vserver 2.0 rc7
[linux-2.6.git] / fs / namei.c
index bef7d99..30d31d2 100644 (file)
@@ -31,7 +31,6 @@
 #include <linux/proc_fs.h>
 #include <linux/vserver/inode.h>
 #include <linux/vserver/debug.h>
-
 #include <asm/namei.h>
 #include <asm/uaccess.h>
 
@@ -152,11 +151,22 @@ char * getname(const char __user * filename)
                        result = ERR_PTR(retval);
                }
        }
-       if (unlikely(current->audit_context) && !IS_ERR(result) && result)
-               audit_getname(result);
+       audit_getname(result);
        return result;
 }
 
+#ifdef CONFIG_AUDITSYSCALL
+void putname(const char *name)
+{
+       if (unlikely(current->audit_context))
+               audit_putname(name);
+       else
+               __putname(name);
+}
+EXPORT_SYMBOL(putname);
+#endif
+
+
 /**
  * generic_permission  -  check for access rights on a Posix-like filesystem
  * @inode:     inode to check access rights for
@@ -506,12 +516,21 @@ fail:
        return PTR_ERR(link);
 }
 
-static inline int __do_follow_link(struct dentry *dentry, struct nameidata *nd)
+struct path {
+       struct vfsmount *mnt;
+       struct dentry *dentry;
+};
+
+static inline int __do_follow_link(struct path *path, struct nameidata *nd)
 {
        int error;
+       struct dentry *dentry = path->dentry;
 
-       touch_atime(nd->mnt, dentry);
+       touch_atime(path->mnt, dentry);
        nd_set_link(nd, NULL);
+
+       if (path->mnt == nd->mnt)
+               mntget(path->mnt);
        error = dentry->d_inode->i_op->follow_link(dentry, nd);
        if (!error) {
                char *s = nd_get_link(nd);
@@ -520,6 +539,8 @@ static inline int __do_follow_link(struct dentry *dentry, struct nameidata *nd)
                if (dentry->d_inode->i_op->put_link)
                        dentry->d_inode->i_op->put_link(dentry, nd);
        }
+       dput(dentry);
+       mntput(path->mnt);
 
        return error;
 }
@@ -531,7 +552,7 @@ static inline int __do_follow_link(struct dentry *dentry, struct nameidata *nd)
  * Without that kind of total limit, nasty chains of consecutive
  * symlinks can cause almost arbitrarily long lookups. 
  */
-static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
+static inline int do_follow_link(struct path *path, struct nameidata *nd)
 {
        int err = -ELOOP;
        if (current->link_count >= MAX_NESTED_LINKS)
@@ -540,17 +561,20 @@ static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
                goto loop;
        BUG_ON(nd->depth >= MAX_NESTED_LINKS);
        cond_resched();
-       err = security_inode_follow_link(dentry, nd);
+       err = security_inode_follow_link(path->dentry, nd);
        if (err)
                goto loop;
        current->link_count++;
        current->total_link_count++;
        nd->depth++;
-       err = __do_follow_link(dentry, nd);
+       err = __do_follow_link(path, nd);
        current->link_count--;
        nd->depth--;
        return err;
 loop:
+       dput(path->dentry);
+       if (path->mnt != nd->mnt)
+               mntput(path->mnt);
        path_release(nd);
        return err;
 }
@@ -578,87 +602,92 @@ int follow_up(struct vfsmount **mnt, struct dentry **dentry)
 /* no need for dcache_lock, as serialization is taken care in
  * namespace.c
  */
-static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
+static int __follow_mount(struct path *path)
 {
        int res = 0;
+       while (d_mountpoint(path->dentry)) {
+               struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
+               if (!mounted)
+                       break;
+               dput(path->dentry);
+               if (res)
+                       mntput(path->mnt);
+               path->mnt = mounted;
+               path->dentry = dget(mounted->mnt_root);
+               res = 1;
+       }
+       return res;
+}
+
+static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
+{
        while (d_mountpoint(*dentry)) {
                struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
                if (!mounted)
                        break;
+               dput(*dentry);
                mntput(*mnt);
                *mnt = mounted;
-               dput(*dentry);
                *dentry = dget(mounted->mnt_root);
-               res = 1;
        }
-       return res;
 }
 
 /* no need for dcache_lock, as serialization is taken care in
  * namespace.c
  */
-static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
+int follow_down(struct vfsmount **mnt, struct dentry **dentry)
 {
        struct vfsmount *mounted;
 
        mounted = lookup_mnt(*mnt, *dentry);
        if (mounted) {
+               dput(*dentry);
                mntput(*mnt);
                *mnt = mounted;
-               dput(*dentry);
                *dentry = dget(mounted->mnt_root);
                return 1;
        }
        return 0;
 }
 
