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/mount.h>
29 #include <linux/audit.h>
30 #include <linux/vs_base.h>
32 #include <asm/namei.h>
33 #include <asm/uaccess.h>
35 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
37 /* [Feb-1997 T. Schoebel-Theuer]
38 * Fundamental changes in the pathname lookup mechanisms (namei)
39 * were necessary because of omirr. The reason is that omirr needs
40 * to know the _real_ pathname, not the user-supplied one, in case
41 * of symlinks (and also when transname replacements occur).
43 * The new code replaces the old recursive symlink resolution with
44 * an iterative one (in case of non-nested symlink chains). It does
45 * this with calls to <fs>_follow_link().
46 * As a side effect, dir_namei(), _namei() and follow_link() are now
47 * replaced with a single function lookup_dentry() that can handle all
48 * the special cases of the former code.
50 * With the new dcache, the pathname is stored at each inode, at least as
51 * long as the refcount of the inode is positive. As a side effect, the
52 * size of the dcache depends on the inode cache and thus is dynamic.
54 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
55 * resolution to correspond with current state of the code.
57 * Note that the symlink resolution is not *completely* iterative.
58 * There is still a significant amount of tail- and mid- recursion in
59 * the algorithm. Also, note that <fs>_readlink() is not used in
60 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
61 * may return different results than <fs>_follow_link(). Many virtual
62 * filesystems (including /proc) exhibit this behavior.
65 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
66 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
67 * and the name already exists in form of a symlink, try to create the new
68 * name indicated by the symlink. The old code always complained that the
69 * name already exists, due to not following the symlink even if its target
70 * is nonexistent. The new semantics affects also mknod() and link() when
71 * the name is a symlink pointing to a non-existant name.
73 * I don't know which semantics is the right one, since I have no access
74 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
75 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
76 * "old" one. Personally, I think the new semantics is much more logical.
77 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
78 * file does succeed in both HP-UX and SunOs, but not in Solaris
79 * and in the old Linux semantics.
82 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
83 * semantics. See the comments in "open_namei" and "do_link" below.
85 * [10-Sep-98 Alan Modra] Another symlink change.
88 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
89 * inside the path - always follow.
90 * in the last component in creation/removal/renaming - never follow.
91 * if LOOKUP_FOLLOW passed - follow.
92 * if the pathname has trailing slashes - follow.
93 * otherwise - don't follow.
94 * (applied in that order).
96 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
97 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
98 * During the 2.4 we need to fix the userland stuff depending on it -
99 * hopefully we will be able to get rid of that wart in 2.5. So far only
100 * XEmacs seems to be relying on it...
103 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
104 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
105 * any extra contention...
108 /* In order to reduce some races, while at the same time doing additional
109 * checking and hopefully speeding things up, we copy filenames to the
110 * kernel data space before using them..
112 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
113 * PATH_MAX includes the nul terminator --RR.
115 static inline int do_getname(const char __user *filename, char *page)
118 unsigned long len = PATH_MAX;
120 if ((unsigned long) filename >= TASK_SIZE) {
121 if (!segment_eq(get_fs(), KERNEL_DS))
123 } else if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
124 len = TASK_SIZE - (unsigned long) filename;
126 retval = strncpy_from_user((char *)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 if (unlikely(current->audit_context) && !IS_ERR(result) && result)
152 audit_getname(result);
159 * is used to check for read/write/execute permissions on a file.
160 * We use "fsuid" for this, letting us set arbitrary permissions
161 * for filesystem access without changing the "normal" uids which
162 * are used for other things..
164 int vfs_permission(struct inode * inode, int mask)
166 umode_t mode = inode->i_mode;
168 if (IS_BARRIER(inode) && !vx_check(0, VX_ADMIN|VX_WATCH))
171 if (mask & MAY_WRITE) {
173 * Nobody gets write access to a read-only fs.
175 if (IS_RDONLY(inode) &&
176 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
180 * Nobody gets write access to an immutable file.
182 if (IS_IMMUTABLE(inode))
186 if (current->fsuid == inode->i_uid)
188 else if (in_group_p(inode->i_gid))
192 * If the DACs are ok we don't need any capability check.
194 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
198 * Read/write DACs are always overridable.
199 * Executable DACs are overridable if at least one exec bit is set.
201 if (!(mask & MAY_EXEC) ||
202 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
203 if (capable(CAP_DAC_OVERRIDE))
207 * Searching includes executable on directories, else just read.
209 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
210 if (capable(CAP_DAC_READ_SEARCH))
216 static inline int xid_permission(struct inode *inode)
218 if (inode->i_xid == 0)
220 if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
225 int permission(struct inode * inode,int mask, struct nameidata *nd)
230 /* Ordinary permission routines do not understand MAY_APPEND. */
231 submask = mask & ~MAY_APPEND;
233 if ((retval = xid_permission(inode)))
235 if (inode->i_op && inode->i_op->permission)
236 retval = inode->i_op->permission(inode, submask, nd);
238 retval = vfs_permission(inode, submask);
242 return security_inode_permission(inode, mask, nd);
246 * get_write_access() gets write permission for a file.
247 * put_write_access() releases this write permission.
248 * This is used for regular files.
249 * We cannot support write (and maybe mmap read-write shared) accesses and
250 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
251 * can have the following values:
252 * 0: no writers, no VM_DENYWRITE mappings
253 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
254 * > 0: (i_writecount) users are writing to the file.
256 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
257 * except for the cases where we don't hold i_writecount yet. Then we need to
258 * use {get,deny}_write_access() - these functions check the sign and refuse
259 * to do the change if sign is wrong. Exclusion between them is provided by
260 * the inode->i_lock spinlock.
