL(page_follow_link_light);
2459 EXPORT_SYMBOL(page_put_link);
2460 EXPORT_SYMBOL(page_readlink);
2461 EXPORT_SYMBOL(page_symlink);
2462 EXPORT_SYMBOL(page_symlink_inode_operations);
2463 EXPORT_SYMBOL(path_lookup);
2464 EXPORT_SYMBOL(path_release);
2465 EXPORT_SYMBOL(path_walk);
2466 EXPORT_SYMBOL(permission);
2467 EXPORT_SYMBOL(unlock_rename);
2468 EXPORT_SYMBOL(vfs_create);
2469 EXPORT_SYMBOL(vfs_follow_link);
2470 EXPORT_SYMBOL(vfs_link);
2471 EXPORT_SYMBOL(vfs_mkdir);
2472 EXPORT_SYMBOL(vfs_mknod);
2473 EXPORT_SYMBOL(generic_permission);
2474 EXPORT_SYMBOL(vfs_readlink);
2475 EXPORT_SYMBOL(vfs_rename);
2476 EXPORT_SYMBOL(vfs_rmdir);
2477 EXPORT_SYMBOL(vfs_symlink);
2478 EXPORT_SYMBOL(vfs_unlink);
2479 EXPORT_SYMBOL(dentry_unhash);
2480 EXPORT_SYMBOL(generic_readlink);