-int follow_down(struct vfsmount **mnt, struct dentry **dentry)
-{
-       return __follow_down(mnt,dentry);
-}
-static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
+static inline void follow_dotdot(struct nameidata *nd)
 {
        while(1) {
                struct vfsmount *parent;
-               struct dentry *old = *dentry;
+               struct dentry *old = nd->dentry;
 
                 read_lock(&current->fs->lock);
-               if (*dentry == current->fs->root &&
-                   *mnt == current->fs->rootmnt) {
+               if (nd->dentry == current->fs->root &&
+                   nd->mnt == current->fs->rootmnt) {
                         read_unlock(&current->fs->lock);
+                       /* for sane '/' avoid follow_mount() */
                        return;
                }
                 read_unlock(&current->fs->lock);
                spin_lock(&dcache_lock);
-               if (*dentry != (*mnt)->mnt_root) {
-                       *dentry = dget((*dentry)->d_parent);
+               if (nd->dentry != nd->mnt->mnt_root) {
+                       nd->dentry = dget(nd->dentry->d_parent);
                        spin_unlock(&dcache_lock);
                        dput(old);
                        break;
                }
                spin_unlock(&dcache_lock);
                spin_lock(&vfsmount_lock);
-               parent = (*mnt)->mnt_parent;
-               if (parent == *mnt) {
+               parent = nd->mnt->mnt_parent;
+               if (parent == nd->mnt) {
                        spin_unlock(&vfsmount_lock);
                        break;
                }
                mntget(parent);
-               *dentry = dget((*mnt)->mnt_mountpoint);
+               nd->dentry = dget(nd->mnt->mnt_mountpoint);
                spin_unlock(&vfsmount_lock);
                dput(old);
-               mntput(*mnt);
-               *mnt = parent;
+               mntput(nd->mnt);
+               nd->mnt = parent;
        }
-       follow_mount(mnt, dentry);
+       follow_mount(&nd->mnt, &nd->dentry);
 }
 
-struct path {
-       struct vfsmount *mnt;
-       struct dentry *dentry;
-};
-
 /*
  *  It's more convoluted than I'd like it to be, but... it's still fairly
  *  small and for now I'd prefer to have fast path as straight as possible.
@@ -689,6 +718,7 @@ static int do_lookup(struct nameidata *nd, struct qstr *name,
 done:
        path->mnt = mnt;
        path->dentry = dentry;
+       __follow_mount(path);
        return 0;
 hidden:
        vxwprintk(1, "xid=%d did lookup hidden %p[#%d,%lu] »%s«.",
@@ -717,13 +747,13 @@ fail:
 
 /*
  * Name resolution.
+ * This is the basic name resolution function, turning a pathname into
+ * the final dentry. We expect 'base' to be positive and a directory.
  *
- * This is the basic name resolution function, turning a pathname
- * into the final dentry.
- *
- * We expect 'base' to be positive and a directory.
+ * Returns 0 and nd will have valid dentry and mnt on success.
+ * Returns error and drops reference to input namei data on failure.
  */
-int fastcall link_path_walk(const char * name, struct nameidata *nd)
+static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
 {
        struct path next;
        struct inode *inode;
@@ -782,7 +812,7 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                        case 2: 
                                if (this.name[1] != '.')
                                        break;
-                               follow_dotdot(&nd->mnt, &nd->dentry);
+                               follow_dotdot(nd);
                                inode = nd->dentry->d_inode;
                                /* fallthrough */
                        case 1:
@@ -802,8 +832,6 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                err = do_lookup(nd, &this, &next);
                if (err)
                        break;
-               /* Check mountpoints.. */
-               follow_mount(&next.mnt, &next.dentry);
 
                err = -ENOENT;
                inode = next.dentry->d_inode;
@@ -814,10 +842,7 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                        goto out_dput;
 