263 int get_write_access(struct inode * inode)
265 spin_lock(&inode->i_lock);
266 if (atomic_read(&inode->i_writecount) < 0) {
267 spin_unlock(&inode->i_lock);
270 atomic_inc(&inode->i_writecount);
271 spin_unlock(&inode->i_lock);
276 int deny_write_access(struct file * file)
278 struct inode *inode = file->f_dentry->d_inode;
280 spin_lock(&inode->i_lock);
281 if (atomic_read(&inode->i_writecount) > 0) {
282 spin_unlock(&inode->i_lock);
285 atomic_dec(&inode->i_writecount);
286 spin_unlock(&inode->i_lock);
291 void path_release(struct nameidata *nd)
298 * umount() mustn't call path_release()/mntput() as that would clear
301 void path_release_on_umount(struct nameidata *nd)
308 * Internal lookup() using the new generic dcache.
311 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
313 struct dentry * dentry = __d_lookup(parent, name);
315 /* lockess __d_lookup may fail due to concurrent d_move()
316 * in some unrelated directory, so try with d_lookup
319 dentry = d_lookup(parent, name);
321 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
322 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
331 * Short-cut version of permission(), for calling by
332 * path_walk(), when dcache lock is held. Combines parts
333 * of permission() and vfs_permission(), and tests ONLY for
334 * MAY_EXEC permission.
336 * If appropriate, check DAC only. If not appropriate, or
337 * short-cut DAC fails, then call permission() to do more
338 * complete permission check.
340 static inline int exec_permission_lite(struct inode *inode,
341 struct nameidata *nd)
343 umode_t mode = inode->i_mode;
345 if (inode->i_op && inode->i_op->permission)
348 if (current->fsuid == inode->i_uid)
350 else if (in_group_p(inode->i_gid))
356 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
359 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
362 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
367 return security_inode_permission(inode, MAY_EXEC, nd);
371 * This is called when everything else fails, and we actually have
372 * to go to the low-level filesystem to find out what we should do..
374 * We get the directory semaphore, and after getting that we also
375 * make sure that nobody added the entry to the dcache in the meantime..
378 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
380 struct dentry * result;
381 struct inode *dir = parent->d_inode;
385 * First re-do the cached lookup just in case it was created
386 * while we waited for the directory semaphore..
388 * FIXME! This could use version numbering or similar to
389 * avoid unnecessary cache lookups.
391 * The "dcache_lock" is purely to protect the RCU list walker
392 * from concurrent renames at this point (we mustn't get false
393 * negatives from the RCU list walk here, unlike the optimistic
396 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
398 result = d_lookup(parent, name);
400 struct dentry * dentry = d_alloc(parent, name);
401 result = ERR_PTR(-ENOMEM);
403 result = dir->i_op->lookup(dir, dentry, nd);
414 * Uhhuh! Nasty case: the cache was re-populated while
415 * we waited on the semaphore. Need to revalidate.
418 if (result->d_op && result->d_op->d_revalidate) {
419 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
421 result = ERR_PTR(-ENOENT);
427 static int __emul_lookup_dentry(const char *, struct nameidata *);
431 walk_init_root(const char *name, struct nameidata *nd)
433 read_lock(¤t->fs->lock);
434 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
435 nd->mnt = mntget(current->fs->altrootmnt);
436 nd->dentry = dget(current->fs->altroot);
437 read_unlock(¤t->fs->lock);
438 if (__emul_lookup_dentry(name,nd))
440 read_lock(¤t->fs->lock);
442 nd->mnt = mntget(current->fs->rootmnt);
443 nd->dentry = dget(current->fs->root);
444 read_unlock(¤t->fs->lock);
448 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
457 if (!walk_init_root(link, nd))
458 /* weird __emul_prefix() stuff did it */
461 res = link_path_walk(link, nd);
463 if (nd->depth || res || nd->last_type!=LAST_NORM)
466 * If it is an iterative symlinks resolution in open_namei() we
467 * have to copy the last component. And all that crap because of
468 * bloody create() on broken symlinks. Furrfu...
471 if (unlikely(!name)) {
475 strcpy(name, nd->last.name);
476 nd->last.name = name;
480 return PTR_ERR(link);
484 * This limits recursive symlink follows to 8, while
485 * limiting consecutive symlinks to 40.
487 * Without that kind of total limit, nasty chains of consecutive
488 * symlinks can cause almost arbitrarily long lookups.
490 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
493 if (current->link_count >= MAX_NESTED_LINKS)
495 if (current->total_link_count >= 40)
497 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
499 err = security_inode_follow_link(dentry, nd);
502 current->link_count++;
503 current->total_link_count++;
505 touch_atime(nd->mnt, dentry);
506 nd_set_link(nd, NULL);
507 err = dentry->d_inode->i_op->follow_link(dentry, nd);
509 char *s = nd_get_link(nd);
511 err = __vfs_follow_link(nd, s);
512 if (dentry->d_inode->i_op->put_link)
513 dentry->d_inode->i_op->put_link(dentry, nd);
515 current->link_count--;
523 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
525 struct vfsmount *parent;
526 struct dentry *mountpoint;
527 spin_lock(&vfsmount_lock);
528 parent=(*mnt)->mnt_parent;
529 if (parent == *mnt) {
530 spin_unlock(&vfsmount_lock);
534 mountpoint=dget((*mnt)->mnt_mountpoint);
535 spin_unlock(&vfsmount_lock);
537 *dentry = mountpoint;
543 /* no need for dcache_lock, as serialization is taken care in
546 static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
549 while (d_mountpoint(*dentry)) {
550 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
556 *dentry = dget(mounted->mnt_root);
562 /* no need for dcache_lock, as serialization is taken care in
565 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
567 struct vfsmount *mounted;
569 mounted = lookup_mnt(*mnt, *dentry);
574 *dentry = dget(mounted->mnt_root);
580 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
582 return __follow_down(mnt,dentry);
585 static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
588 struct vfsmount *parent;
589 struct dentry *old = *dentry;
591 read_lock(¤t->fs->lock);
592 if (*dentry == current->fs->root &&
593 *mnt == current->fs->rootmnt) {
594 read_unlock(¤t->fs->lock);
597 read_unlock(¤t->fs->lock);
598 spin_lock(&dcache_lock);
599 if (*dentry != (*mnt)->mnt_root) {
600 *dentry = dget((*dentry)->d_parent);
601 spin_unlock(&dcache_lock);
605 spin_unlock(&dcache_lock);
606 spin_lock(&vfsmount_lock);
607 parent = (*mnt)->mnt_parent;
608 if (parent == *mnt) {
609 spin_unlock(&vfsmount_lock);
613 *dentry = dget((*mnt)->mnt_mountpoint);
614 spin_unlock(&vfsmount_lock);
619 follow_mount(mnt, dentry);
623 struct vfsmount *mnt;
624 struct dentry *dentry;
628 * It's more convoluted than I'd like it to be, but... it's still fairly
629 * small and for now I'd prefer to have fast path as straight as possible.
