Pascal DeMilly
2003-Jan-24 22:16 UTC
[Shorewall-users] icmp: w.x.y.z unreachable need to defrag (mtu 296)
Hi, I have a setup that consist of 2 firewalls connected over dialup and PPP. Each side of the ppp are protected by shorewall. One side of the PPP masquerades everything not addressed to the local network to its eth0 (the net). fw1 <---- ppp (dialup) -----> fw0 <----- NET When making an http request to a site on the Internet from the machine not directly connected to the net (fw1), the request hangs. I have checked that I can ping addresses on the net and that my DNS resolution works correctly. Running tcpdump on the firewall connected to the net (fw0), I see that the request goes and comes back but the firewall sends an icmp unreachable need to defrag packet back to the sender, in that case the web site that I was trying to look at. I verified by running a tcpdump outside of fw0 that this icmp message was indeed leaving the firewall and not being dropped there. Now my question. I have set CLAMPMSS to yes thinking that this should solve the problem, but it didn''t. I also setup a transparent squid cache on fw0 and it seems to be a lot better (at least faster) but I can still see some icmp unreachable need to defrag packets. Any idea what else I could try? Thanks in advance Pascal -- Pascal DeMilly <list.shorewall@newgenesys.com>
Tom Eastep
2003-Jan-25 06:48 UTC
[Shorewall-users] icmp: w.x.y.z unreachable need to defrag (mtu 296)
--On Friday, January 24, 2003 10:21 PM -0800 Pascal DeMilly <list.shorewall@newgenesys.com> wrote:> > Now my question. I have set CLAMPMSS to yes thinking that this should > solve the problem, but it didn''t. I also setup a transparent squid cache > on fw0 and it seems to be a lot better (at least faster) but I can still > see some icmp unreachable need to defrag packets. Any idea what else I > could try? >What values have you set for ''mtu'' and ''mru'' in /etc/ppp/options? When I used a fw-based PPTP link, I used 1000 for both. -Tom -- Tom Eastep \ Shorewall - iptables made easy AIM: teastep \ http://www.shorewall.net ICQ: #60745924 \ teastep@shorewall.net
Tom Eastep
2003-Jan-25 09:08 UTC
[Shorewall-users] icmp: w.x.y.z unreachable need to defrag (mtu 296)
--On Saturday, January 25, 2003 6:48 AM -0800 Tom Eastep <teastep@shorewall.net> wrote:> > > --On Friday, January 24, 2003 10:21 PM -0800 Pascal DeMilly > <list.shorewall@newgenesys.com> wrote: > >> >> Now my question. I have set CLAMPMSS to yes thinking that this should >> solve the problem, but it didn''t. I also setup a transparent squid cache >> on fw0 and it seems to be a lot better (at least faster) but I can still >> see some icmp unreachable need to defrag packets. Any idea what else I >> could try? >> > > What values have you set for ''mtu'' and ''mru'' in /etc/ppp/options? When I > used a fw-based PPTP link, I used 1000 for both. >I was reviewing your post again and I notice that the ppp link is dialup -- IIRC, those should use a mtu and mru in the 500-600 byte range. Possibly someone else who uses dial-up can comment... -Tom -- Tom Eastep \ Shorewall - iptables made easy AIM: teastep \ http://www.shorewall.net ICQ: #60745924 \ teastep@shorewall.net
Larry Platzek
2003-Jan-25 18:07 UTC
[Shorewall-users] icmp: w.x.y.z unreachable need to defrag (mtu 296)
Tom: I am using a dialup modem and have a mtu of 1500 this according to "ip addr" this is runnung on a Bering-Uclib-1.0.1 using the ppp-filter.lrp I have not tried specifying MTU or MRU. I am just starting to update my systems need to try some other ppp options and document them. Hope to be up being active soon on this and the LEAF lists. Larry Platzek larryp@inow.com On Sat, 25 Jan 2003, Tom Eastep wrote:> Date: Sat, 25 Jan 2003 09:08:17 -0800 > From: Tom Eastep <teastep@shorewall.net> > To: Shorewall users list <shorewall-users@shorewall.net> > Subject: Re: [Shorewall-users] icmp: w.x.y.z unreachable need to defrag > (mtu 296) > > > > --On Saturday, January 25, 2003 6:48 AM -0800 Tom Eastep > <teastep@shorewall.net> wrote: > > > > > > > --On Friday, January 24, 2003 10:21 PM -0800 Pascal DeMilly > > <list.shorewall@newgenesys.com> wrote: > > > >> > >> Now my question. I have set CLAMPMSS to yes thinking that this should > >> solve the problem, but it didn''t. I also setup a transparent squid cache > >> on fw0 and it seems to be a lot better (at least faster) but I can still > >> see some icmp unreachable need to defrag packets. Any idea what else I > >> could try? > >> > > > > What values have you set for ''mtu'' and ''mru'' in /etc/ppp/options? When I > > used a fw-based PPTP link, I used 1000 for both. > > > > I was reviewing your post again and I notice that the ppp link is dialup -- > IIRC, those should use a mtu and mru in the 500-600 byte range. Possibly > someone else who uses dial-up can comment... > > -Tom > -- > Tom Eastep \ Shorewall - iptables made easy > AIM: teastep \ http://www.shorewall.net > ICQ: #60745924 \ teastep@shorewall.net > _______________________________________________ > Shorewall-users mailing list > Shorewall-users@lists.shorewall.net > http://lists.shorewall.net/mailman/listinfo/shorewall-users >
