Subject: kern/20039: mbuf leak using rp-pppoe
To: None <email@example.com>
From: None <firstname.lastname@example.org>
Date: 01/25/2003 02:06:12
>Synopsis: mbuf leak using rp-pppoe
>Arrival-Date: Fri Jan 24 07:07:00 PST 2003
>Originator: Simon Burge
>Release: NetBSD 1.6_STABLE from December 1, 2002
System: NetBSD way 1.6_STABLE NetBSD 1.6_STABLE (WAY)
#1: Sun Dec 1 01:25:28 EST 2002
rp-pppoe 2.6 or 3.5
Using rp-pppoe (version doesn't seem to matter) on a firewall
type box, I'm seeing an mbuf leak that seems to correspond
to roughly one mbuf for each new outbound TCP connection
over the ppp interface. These mbufs aren't freed after
the TCP connection finishes.
Here's a short log:
way:~ 22> netstat -m | grep 'mbufs allocated to data'
360 mbufs allocated to data
way:~ 23> ssh cvs.netbsd.org uptime
2:34PM up 33 days, 3:53, 8 users, load averages: 1.59, 1.97, 1.83
way:~ 24> netstat -m | grep 'mbufs allocated to data'
361 mbufs allocated to data
way:~ 25> ssh cvs.netbsd.org uptime
2:34PM up 33 days, 3:53, 8 users, load averages: 1.53, 1.94, 1.82
way:~ 26> netstat -m | grep 'mbufs allocated to data'
362 mbufs allocated to data
way:~ 27> ssh cvs.netbsd.org uptime
2:34PM up 33 days, 3:54, 8 users, load averages: 1.40, 1.89, 1.81
way:~ 28> netstat -m | grep 'mbufs allocated to data'
363 mbufs allocated to data
This box used to run the 1.5 branch using the same rp-pppoe
package and had no mbuf leak problems.
A friend has a box running 1.6 STABLE with a similar firewall
configuration (talking to the same ADSL provider) and
doesn't have the problem, while another friend has a
workstation type box with rp-pppoe and does see the problem
(also with 1.6 STABLE). This doesn't make it any easier
to track down the problem :(
Use rp-pppoe, start outgoing connections (but not in all