630 * It _is_ time-critical.
632 static int do_lookup(struct nameidata *nd, struct qstr *name,
635 struct vfsmount *mnt = nd->mnt;
636 struct dentry *dentry = __d_lookup(nd->dentry, name);
640 if (dentry->d_op && dentry->d_op->d_revalidate)
641 goto need_revalidate;
644 path->dentry = dentry;
648 dentry = real_lookup(nd->dentry, name, nd);
654 if (dentry->d_op->d_revalidate(dentry, nd))
656 if (d_invalidate(dentry))
662 return PTR_ERR(dentry);
668 * This is the basic name resolution function, turning a pathname
669 * into the final dentry.
671 * We expect 'base' to be positive and a directory.
673 int fastcall link_path_walk(const char * name, struct nameidata *nd)
678 unsigned int lookup_flags = nd->flags;
680 atomic = (lookup_flags & LOOKUP_ATOMIC);
687 inode = nd->dentry->d_inode;
689 lookup_flags = LOOKUP_FOLLOW;
691 /* At this point we know we have a real path component. */
697 err = exec_permission_lite(inode, nd);
698 if (err == -EAGAIN) {
699 err = permission(inode, MAY_EXEC, nd);
705 c = *(const unsigned char *)name;
707 hash = init_name_hash();
710 hash = partial_name_hash(c, hash);
711 c = *(const unsigned char *)name;
712 } while (c && (c != '/'));
713 this.len = name - (const char *) this.name;
714 this.hash = end_name_hash(hash);
716 /* remove trailing slashes? */
719 while (*++name == '/');
721 goto last_with_slashes;
724 * "." and ".." are special - ".." especially so because it has
725 * to be able to know about the current root directory and
726 * parent relationships.
728 if (this.name[0] == '.') switch (this.len) {
732 if (this.name[1] != '.')
734 follow_dotdot(&nd->mnt, &nd->dentry);
735 inode = nd->dentry->d_inode;
741 * See if the low-level filesystem might want
742 * to use its own hash..
744 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
745 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
749 err = -EWOULDBLOCKIO;
752 nd->flags |= LOOKUP_CONTINUE;
753 /* This does the actual lookups.. */
754 err = do_lookup(nd, &this, &next);
757 /* Check mountpoints.. */
758 follow_mount(&next.mnt, &next.dentry);
761 inode = next.dentry->d_inode;
768 if (inode->i_op->follow_link) {
770 err = do_follow_link(next.dentry, nd);
776 inode = nd->dentry->d_inode;
785 nd->dentry = next.dentry;
788 if (!inode->i_op->lookup)
791 /* here ends the main loop */
794 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
796 nd->flags &= ~LOOKUP_CONTINUE;
797 if (lookup_flags & LOOKUP_PARENT)
799 if (this.name[0] == '.') switch (this.len) {
803 if (this.name[1] != '.')
805 follow_dotdot(&nd->mnt, &nd->dentry);
806 inode = nd->dentry->d_inode;
811 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
812 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
816 err = -EWOULDBLOCKIO;
819 err = do_lookup(nd, &this, &next);
822 follow_mount(&next.mnt, &next.dentry);
823 inode = next.dentry->d_inode;
824 if ((lookup_flags & LOOKUP_FOLLOW)
825 && inode && inode->i_op && inode->i_op->follow_link) {
827 err = do_follow_link(next.dentry, nd);
832 inode = nd->dentry->d_inode;
836 nd->dentry = next.dentry;
841 if (lookup_flags & LOOKUP_DIRECTORY) {
843 if (!inode->i_op || !inode->i_op->lookup)
849 nd->last_type = LAST_NORM;
850 if (this.name[0] != '.')
853 nd->last_type = LAST_DOT;
854 else if (this.len == 2 && this.name[1] == '.')
855 nd->last_type = LAST_DOTDOT;
860 * We bypassed the ordinary revalidation routines.
861 * We may need to check the cached dentry for staleness.
863 if (nd->dentry && nd->dentry->d_sb &&
864 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
866 /* Note: we do not d_invalidate() */
867 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
881 int fastcall path_walk(const char * name, struct nameidata *nd)
883 current->total_link_count = 0;
884 return link_path_walk(name, nd);
888 /* returns 1 if everything is done */
889 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
891 if (path_walk(name, nd))
892 return 0; /* something went wrong... */
894 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
895 struct dentry *old_dentry = nd->dentry;
896 struct vfsmount *old_mnt = nd->mnt;
897 struct qstr last = nd->last;
898 int last_type = nd->last_type;
900 * NAME was not found in alternate root or it's a directory. Try to find
901 * it in the normal root:
903 nd->last_type = LAST_ROOT;
904 read_lock(¤t->fs->lock);
905 nd->mnt = mntget(current->fs->rootmnt);
906 nd->dentry = dget(current->fs->root);
907 read_unlock(¤t->fs->lock);
908 if (path_walk(name, nd) == 0) {
909 if (nd->dentry->d_inode) {
916 nd->dentry = old_dentry;
919 nd->last_type = last_type;
924 void set_fs_altroot(void)
926 char *emul = __emul_prefix();
928 struct vfsmount *mnt = NULL, *oldmnt;
929 struct dentry *dentry = NULL, *olddentry;
934 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
940 write_lock(¤t->fs->lock);
941 oldmnt = current->fs->altrootmnt;
942 olddentry = current->fs->altroot;
943 current->fs->altrootmnt = mnt;
944 current->fs->altroot = dentry;
945 write_unlock(¤t->fs->lock);
952 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
956 nd->last_type = LAST_ROOT; /* if there are only slashes... */
960 read_lock(¤t->fs->lock);
962 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
963 nd->mnt = mntget(current->fs->altrootmnt);
964 nd->dentry = dget(current->fs->altroot);
965 read_unlock(¤t->fs->lock);
966 if (__emul_lookup_dentry(name,nd))
968 read_lock(¤t->fs->lock);
970 nd->mnt = mntget(current->fs->rootmnt);
971 nd->dentry = dget(current->fs->root);
973 nd->mnt = mntget(current->fs->pwdmnt);
974 nd->dentry = dget(current->fs->pwd);
976 read_unlock(¤t->fs->lock);
977 current->total_link_count = 0;
978 retval = link_path_walk(name, nd);
979 if (unlikely(current->audit_context
980 && nd && nd->dentry && nd->dentry->d_inode))
982 nd->dentry->d_inode->i_ino,
983 nd->dentry->d_inode->i_rdev);
988 * Restricted form of lookup. Doesn't follow links, single-component only,
989 * needs parent already locked. Doesn't follow mounts.