Tom Eastep
2003-Jan-25 18:09 UTC
[Shorewall-users] icmp: w.x.y.z unreachable need to defrag (mtu 296)
--On Saturday, January 25, 2003 6:06 PM -0800 Larry Platzek <larryp@inow.com> wrote:> Tom: > I am using a dialup modem and have a mtu of 1500 this according to > "ip addr" this is runnung on a Bering-Uclib-1.0.1 using the ppp-filter.lrp > I have not tried specifying MTU or MRU. > > I am just starting to update my systems need to try some other ppp options > and document them. > Hope to be up being active soon on this and the LEAF lists. >Thanks, Larry. -Tom -- Tom Eastep \ Shorewall - iptables made easy Shoreline, \ http://www.shorewall.net Washington USA \ teastep@shorewall.net
Pascal DeMilly
2003-Jan-26 00:27 UTC
[Shorewall-users] icmp: w.x.y.z unreachable need to defrag (mtu 296)
I did some more debugging and one thing that threw me off-track at first (I thought shorewall was blocking this particular icmp message) was actually very telling. I didn''t include a trace in my previous message because I thought it was irrelevant and primarily because it is on a testing subnet non easily accessible from my mail reader. But from that trace, I could see that the icmp message returned to the web server has in its payload, the IP address of the originating machine behind my ppp connection instead of as one should expect the NAT gateway, and that didn''t seem right. I googled on that find and found that interesting e-mail on the netfilter mailing-list. http://lists.netfilter.org/pipermail/netfilter-devel/2002-October/009511.html which explain the problem. Now as far as the MTU I was using 296 (256+40) which is according to some RFC I read the recommended value for dialup modem. I since upgraded it to 552 (512+40) because I am using V.92 modems and get quite frequently over 36K throughput. I have been writing this e-mail over the last day, trying to make sure to test everything before posting and I have found the problem. In my PPP option file I had set the MRU but not the MTU. So even if ifconfig shows the expected MTU always verify. Setting the MRU and MTU to the same value got ride off the ICMP messages but nevertheless it seems that there is an implementation problem with it in netfilter. Thanks to everybody and particularly as always to Tom. Pascal On Fri, 2003-01-24 at 22:21, Pascal DeMilly wrote:> Hi, > > I have a setup that consist of 2 firewalls connected over dialup and > PPP. Each side of the ppp are protected by shorewall. One side of the > PPP masquerades everything not addressed to the local network to its > eth0 (the net). > > fw1 <---- ppp (dialup) -----> fw0 <----- NET > > When making an http request to a site on the Internet from the machine > not directly connected to the net (fw1), the request hangs. I have > checked that I can ping addresses on the net and that my DNS resolution > works correctly. > > Running tcpdump on the firewall connected to the net (fw0), I see that > the request goes and comes back but the firewall sends an icmp > unreachable need to defrag packet back to the sender, in that case the > web site that I was trying to look at. I verified by running a tcpdump > outside of fw0 that this icmp message was indeed leaving the firewall > and not being dropped there. > > Now my question. I have set CLAMPMSS to yes thinking that this should > solve the problem, but it didn''t. I also setup a transparent squid cache > on fw0 and it seems to be a lot better (at least faster) but I can still > see some icmp unreachable need to defrag packets. Any idea what else I > could try? > > Thanks in advance > > Pascal-- Pascal DeMilly <list.shorewall@newgenesys.com>
Tom Eastep
2003-Jan-26 06:48 UTC
[Shorewall-users] icmp: w.x.y.z unreachable need to defrag (mtu 296)
--On Sunday, January 26, 2003 12:33 AM -0800 Pascal DeMilly <list.shorewall@newgenesys.com> wrote:> In my > PPP option file I had set the MRU but not the MTU. So even if ifconfig > shows the expected MTU always verify. Setting the MRU and MTU to the > same value got ride off the ICMP messages but nevertheless it seems that > there is an implementation problem with it in netfilter.Or ifconfig is showing you the wrong information... :-) -Tom -- Tom Eastep \ Shorewall - iptables made easy Shoreline, \ http://www.shorewall.net Washington USA \ teastep@shorewall.net