[PATCH] BKL shifted into ->mknod()

Documentation/filesystems/Locking

@@ -52,7 +52,7 @@ locking rules:
 lookup:		no	yes
 create:		no	yes
 link:		yes	yes
-mknod:		yes	yes
+mknod:		no	yes
 mkdir:		yes	yes
 unlink:		yes	yes (both)
 rmdir:		yes	yes (both)	(see below)

Documentation/filesystems/porting

@@ -80,17 +80,8 @@ can relax your locking.
 ---
 [mandatory]
 
-->lookup() is called without BKL now.  Grab it on the entry, drop upon return
-- that will guarantee the same locking you used to have.  If your ->lookup()
-or its parts do not need BKL - better yet, now you can shift lock_kernel()/
-unlock_kernel() so that they would protect exactly what needs to be protected.
-
----
-[mandatory]
-
-->truncate() is called without BKL now (same as above).
-
----
-[mandatory]
-
-->create() is called without BKL now (same as above).
+->lookup(), ->truncate(), ->create() and ->mknod() are called without BKL now.
+Grab it on the entry, drop upon return - that will guarantee the same
+locking you used to have.  If your ->method or its parts do not need
+BKL - better yet, now you can shift lock_kernel()/ unlock_kernel()
+so that they would protect exactly what needs to be protected.

fs/coda/dir.c

@@ -267,13 +267,16 @@ static int coda_mknod(struct inode *dir, struct dentry *de, int mode, int rdev)
 	if ( coda_hasmknod == 0 )
 		return -EIO;
 
+	lock_kernel();
 	coda_vfs_stat.create++;
 
 	CDEBUG(D_INODE, "name: %s, length %d, mode %o, rdev %x\n",
 	       name, length, mode, rdev);
 
-	if (coda_isroot(dir) && coda_iscontrol(name, length))
+	if (coda_isroot(dir) && coda_iscontrol(name, length)) {
+		unlock_kernel();
 		return -EPERM;
+	}
 
 	error = venus_create(dir->i_sb, coda_i2f(dir), name, length, 
 				0, mode, rdev, &newfid, &attrs);
@@ -282,6 +285,7 @@ static int coda_mknod(struct inode *dir, struct dentry *de, int mode, int rdev)
 		CDEBUG(D_INODE, "mknod: %s, result %d\n",
 		       coda_f2s(&newfid), error); 
 		d_drop(de);
+		unlock_kernel();
 		return error;
 	}
 
@@ -289,11 +293,13 @@ static int coda_mknod(struct inode *dir, struct dentry *de, int mode, int rdev)
 	if ( error ) {
 		d_drop(de);
 		result = NULL;
+		unlock_kernel();
 		return error;
 	}
 
 	/* invalidate the directory cnode's attributes */
 	coda_dir_changed(dir, 0);
+	unlock_kernel();
 	d_instantiate(de, result);
         return 0;
 }			     

fs/devfs/base.c

@@ -3151,31 +3151,43 @@ static int devfs_mknod (struct inode *dir, struct dentry *dentry, int mode,
 
     DPRINTK (DEBUG_I_MKNOD, "(%s): mode: 0%o  dev: %d\n",
 	     dentry->d_name.name, mode, rdev);
+    lock_kernel();
     parent = get_devfs_entry_from_vfs_inode (dir);
-    if (parent == NULL) return -ENOENT;
+    if (parent == NULL) {
+	unlock_kernel();
+	return -ENOENT;
+    }
     de = _devfs_alloc_entry (dentry->d_name.name, dentry->d_name.len, mode);
-    if (!de) return -ENOMEM;
+    if (!de) {
+	unlock_kernel();
+	return -ENOMEM;
+    }
     de->vfs_deletable = TRUE;
     if ( S_ISBLK (mode) || S_ISCHR (mode) )
     {
 	de->u.fcb.u.device.major = MAJOR (rdev);
 	de->u.fcb.u.device.minor = MINOR (rdev);
     }
-    if ( ( err = _devfs_append_entry (parent, de, FALSE, NULL) ) != 0 )
+    if ( ( err = _devfs_append_entry (parent, de, FALSE, NULL) ) != 0 ) {
+	unlock_kernel();
 	return err;
+    }
     de->inode.uid = current->euid;
     de->inode.gid = current->egid;
     de->inode.atime = CURRENT_TIME;
     de->inode.mtime = CURRENT_TIME;
     de->inode.ctime = CURRENT_TIME;
-    if ( ( inode = _devfs_get_vfs_inode (dir->i_sb, de, dentry) ) == NULL )
+    if ( ( inode = _devfs_get_vfs_inode (dir->i_sb, de, dentry) ) == NULL ) {
+	unlock_kernel();
 	return -ENOMEM;
+    }
     DPRINTK (DEBUG_I_MKNOD, ":   new VFS inode(%u): %p  dentry: %p\n",
 	     de->inode.ino, inode, dentry);
     d_instantiate (dentry, inode);
     if ( !is_devfsd_or_child (fs_info) )
 	devfsd_notify_de (de, DEVFSD_NOTIFY_CREATE, inode->i_mode,
 			  inode->i_uid, inode->i_gid, fs_info, 0);
+    unlock_kernel();
     return 0;
 }   /*  End Function devfs_mknod  */
 

fs/ext3/namei.c

@@ -487,9 +487,12 @@ static int ext3_mknod (struct inode * dir, struct dentry *dentry,
 	struct inode *inode;
 	int err;
 