992 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
994 struct dentry * dentry;
998 inode = base->d_inode;
999 err = permission(inode, MAY_EXEC, nd);
1000 dentry = ERR_PTR(err);
1005 * See if the low-level filesystem might want
1006 * to use its own hash..
1008 if (base->d_op && base->d_op->d_hash) {
1009 err = base->d_op->d_hash(base, name);
1010 dentry = ERR_PTR(err);
1015 dentry = cached_lookup(base, name, nd);
1017 struct dentry *new = d_alloc(base, name);
1018 dentry = ERR_PTR(-ENOMEM);
1021 dentry = inode->i_op->lookup(inode, new, nd);
1031 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1033 return __lookup_hash(name, base, NULL);
1037 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1048 hash = init_name_hash();
1050 c = *(const unsigned char *)name++;
1051 if (c == '/' || c == '\0')
1053 hash = partial_name_hash(c, hash);
1055 this.hash = end_name_hash(hash);
1057 return lookup_hash(&this, base);
1059 return ERR_PTR(-EACCES);
1065 * is used by most simple commands to get the inode of a specified name.
1066 * Open, link etc use their own routines, but this is enough for things
1069 * namei exists in two versions: namei/lnamei. The only difference is
1070 * that namei follows links, while lnamei does not.
1073 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1075 char *tmp = getname(name);
1076 int err = PTR_ERR(tmp);
1079 err = path_lookup(tmp, flags, nd);
1086 * It's inline, so penalty for filesystems that don't use sticky bit is
1089 static inline int check_sticky(struct inode *dir, struct inode *inode)
1091 if (!(dir->i_mode & S_ISVTX))
1093 if (inode->i_uid == current->fsuid)
1095 if (dir->i_uid == current->fsuid)
1097 return !capable(CAP_FOWNER);
1101 * Check whether we can remove a link victim from directory dir, check
1102 * whether the type of victim is right.
1103 * 1. We can't do it if dir is read-only (done in permission())
1104 * 2. We should have write and exec permissions on dir
1105 * 3. We can't remove anything from append-only dir
1106 * 4. We can't do anything with immutable dir (done in permission())
1107 * 5. If the sticky bit on dir is set we should either
1108 * a. be owner of dir, or
1109 * b. be owner of victim, or
1110 * c. have CAP_FOWNER capability
1111 * 6. If the victim is append-only or immutable we can't do antyhing with
1112 * links pointing to it.
1113 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1114 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1115 * 9. We can't remove a root or mountpoint.
1116 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1117 * nfs_async_unlink().
1119 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1122 if (!victim->d_inode)
1124 if (victim->d_parent->d_inode != dir)
1127 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1132 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1133 IS_IXORUNLINK(victim->d_inode))
1136 if (!S_ISDIR(victim->d_inode->i_mode))
1138 if (IS_ROOT(victim))
1140 } else if (S_ISDIR(victim->d_inode->i_mode))
1142 if (IS_DEADDIR(dir))
1144 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1149 /* Check whether we can create an object with dentry child in directory
1151 * 1. We can't do it if child already exists (open has special treatment for
1152 * this case, but since we are inlined it's OK)
1153 * 2. We can't do it if dir is read-only (done in permission())
1154 * 3. We should have write and exec permissions on dir
1155 * 4. We can't do it if dir is immutable (done in permission())
1157 static inline int may_create(struct inode *dir, struct dentry *child,
1158 struct nameidata *nd)
1162 if (IS_DEADDIR(dir))
1164 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1168 * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1171 * O_DIRECTORY translates into forcing a directory lookup.
1173 static inline int lookup_flags(unsigned int f)
1175 unsigned long retval = LOOKUP_FOLLOW;
1178 retval &= ~LOOKUP_FOLLOW;
1180 if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1181 retval &= ~LOOKUP_FOLLOW;
1183 if (f & O_DIRECTORY)
1184 retval |= LOOKUP_DIRECTORY;
1185 if (f & O_ATOMICLOOKUP)
1186 retval |= LOOKUP_ATOMIC;
1192 * p1 and p2 should be directories on the same fs.