                if (inode->i_op->follow_link) {
-                       mntget(next.mnt);
-                       err = do_follow_link(next.dentry, nd);
-                       dput(next.dentry);
-                       mntput(next.mnt);
+                       err = do_follow_link(&next, nd);
                        if (err)
                                goto return_err;
                        err = -ENOENT;
@@ -829,6 +854,8 @@ int fastcall link_path_walk(const char * name, struct nameidata *nd)
                                break;
                } else {
                        dput(nd->dentry);
+                       if (nd->mnt != next.mnt)
+                               mntput(nd->mnt);
                        nd->mnt = next.mnt;
                        nd->dentry = next.dentry;
                }
@@ -850,7 +877,7 @@ last_component:
                        case 2: 
                                if (this.name[1] != '.')
                                        break;
-                               follow_dotdot(&nd->mnt, &nd->dentry);
+                               follow_dotdot(nd);
                                inode = nd->dentry->d_inode;
                                /* fallthrough */
                        case 1:
@@ -864,19 +891,17 @@ last_component:
                err = do_lookup(nd, &this, &next);
                if (err)
                        break;
-               follow_mount(&next.mnt, &next.dentry);
                inode = next.dentry->d_inode;
                if ((lookup_flags & LOOKUP_FOLLOW)
                    && inode && inode->i_op && inode->i_op->follow_link) {
-                       mntget(next.mnt);
-                       err = do_follow_link(next.dentry, nd);
-                       dput(next.dentry);
-                       mntput(next.mnt);
+                       err = do_follow_link(&next, nd);
                        if (err)
                                goto return_err;
                        inode = nd->dentry->d_inode;
                } else {
                        dput(nd->dentry);
+                       if (nd->mnt != next.mnt)
+                               mntput(nd->mnt);
                        nd->mnt = next.mnt;
                        nd->dentry = next.dentry;
                }
@@ -916,6 +941,8 @@ return_base:
                return 0;
 out_dput:
                dput(next.dentry);
+               if (nd->mnt != next.mnt)
+                       mntput(next.mnt);
                break;
        }
        path_release(nd);
@@ -923,14 +950,47 @@ return_err:
        return err;
 }
 
+/*
+ * Wrapper to retry pathname resolution whenever the underlying
+ * file system returns an ESTALE.
+ *
+ * Retry the whole path once, forcing real lookup requests
+ * instead of relying on the dcache.
+ */
+int fastcall link_path_walk(const char *name, struct nameidata *nd)
+{
+       struct nameidata save = *nd;
+       int result;
+
+       /* make sure the stuff we saved doesn't go away */
+       dget(save.dentry);
+       mntget(save.mnt);
+
+       result = __link_path_walk(name, nd);
+       if (result == -ESTALE) {
+               *nd = save;
+               dget(nd->dentry);
+               mntget(nd->mnt);
+               nd->flags |= LOOKUP_REVAL;
+               result = __link_path_walk(name, nd);
+       }
+
+       dput(save.dentry);
+       mntput(save.mnt);
+
+       return result;
+}
+
 int fastcall path_walk(const char * name, struct nameidata *nd)
 {
        current->total_link_count = 0;
        return link_path_walk(name, nd);
 }
 
-/* SMP-safe */
-/* returns 1 if everything is done */
+/* 
+ * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
+ * everything is done. Returns 0 and drops input nd, if lookup failed;
+ */
 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
 {
        if (path_walk(name, nd))
@@ -994,9 +1054,10 @@ set_it:
        }
 }
 
+/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
 {
-       int retval;
+       int retval = 0;
 
        nd->last_type = LAST_ROOT; /* if there are only slashes... */
        nd->flags = flags;
@@ -1009,7 +1070,7 @@ int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata
                        nd->dentry = dget(current->fs->altroot);
                        read_unlock(&current->fs->lock);
                        if (__emul_lookup_dentry(name,nd))
-                               return 0;
+                               goto out; /* found in altroot */
                        read_lock(&current->fs->lock);
                }
                nd->mnt = mntget(current->fs->rootmnt);
@@ -1021,11 +1082,10 @@ int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata
        read_unlock(&current->fs->lock);
        current->total_link_count = 0;
        retval = link_path_walk(name, nd);
+out:
        if (unlikely(current->audit_context
                     && nd && nd->dentry && nd->dentry->d_inode))
-               audit_inode(name,
-                           nd->dentry->d_inode->i_ino,
-                           nd->dentry->d_inode->i_rdev);
+               audit_inode(name, nd->dentry->d_inode);
        return retval;
 }
 
@@ -1396,7 +1456,7 @@ int may_open(struct nameidata *nd, int acc_mode, int flag)
 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
 {
        int acc_mode, error = 0;
-       struct dentry *dentry;
+       struct path path;
        struct dentry *dir;
        int count = 0;
 
@@ -1440,23 +1500,24 @@ int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
        dir = nd->dentry;
        nd->flags &= ~LOOKUP_PARENT;
        down(&dir->d_inode->i_sem);
-       dentry = __lookup_hash(&nd->last, nd->dentry, nd);
+       path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
+       path.mnt = nd->mnt;
 
 do_last:
-       error = PTR_ERR(dentry);
-       if (IS_ERR(dentry)) {
+       error = PTR_ERR(path.dentry);
+       if (IS_ERR(path.dentry)) {
                up(&dir->d_inode->i_sem);
                goto exit;
        }
 
