Subject: Memory leak in nfsrv_readdir
To: None <tech-kern@netbsd.org, tech-net@netbsd.org>
From: Mark Davies <mark@mcs.vuw.ac.nz>
List: tech-kern
Date: 07/20/2006 16:16:38
A couple of weeks ago I posted to current-users about one of my servers that 
kept locking up every few days. 
(http://mail-index.netbsd.org/current-users/2006/07/06/0007.html)
After some very useful assistance by Juergen Hannken-Illjes we've tracked down 
that the actual problem is a memory leak in nfsrv_readdir().  It looks like 
this leak has been there since Lite2 was merged in 8 years ago.

Whats happening is that nfsrv_readdir() calls VOP_READDIR which malloc's some 
space to hold the directory seek cookies.  Normally the cookies are freed at 
the end of the routine but somtimes the code jumps back to the "again" label 
calling VOP_READDIR again so mallocing new cookies without ever freeing the 
old ones.

The below patch fixes the problem in the same way as FreeBSD does.  OK to 
commit?

I still don't know what it is about that particular server of mine that is 
tickling the bug much more frequently than other boxes.

cheers
mark

Index: nfs_serv.c
===================================================================
RCS file: /src/cvs/netbsd/src/sys/nfs/nfs_serv.c,v
retrieving revision 1.98
diff -u -r1.98 nfs_serv.c
--- nfs_serv.c	6 Oct 2005 10:23:01 -0000	1.98
+++ nfs_serv.c	20 Jul 2006 03:35:09 -0000
@@ -2668,6 +2668,11 @@
 	eofflag = 0;
 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 
+	if (cookies) {
+		free((caddr_t)cookies, M_TEMP);
+		cookies = NULL;
+	}
+
 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
 
 	off = (off_t)io.uio_offset;
@@ -2928,6 +2933,11 @@
 
 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 
+	if (cookies) {
+		free((caddr_t)cookies, M_TEMP);
+		cookies = NULL;
+	}
+
 	error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
 
 	off = (u_quad_t)io.uio_offset;