1194 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1199 down(&p1->d_inode->i_sem);
1203 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1205 for (p = p1; p->d_parent != p; p = p->d_parent) {
1206 if (p->d_parent == p2) {
1207 down(&p2->d_inode->i_sem);
1208 down(&p1->d_inode->i_sem);
1213 for (p = p2; p->d_parent != p; p = p->d_parent) {
1214 if (p->d_parent == p1) {
1215 down(&p1->d_inode->i_sem);
1216 down(&p2->d_inode->i_sem);
1221 down(&p1->d_inode->i_sem);
1222 down(&p2->d_inode->i_sem);
1226 void unlock_rename(struct dentry *p1, struct dentry *p2)
1228 up(&p1->d_inode->i_sem);
1230 up(&p2->d_inode->i_sem);
1231 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1235 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1236 struct nameidata *nd)
1238 int error = may_create(dir, dentry, nd);
1243 if (!dir->i_op || !dir->i_op->create)
1244 return -EACCES; /* shouldn't it be ENOSYS? */
1247 error = security_inode_create(dir, dentry, mode);
1251 error = dir->i_op->create(dir, dentry, mode, nd);
1253 inode_dir_notify(dir, DN_CREATE);
1254 security_inode_post_create(dir, dentry, mode);
1259 int may_open(struct nameidata *nd, int acc_mode, int flag)
1261 struct dentry *dentry = nd->dentry;
1262 struct inode *inode = dentry->d_inode;
1268 if (S_ISLNK(inode->i_mode))
1271 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1274 error = permission(inode, acc_mode, nd);
1279 * FIFO's, sockets and device files are special: they don't
1280 * actually live on the filesystem itself, and as such you
1281 * can write to them even if the filesystem is read-only.
1283 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1285 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1286 if (nd->mnt->mnt_flags & MNT_NODEV)
1290 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1293 * An append-only file must be opened in append mode for writing.
1295 if (IS_APPEND(inode)) {
1296 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1302 /* O_NOATIME can only be set by the owner or superuser */
1303 if (flag & O_NOATIME)
1304 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1308 * Ensure there are no outstanding leases on the file.
1310 error = break_lease(inode, flag);
1314 if (flag & O_TRUNC) {
1315 error = get_write_access(inode);
1320 * Refuse to truncate files with mandatory locks held on them.
1322 error = locks_verify_locked(inode);
1326 error = do_truncate(dentry, 0);
1328 put_write_access(inode);
1332 if (flag & FMODE_WRITE)
1341 * namei for open - this is in fact almost the whole open-routine.
1343 * Note that the low bits of "flag" aren't the same as in the open
1344 * system call - they are 00 - no permissions needed
1345 * 01 - read permission needed
1346 * 10 - write permission needed
1347 * 11 - read/write permissions needed
1348 * which is a lot more logical, and also allows the "no perm" needed
1349 * for symlinks (where the permissions are checked later).
1352 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1354 int acc_mode, error = 0;
1355 struct dentry *dentry;
1359 acc_mode = ACC_MODE(flag);
1361 /* Allow the LSM permission hook to distinguish append
1362 access from general write access. */
1363 if (flag & O_APPEND)
1364 acc_mode |= MAY_APPEND;
1366 /* Fill in the open() intent data */
1367 nd->intent.open.flags = flag;
1368 nd->intent.open.create_mode = mode;
1371 * The simplest case - just a plain lookup.
1373 if (!(flag & O_CREAT)) {
1374 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1381 * Create - we need to know the parent.
1383 error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1388 * We have the parent and last component. First of all, check
1389 * that we are not asked to creat(2) an obvious directory - that
1393 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1397 nd->flags &= ~LOOKUP_PARENT;
1398 down(&dir->d_inode->i_sem);
1399 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1402 error = PTR_ERR(dentry);
1403 if (IS_ERR(dentry)) {
1404 up(&dir->d_inode->i_sem);
1408 /* Negative dentry, just create the file */
1409 if (!dentry->d_inode) {
1410 if (!IS_POSIXACL(dir->d_inode))
1411 mode &= ~current->fs->umask;
1412 error = vfs_create(dir->d_inode, dentry, mode, nd);
1413 up(&dir->d_inode->i_sem);
1415 nd->dentry = dentry;
1418 /* Don't check for write permission, don't truncate */
1425 * It already exists.
1427 up(&dir->d_inode->i_sem);
1433 if (d_mountpoint(dentry)) {
1435 if (flag & O_NOFOLLOW)
1437 while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
1440 if (!dentry->d_inode)
1442 if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1446 nd->dentry = dentry;
1448 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1451 error = may_open(nd, acc_mode, flag);
1464 if (flag & O_NOFOLLOW)
1467 * This is subtle. Instead of calling do_follow_link() we do the
1468 * thing by hands. The reason is that this way we have zero link_count
1469 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1470 * After that we have the parent and last component, i.e.
1471 * we are in the same situation as after the first path_walk().
1472 * Well, almost - if the last component is normal we get its copy
1473 * stored in nd->last.name and we will have to putname() it when we
1474 * are done. Procfs-like symlinks just set LAST_BIND.
1476 nd->flags |= LOOKUP_PARENT;
1477 error = security_inode_follow_link(dentry, nd);
1480 touch_atime(nd->mnt, dentry);
1481 nd_set_link(nd, NULL);
1482 error = dentry->d_inode->i_op->follow_link(dentry, nd);
1484 char *s = nd_get_link(nd);
1486 error = __vfs_follow_link(nd, s);
1487 if (dentry->d_inode->i_op->put_link)
1488 dentry->d_inode->i_op->put_link(dentry, nd);
1493 nd->flags &= ~LOOKUP_PARENT;
1494 if (nd->last_type == LAST_BIND) {
1495 dentry = nd->dentry;
1499 if (nd->last_type != LAST_NORM)
1501 if (nd->last.name[nd->last.len]) {
1502 putname(nd->last.name);
1507 putname(nd->last.name);
1511 down(&dir->d_inode->i_sem);
1512 dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1513 putname(nd->last.name);
1518 * lookup_create - lookup a dentry, creating it if it doesn't exist
1519 * @nd: nameidata info
1520 * @is_dir: directory flag
1522 * Simple function to lookup and return a dentry and create it
1523 * if it doesn't exist. Is SMP-safe.