+	lock_kernel();
 	handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS + 3);
-	if (IS_ERR(handle))
+	if (IS_ERR(handle)) {
+		unlock_kernel();
 		return PTR_ERR(handle);
+	}
 
 	if (IS_SYNC(dir))
 		handle->h_sync = 1;
@@ -502,6 +505,7 @@ static int ext3_mknod (struct inode * dir, struct dentry *dentry,
 		err = ext3_add_nondir(handle, dentry, inode);
 	}
 	ext3_journal_stop(handle, dir);
+	unlock_kernel();
 	return err;
 }
 
@@ -1129,6 +1133,6 @@ struct inode_operations ext3_dir_inode_operations = {
 	symlink:	ext3_symlink,		/* BKL held */
 	mkdir:		ext3_mkdir,		/* BKL held */
 	rmdir:		ext3_rmdir,		/* BKL held */
-	mknod:		ext3_mknod,		/* BKL held */
+	mknod:		ext3_mknod,
 	rename:		ext3_rename,		/* BKL held */
 };

fs/hpfs/namei.c

@@ -182,6 +182,7 @@ int hpfs_mknod(struct inode *dir, struct dentry *dentry, int mode, int rdev)
 	int err;
 	if ((err = hpfs_chk_name((char *)name, &len))) return err==-ENOENT ? -EINVAL : err;
 	if (dir->i_sb->s_hpfs_eas < 2) return -EPERM;
+	lock_kernel();
 	if (!(fnode = hpfs_alloc_fnode(dir->i_sb, hpfs_i(dir)->i_dno, &fno, &bh))) goto bail;
 	memset(&dee, 0, sizeof dee);
 	if (!(mode & 0222)) dee.read_only = 1;
@@ -196,6 +197,7 @@ int hpfs_mknod(struct inode *dir, struct dentry *dentry, int mode, int rdev)
 		brelse(bh);
 		hpfs_free_sectors(dir->i_sb, fno, 1);
 		hpfs_unlock_inode(dir);
+		unlock_kernel();
 		return -EEXIST;
 	}
 	fnode->len = len;
@@ -221,12 +223,14 @@ int hpfs_mknod(struct inode *dir, struct dentry *dentry, int mode, int rdev)
 	hpfs_unlock_iget(dir->i_sb);
 	hpfs_unlock_inode(dir);
 	brelse(bh);
+	unlock_kernel();
 	return 0;
-	bail1:
+bail1:
 	brelse(bh);
 	hpfs_free_sectors(dir->i_sb, fno, 1);
 	hpfs_unlock_inode(dir);
-	bail:
+bail:
+	unlock_kernel();
 	return -ENOSPC;
 }
 

fs/jffs/inode-v23.c

@@ -1052,6 +1052,7 @@ jffs_mknod(struct inode *dir, struct dentry *dentry, int mode, int rdev)
 
 	D1(printk("***jffs_mknod()\n"));
 
+	lock_kernel();
 	dir_f = (struct jffs_file *)dir->u.generic_ip;
 	c = dir_f->c;
 
@@ -1123,6 +1124,7 @@ jffs_mknod(struct inode *dir, struct dentry *dentry, int mode, int rdev)
 jffs_mknod_end:
 	D3(printk (KERN_NOTICE "mknod(): up biglock\n"));
 	up(&c->fmc->biglock);
+	unlock_kernel();
 	return result;
 } /* jffs_mknod()  */
 

fs/jffs2/dir.c

@@ -802,9 +802,12 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 	__u32 writtenlen;
 	int ret;
 
+	lock_kernel();
 	ri = jffs2_alloc_raw_inode();
-	if (!ri)
+	if (!ri) {
+		unlock_kernel();
 		return -ENOMEM;
+	}
 	
 	c = JFFS2_SB_INFO(dir_i->i_sb);
 	
@@ -822,6 +825,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 
 	if (ret) {
 		jffs2_free_raw_inode(ri);
+		unlock_kernel();
 		return ret;
 	}
 