        /* Negative dentry, just create the file */
-       if (!dentry->d_inode) {
+       if (!path.dentry->d_inode) {
                if (!IS_POSIXACL(dir->d_inode))
                        mode &= ~current->fs->umask;
-               error = vfs_create(dir->d_inode, dentry, mode, nd);
+               error = vfs_create(dir->d_inode, path.dentry, mode, nd);
                up(&dir->d_inode->i_sem);
                dput(nd->dentry);
-               nd->dentry = dentry;
+               nd->dentry = path.dentry;
                if (error)
                        goto exit;
                /* Don't check for write permission, don't truncate */
@@ -1474,22 +1535,24 @@ do_last:
        if (flag & O_EXCL)
                goto exit_dput;
 
-       if (d_mountpoint(dentry)) {
+       if (__follow_mount(&path)) {
                error = -ELOOP;
                if (flag & O_NOFOLLOW)
                        goto exit_dput;
-               while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
        }
        error = -ENOENT;
-       if (!dentry->d_inode)
+       if (!path.dentry->d_inode)
                goto exit_dput;
-       if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
+       if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
                goto do_link;
 
        dput(nd->dentry);
-       nd->dentry = dentry;
+       nd->dentry = path.dentry;
+       if (nd->mnt != path.mnt)
+               mntput(nd->mnt);
+       nd->mnt = path.mnt;
        error = -EISDIR;
-       if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
+       if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
                goto exit;
 ok:
        error = may_open(nd, acc_mode, flag);
@@ -1498,7 +1561,9 @@ ok:
        return 0;
 
 exit_dput:
-       dput(dentry);
+       dput(path.dentry);
+       if (nd->mnt != path.mnt)
+               mntput(path.mnt);
 exit:
        path_release(nd);
        return error;
@@ -1518,18 +1583,15 @@ do_link:
         * are done. Procfs-like symlinks just set LAST_BIND.
         */
        nd->flags |= LOOKUP_PARENT;
-       error = security_inode_follow_link(dentry, nd);
+       error = security_inode_follow_link(path.dentry, nd);
        if (error)
                goto exit_dput;
-       error = __do_follow_link(dentry, nd);
-       dput(dentry);
+       error = __do_follow_link(&path, nd);
        if (error)
                return error;
        nd->flags &= ~LOOKUP_PARENT;
-       if (nd->last_type == LAST_BIND) {
-               dentry = nd->dentry;
+       if (nd->last_type == LAST_BIND)
                goto ok;
-       }
        error = -EISDIR;
        if (nd->last_type != LAST_NORM)
                goto exit;
@@ -1544,7 +1606,8 @@ do_link:
        }
        dir = nd->dentry;
        down(&dir->d_inode->i_sem);
-       dentry = __lookup_hash(&nd->last, nd->dentry, nd);
+       path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
+       path.mnt = nd->mnt;
        putname(nd->last.name);
        goto do_last;
 }
@@ -1578,6 +1641,7 @@ enoent:
 fail:
        return dentry;
 }
+EXPORT_SYMBOL_GPL(lookup_create);
 
 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
 {
@@ -1727,17 +1791,13 @@ out:
 void dentry_unhash(struct dentry *dentry)
 {
        dget(dentry);
-       spin_lock(&dcache_lock);
-       switch (atomic_read(&dentry->d_count)) {
-       default:
-               spin_unlock(&dcache_lock);
+       if (atomic_read(&dentry->d_count))
                shrink_dcache_parent(dentry);
-               spin_lock(&dcache_lock);
-               if (atomic_read(&dentry->d_count) != 2)
-                       break;
-       case 2:
+       spin_lock(&dcache_lock);
+       spin_lock(&dentry->d_lock);
+       if (atomic_read(&dentry->d_count) == 2)
                __d_drop(dentry);
-       }
+       spin_unlock(&dentry->d_lock);
        spin_unlock(&dcache_lock);
 }
 
@@ -2073,8 +2133,8 @@ exit:
  *        ->i_sem on parents, which works but leads to some truely excessive
  *        locking].
  */
-int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
-              struct inode *new_dir, struct dentry *new_dentry)
+static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
+                         struct inode *new_dir, struct dentry *new_dentry)
 {
        int error = 0;
        struct inode *target;
@@ -2118,8 +2178,8 @@ int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
        return error;
 }
 
-int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
-              struct inode *new_dir, struct dentry *new_dentry)
+static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
+                           struct inode *new_dir, struct dentry *new_dentry)
 {
        struct inode *target;
        int error;