1525 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1527 struct dentry *dentry;
1529 down(&nd->dentry->d_inode->i_sem);
1530 dentry = ERR_PTR(-EEXIST);
1531 if (nd->last_type != LAST_NORM)
1533 nd->flags &= ~LOOKUP_PARENT;
1534 dentry = lookup_hash(&nd->last, nd->dentry);
1537 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1542 dentry = ERR_PTR(-ENOENT);
1547 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1549 int error = may_create(dir, dentry, NULL);
1554 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1557 if (!dir->i_op || !dir->i_op->mknod)
1560 error = security_inode_mknod(dir, dentry, mode, dev);
1565 error = dir->i_op->mknod(dir, dentry, mode, dev);
1567 inode_dir_notify(dir, DN_CREATE);
1568 security_inode_post_mknod(dir, dentry, mode, dev);
1573 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1577 struct dentry * dentry;
1578 struct nameidata nd;
1582 tmp = getname(filename);
1584 return PTR_ERR(tmp);
1586 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1589 dentry = lookup_create(&nd, 0);
1590 error = PTR_ERR(dentry);
1592 if (!IS_POSIXACL(nd.dentry->d_inode))
1593 mode &= ~current->fs->umask;
1594 if (!IS_ERR(dentry)) {
1595 switch (mode & S_IFMT) {
1596 case 0: case S_IFREG:
1597 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1599 case S_IFCHR: case S_IFBLK:
1600 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1601 new_decode_dev(dev));
1603 case S_IFIFO: case S_IFSOCK:
1604 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1614 up(&nd.dentry->d_inode->i_sem);
1622 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1624 int error = may_create(dir, dentry, NULL);
1629 if (!dir->i_op || !dir->i_op->mkdir)
1632 mode &= (S_IRWXUGO|S_ISVTX);
1633 error = security_inode_mkdir(dir, dentry, mode);
1638 error = dir->i_op->mkdir(dir, dentry, mode);
1640 inode_dir_notify(dir, DN_CREATE);
1641 security_inode_post_mkdir(dir,dentry, mode);
1646 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1651 tmp = getname(pathname);
1652 error = PTR_ERR(tmp);
1654 struct dentry *dentry;
1655 struct nameidata nd;
1657 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1660 dentry = lookup_create(&nd, 1);
1661 error = PTR_ERR(dentry);
1662 if (!IS_ERR(dentry)) {
1663 if (!IS_POSIXACL(nd.dentry->d_inode))
1664 mode &= ~current->fs->umask;
1665 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1668 up(&nd.dentry->d_inode->i_sem);
1678 * We try to drop the dentry early: we should have
1679 * a usage count of 2 if we're the only user of this
1680 * dentry, and if that is true (possibly after pruning
1681 * the dcache), then we drop the dentry now.
1683 * A low-level filesystem can, if it choses, legally
1686 * if (!d_unhashed(dentry))
1689 * if it cannot handle the case of removing a directory
1690 * that is still in use by something else..
1692 static void d_unhash(struct dentry *dentry)
1695 spin_lock(&dcache_lock);
1696 switch (atomic_read(&dentry->d_count)) {
1698 spin_unlock(&dcache_lock);
1699 shrink_dcache_parent(dentry);
1700 spin_lock(&dcache_lock);
1701 if (atomic_read(&dentry->d_count) != 2)
1706 spin_unlock(&dcache_lock);
1709 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1711 int error = may_delete(dir, dentry, 1);
1716 if (!dir->i_op || !dir->i_op->rmdir)
1721 down(&dentry->d_inode->i_sem);
1723 if (d_mountpoint(dentry))
1726 error = security_inode_rmdir(dir, dentry);
1728 error = dir->i_op->rmdir(dir, dentry);
1730 dentry->d_inode->i_flags |= S_DEAD;
1733 up(&dentry->d_inode->i_sem);
1735 inode_dir_notify(dir, DN_DELETE);
1743 asmlinkage long sys_rmdir(const char __user * pathname)
1747 struct dentry *dentry;
1748 struct nameidata nd;
1750 name = getname(pathname);
1752 return PTR_ERR(name);
1754 error = path_lookup(name, LOOKUP_PARENT, &nd);
1758 switch(nd.last_type) {
1769 down(&nd.dentry->d_inode->i_sem);
1770 dentry = lookup_hash(&nd.last, nd.dentry);
1771 error = PTR_ERR(dentry);
1772 if (!IS_ERR(dentry)) {
1773 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1776 up(&nd.dentry->d_inode->i_sem);
1784 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1786 int error = may_delete(dir, dentry, 0);
1791 if (!dir->i_op || !dir->i_op->unlink)
1796 down(&dentry->d_inode->i_sem);
1797 if (d_mountpoint(dentry))
1800 error = security_inode_unlink(dir, dentry);
1802 error = dir->i_op->unlink(dir, dentry);
1804 up(&dentry->d_inode->i_sem);
1806 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1807 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1809 inode_dir_notify(dir, DN_DELETE);
1815 * Make sure that the actual truncation of the file will occur outside its
1816 * directory's i_sem. Truncate can take a long time if there is a lot of
1817 * writeout happening, and we don't want to prevent access to the directory
1818 * while waiting on the I/O.
1820 asmlinkage long sys_unlink(const char __user * pathname)
1824 struct dentry *dentry;
1825 struct nameidata nd;
1826 struct inode *inode = NULL;
1828 name = getname(pathname);
1830 return PTR_ERR(name);
1832 error = path_lookup(name, LOOKUP_PARENT, &nd);
1836 if (nd.last_type != LAST_NORM)
1838 down(&nd.dentry->d_inode->i_sem);
1839 dentry = lookup_hash(&nd.last, nd.dentry);
1840 error = PTR_ERR(dentry);
1841 if (!IS_ERR(dentry)) {
1842 /* Why not before? Because we want correct error value */
1843 if (nd.last.name[nd.last.len])
1845 inode = dentry->d_inode;
1847 atomic_inc(&inode->i_count);
1848 error = vfs_unlink(nd.dentry->d_inode, dentry);
1852 up(&nd.dentry->d_inode->i_sem);
1859 iput(inode); /* truncate the inode here */
1863 error = !dentry->d_inode ? -ENOENT :
1864 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1868 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1870 int error = may_create(dir, dentry, NULL);
1875 if (!dir->i_op || !dir->i_op->symlink)
1878 error = security_inode_symlink(dir, dentry, oldname);
1883 error = dir->i_op->symlink(dir, dentry, oldname);
1885 inode_dir_notify(dir, DN_CREATE);
1886 security_inode_post_symlink(dir, dentry, oldname);
1891 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1897 from = getname(oldname);
1899 return PTR_ERR(from);
1900 to = getname(newname);
1901 error = PTR_ERR(to);
1903 struct dentry *dentry;
1904 struct nameidata nd;
1906 error = path_lookup(to, LOOKUP_PARENT, &nd);
1909 dentry = lookup_create(&nd, 0);
1910 error = PTR_ERR(dentry);
1911 if (!IS_ERR(dentry)) {
1912 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1915 up(&nd.dentry->d_inode->i_sem);
1924 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1926 struct inode *inode = old_dentry->d_inode;
1932 error = may_create(dir, new_dentry, NULL);
1936 if (dir->i_sb != inode->i_sb)
1940 * A link to an append-only or immutable file cannot be created.