@@ -830,6 +834,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 	if (IS_ERR(inode)) {
 		jffs2_free_raw_inode(ri);
 		jffs2_complete_reservation(c);
+		unlock_kernel();
 		return PTR_ERR(inode);
 	}
 	inode->i_op = &jffs2_file_inode_operations;
@@ -854,6 +859,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 		up(&f->sem);
 		jffs2_complete_reservation(c);
 		jffs2_clear_inode(inode);
+		unlock_kernel();
 		return PTR_ERR(fn);
 	}
 	/* No data here. Only a metadata node, which will be 
@@ -874,6 +880,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 		if (ret) {
 			/* Eep. */
 			jffs2_clear_inode(inode);
+			unlock_kernel();
 			return ret;
 		}
 	}
@@ -883,6 +890,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 		/* Argh. Now we treat it like a normal delete */
 		jffs2_complete_reservation(c);
 		jffs2_clear_inode(inode);
+		unlock_kernel();
 		return -ENOMEM;
 	}
 
@@ -916,6 +924,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 		   as if it were the final unlink() */
 		up(&dir_f->sem);
 		jffs2_clear_inode(inode);
+		unlock_kernel();
 		return PTR_ERR(fd);
 	}
 
@@ -923,6 +932,7 @@ static int jffs2_mknod (struct inode *dir_i, struct dentry *dentry, int mode, in
 	   one if necessary. */
 	jffs2_add_fd_to_list(c, fd, &dir_f->dents);
 	up(&dir_f->sem);
+	unlock_kernel();
 
 	d_instantiate(dentry, inode);
 	return 0;

fs/namei.c

@@ -1262,9 +1262,7 @@ int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
 		return -EPERM;
 
 	DQUOT_INIT(dir);
-	lock_kernel();
 	error = dir->i_op->mknod(dir, dentry, mode, dev);
-	unlock_kernel();
 	if (!error)
 		inode_dir_notify(dir, DN_CREATE);
 	return error;

fs/nfs/dir.c

@@ -670,6 +670,7 @@ static int nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, int rde
 	attr.ia_mode = mode;
 	attr.ia_valid = ATTR_MODE;
 
+	lock_kernel();
 	nfs_zap_caches(dir);
 	error = NFS_PROTO(dir)->mknod(dir, &dentry->d_name, &attr, rdev,
 					&fhandle, &fattr);
@@ -677,6 +678,7 @@ static int nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, int rde
 		error = nfs_instantiate(dentry, &fhandle, &fattr);
 	if (error || fhandle.size == 0)
 		d_drop(dentry);
+	unlock_kernel();
 	return error;
 }
 

fs/reiserfs/namei.c

@@ -588,6 +588,7 @@ static int reiserfs_mknod (struct inode * dir, struct dentry *dentry, int mode,
     if (!inode) {
 	return -ENOMEM ;
     }
+    lock_kernel();
     journal_begin(&th, dir->i_sb, jbegin_count) ;
     windex = push_journal_writer("reiserfs_mknod") ;
 
@@ -595,6 +596,7 @@ static int reiserfs_mknod (struct inode * dir, struct dentry *dentry, int mode,
     if (!inode) {
 	pop_journal_writer(windex) ;
 	journal_end(&th, dir->i_sb, jbegin_count) ;
+	unlock_kernel();
 	return retval;
     }
 
@@ -614,12 +616,14 @@ static int reiserfs_mknod (struct inode * dir, struct dentry *dentry, int mode,
 	pop_journal_writer(windex) ;
 	journal_end(&th, dir->i_sb, jbegin_count) ;
 	iput (inode);
+	unlock_kernel();
 	return retval;
     }
 
     d_instantiate(dentry, inode);
     pop_journal_writer(windex) ;
     journal_end(&th, dir->i_sb, jbegin_count) ;
+    unlock_kernel();
     return 0;
 }
 

fs/udf/namei.c

@@ -679,6 +679,7 @@ static int udf_mknod(struct inode * dir, struct dentry * dentry, int mode, int r
 	int err;
 	struct FileIdentDesc cfi, *fi;
 
+	lock_kernel();
 	err = -EIO;
 	inode = udf_new_inode(dir, mode, &err);
 	if (!inode)
@@ -691,6 +692,7 @@ static int udf_mknod(struct inode * dir, struct dentry * dentry, int mode, int r
 		inode->i_nlink --;
 		mark_inode_dirty(inode);
 		iput(inode);
+		unlock_kernel();
 		return err;
 	}
 	cfi.icb.extLength = cpu_to_le32(inode->i_sb->s_blocksize);
@@ -710,6 +712,7 @@ static int udf_mknod(struct inode * dir, struct dentry * dentry, int mode, int r
 	d_instantiate(dentry, inode);
 	err = 0;
 out:
+	unlock_kernel();
 	return err;
 }
 

fs/ufs/namei.c

@@ -114,7 +114,9 @@ static int ufs_mknod (struct inode * dir, struct dentry *dentry, int mode, int r
 	if (!IS_ERR(inode)) {
 		init_special_inode(inode, mode, rdev);
 		mark_inode_dirty(inode);
+		lock_kernel();
 		err = ufs_add_nondir(dentry, inode);
+		unlock_kernel();
 	}
 	return err;
 }