1942 if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
1944 if (!dir->i_op || !dir->i_op->link)
1946 if (S_ISDIR(old_dentry->d_inode->i_mode))
1949 error = security_inode_link(old_dentry, dir, new_dentry);
1953 down(&old_dentry->d_inode->i_sem);
1955 error = dir->i_op->link(old_dentry, dir, new_dentry);
1956 up(&old_dentry->d_inode->i_sem);
1958 inode_dir_notify(dir, DN_CREATE);
1959 security_inode_post_link(old_dentry, dir, new_dentry);
1965 * Hardlinks are often used in delicate situations. We avoid
1966 * security-related surprises by not following symlinks on the
1969 * We don't follow them on the oldname either to be compatible
1970 * with linux 2.0, and to avoid hard-linking to directories
1971 * and other special files. --ADM
1973 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
1975 struct dentry *new_dentry;
1976 struct nameidata nd, old_nd;
1980 to = getname(newname);
1984 error = __user_walk(oldname, 0, &old_nd);
1987 error = path_lookup(to, LOOKUP_PARENT, &nd);
1991 if (old_nd.mnt != nd.mnt)
1993 new_dentry = lookup_create(&nd, 0);
1994 error = PTR_ERR(new_dentry);
1995 if (!IS_ERR(new_dentry)) {
1996 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
1999 up(&nd.dentry->d_inode->i_sem);
2003 path_release(&old_nd);
2011 * The worst of all namespace operations - renaming directory. "Perverted"
2012 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2014 * a) we can get into loop creation. Check is done in is_subdir().
2015 * b) race potential - two innocent renames can create a loop together.
2016 * That's where 4.4 screws up. Current fix: serialization on
2017 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2019 * c) we have to lock _three_ objects - parents and victim (if it exists).
2020 * And that - after we got ->i_sem on parents (until then we don't know
2021 * whether the target exists). Solution: try to be smart with locking
2022 * order for inodes. We rely on the fact that tree topology may change
2023 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2024 * move will be locked. Thus we can rank directories by the tree
2025 * (ancestors first) and rank all non-directories after them.
2026 * That works since everybody except rename does "lock parent, lookup,
2027 * lock child" and rename is under ->s_vfs_rename_sem.
2028 * HOWEVER, it relies on the assumption that any object with ->lookup()
2029 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2030 * we'd better make sure that there's no link(2) for them.
2031 * d) some filesystems don't support opened-but-unlinked directories,
2032 * either because of layout or because they are not ready to deal with
2033 * all cases correctly. The latter will be fixed (taking this sort of
2034 * stuff into VFS), but the former is not going away. Solution: the same
2035 * trick as in rmdir().
2036 * e) conversion from fhandle to dentry may come in the wrong moment - when
2037 * we are removing the target. Solution: we will have to grab ->i_sem
2038 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2039 * ->i_sem on parents, which works but leads to some truely excessive
2042 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2043 struct inode *new_dir, struct dentry *new_dentry)
2046 struct inode *target;
2049 * If we are going to change the parent - check write permissions,
2050 * we'll need to flip '..'.
2052 if (new_dir != old_dir) {
2053 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2058 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2062 target = new_dentry->d_inode;
2064 down(&target->i_sem);
2065 d_unhash(new_dentry);
2067 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2070 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2073 target->i_flags |= S_DEAD;
2075 if (d_unhashed(new_dentry))
2076 d_rehash(new_dentry);
2080 d_move(old_dentry,new_dentry);
2081 security_inode_post_rename(old_dir, old_dentry,
2082 new_dir, new_dentry);
2087 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2088 struct inode *new_dir, struct dentry *new_dentry)
2090 struct inode *target;
2093 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2098 target = new_dentry->d_inode;
2100 down(&target->i_sem);
2101 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2104 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2106 /* The following d_move() should become unconditional */
2107 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2108 d_move(old_dentry, new_dentry);
2109 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2117 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2118 struct inode *new_dir, struct dentry *new_dentry)
2121 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2123 if (old_dentry->d_inode == new_dentry->d_inode)
2126 error = may_delete(old_dir, old_dentry, is_dir);
2130 if (!new_dentry->d_inode)
2131 error = may_create(new_dir, new_dentry, NULL);
2133 error = may_delete(new_dir, new_dentry, is_dir);
2137 if (!old_dir->i_op || !old_dir->i_op->rename)
2140 DQUOT_INIT(old_dir);
2141 DQUOT_INIT(new_dir);
2144 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2146 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2148 if (old_dir == new_dir)
2149 inode_dir_notify(old_dir, DN_RENAME);
2151 inode_dir_notify(old_dir, DN_DELETE);
2152 inode_dir_notify(new_dir, DN_CREATE);
2158 static inline int do_rename(const char * oldname, const char * newname)
2161 struct dentry * old_dir, * new_dir;
2162 struct dentry * old_dentry, *new_dentry;
2163 struct dentry * trap;
2164 struct nameidata oldnd, newnd;
2166 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2170 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2175 if (oldnd.mnt != newnd.mnt)
2178 old_dir = oldnd.dentry;
2180 if (oldnd.last_type != LAST_NORM)
2183 new_dir = newnd.dentry;
2184 if (newnd.last_type != LAST_NORM)
2187 trap = lock_rename(new_dir, old_dir);
2189 old_dentry = lookup_hash(&oldnd.last, old_dir);
2190 error = PTR_ERR(old_dentry);
2191 if (IS_ERR(old_dentry))
2193 /* source must exist */
2195 if (!old_dentry->d_inode)
2197 /* unless the source is a directory trailing slashes give -ENOTDIR */
2198 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2200 if (oldnd.last.name[oldnd.last.len])
2202 if (newnd.last.name[newnd.last.len])
2205 /* source should not be ancestor of target */
2207 if (old_dentry == trap)
2209 new_dentry = lookup_hash(&newnd.last, new_dir);
2210 error = PTR_ERR(new_dentry);
2211 if (IS_ERR(new_dentry))
2213 /* target should not be an ancestor of source */
2215 if (new_dentry == trap)
2218 error = vfs_rename(old_dir->d_inode, old_dentry,
2219 new_dir->d_inode, new_dentry);
2225 unlock_rename(new_dir, old_dir);
2227 path_release(&newnd);
2229 path_release(&oldnd);
2234 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2240 from = getname(oldname);
2242 return PTR_ERR(from);
2243 to = getname(newname);
2244 error = PTR_ERR(to);
2246 error = do_rename(from,to);
2253 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2257 len = PTR_ERR(link);
2262 if (len > (unsigned) buflen)
2264 if (copy_to_user(buffer, link, len))
2271 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2272 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2273 * using) it for any given inode is up to filesystem.
2275 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2277 struct nameidata nd;
2280 res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2282 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2283 if (dentry->d_inode->i_op->put_link)
2284 dentry->d_inode->i_op->put_link(dentry, &nd);
2289 int vfs_follow_link(struct nameidata *nd, const char *link)
2291 return __vfs_follow_link(nd, link);
2294 /* get the link contents into pagecache */
2295 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2298 struct address_space *mapping = dentry->d_inode->i_mapping;
2299 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2303 wait_on_page_locked(page);
2304 if (!PageUptodate(page))
2310 page_cache_release(page);
2311 return ERR_PTR(-EIO);
2317 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2319 struct page *page = NULL;
2320 char *s = page_getlink(dentry, &page);
2321 int res = vfs_readlink(dentry,buffer,buflen,s);
2324 page_cache_release(page);
2329 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2332 nd_set_link(nd, page_getlink(dentry, &page));
2336 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2338 if (!IS_ERR(nd_get_link(nd))) {
2340 page = find_get_page(dentry->d_inode->i_mapping, 0);
2344 page_cache_release(page);
2345 page_cache_release(page);
2349 int page_follow_link(struct dentry *dentry, struct nameidata *nd)
2351 struct page *page = NULL;
2352 char *s = page_getlink(dentry, &page);
2353 int res = __vfs_follow_link(nd, s);
2356 page_cache_release(page);
2361 int page_symlink(struct inode *inode, const char *symname, int len)
2363 struct address_space *mapping = inode->i_mapping;
2364 struct page *page = grab_cache_page(mapping, 0);
2370 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2373 kaddr = kmap_atomic(page, KM_USER0);
2374 memcpy(kaddr, symname, len-1);
2375 kunmap_atomic(kaddr, KM_USER0);
2376 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2378 * Notice that we are _not_ going to block here - end of page is
2379 * unmapped, so this will only try to map the rest of page, see
2380 * that it is unmapped (typically even will not look into inode -
2381 * ->i_size will be enough for everything) and zero it out.
2382 * OTOH it's obviously correct and should make the page up-to-date.
2384 if (!PageUptodate(page)) {
2385 err = mapping->a_ops->readpage(NULL, page);
2386 wait_on_page_locked(page);
2390 page_cache_release(page);
2393 mark_inode_dirty(inode);
2397 page_cache_release(page);
2402 struct inode_operations page_symlink_inode_operations = {
2403 .readlink = generic_readlink,
2404 .follow_link = page_follow_link_light,
2405 .put_link = page_put_link,
2408 EXPORT_SYMBOL(__user_walk);
2409 EXPORT_SYMBOL(follow_down);
2410 EXPORT_SYMBOL(follow_up);
2411 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2412 EXPORT_SYMBOL(getname);
2413 EXPORT_SYMBOL(lock_rename);
2414 EXPORT_SYMBOL(lookup_create);
2415 EXPORT_SYMBOL(lookup_hash);
2416 EXPORT_SYMBOL(lookup_one_len);
2417 EXPORT_SYMBOL(page_follow_link);
2418 EXPORT_SYMBOL(page_follow_link_light);
2419 EXPORT_SYMBOL(page_put_link);
2420 EXPORT_SYMBOL(page_readlink);
2421 EXPORT_SYMBOL(page_symlink);
2422 EXPORT_SYMBOL(page_symlink_inode_operations);
2423 EXPORT_SYMBOL(path_lookup);
2424 EXPORT_SYMBOL(path_release);
2425 EXPORT_SYMBOL(path_walk);
2426 EXPORT_SYMBOL(permission);
2427 EXPORT_SYMBOL(unlock_rename);
2428 EXPORT_SYMBOL(vfs_create);
2429 EXPORT_SYMBOL(vfs_follow_link);
2430 EXPORT_SYMBOL(vfs_link);
2431 EXPORT_SYMBOL(vfs_mkdir);
2432 EXPORT_SYMBOL(vfs_mknod);
2433 EXPORT_SYMBOL(vfs_permission);
2434 EXPORT_SYMBOL(vfs_readlink);
2435 EXPORT_SYMBOL(vfs_rename);
2436 EXPORT_SYMBOL(vfs_rmdir);
2437 EXPORT_SYMBOL(vfs_symlink);
2438 EXPORT_SYMBOL(vfs_unlink);
2439 EXPORT_SYMBOL(generic_readlink);