[OpenVPN home] [Date Prev] [Date Index] [Date Next]
[OpenVPN mailing lists] [Thread Prev] [Thread Index] [Thread Next]
Google
 
Web openvpn.net

Re: [Openvpn-users] TAP driver stalls on windows [too]?


  • Subject: Re: [Openvpn-users] TAP driver stalls on windows [too]?
  • From: "James Yonan" <jim@xxxxxxxxx>
  • Date: Wed, 1 Oct 2003 11:45:52 -0000

Luc Van der Veken <lucvdv@xxxxxxxx> said:

> Sorry for the length of this, but I wanted to be as complete as
> possible.
> 
> In short: the problem is [apparently] a TAP device that stops working
> (on one machine only, the others seem to be fine).
> 
> I'm still running 1.5 beta 6, just downloaded beta 8 but not installed
> yet.
> 
> For testing, I try to keep two OpenVPN tunnels open 24/24 and 7/7
> between Win2000 machines.  A third will be added in a week or so, and
> a fourth is only used every now and then (between my home machine and
> the office, XP to Win2000).
> 
> There's only little traffic through the tunnels: an e-mail, auto-sent
> every 4 hours to verify that the tunnel is still there; and through
> one (the one that fails) I also log on to Terminal Server and access a
> SQL server from time to time (normally less than once per day), but
> that doesn't seem to be related to the problem: it always occurred
> when the tunnel was idle (just sending its keepalives).
> 
> 
> One of the tunnels is established over a LAN, without any routers
> between the two endpoints.  It uses the default MTU.
> This one is still up after two weeks (I test it by sending e-mails to
> the tunnel's TAP IP address instead of the machine's normal IP
> address).  It's between Win2000 Pro (client side) and Win2000 Server.
> 
> 
> The other connects a DSL client (fixed IP, server side of the tunnel)
> with a cable internet client (dynamic IP, but it hasn't changed yet
> for as long as the test runs).  This connects two Win2000 Servers.
> 
> There's a firewall at both ends: on the server side that's a
> Watchguard Firebox II set up to NAT the OpenVPN port to the machine
> where the server side is running; on the client side it's a cheap
> consumer-type D-Link firewall set up to block all but the tunnel's
> port.
> This tunnel uses a lower MTU because that was the only way to get RDP
> (terminal server) connections to work through it.  The funny thing is
> that this is not necessary on a connection from my home machine,
> through the same cable internet provider (differences: different type
> of cable modem, different brand of firewall, and IP address in another
> range).
> 
> 
> This tunnel just won't stay up, and it looks like it's the TAP driver
> that keeps going down.  OpenVPN is set to auto-restart when the tunnel
> goes away: it restarts, but it fails to re-establish the tunnel.
> Restarting the OpenVPN service doesn't help.
> 
> After disabling and re-enabling the TAP driver in device management
> the tunnel is re-established immediately (it says so in the logs),
> _but_ no traffic is possible through it: it requires a reboot before
> that starts working again.

You might try beta8 (or beta9 which will be released shortly with new OpenSSL
0.9.7c DLLS in response to the security advisory).

A great deal of cleanup & stability work has gone into the driver between
beta7 and beta8.

There's also better diagnostics, e.g. if you run beta8 or higher from a
console window, F2 will print a driver status line which contains information
that would be useful in troubleshooting possible driver bugs.

> The longest this tunnel has stayed up was 8 days, but the second
> longest was only 24 hours.  The last time it was restarted was
> yesterday at 10:10, the logs show an attempted auto-restart (that
> failed to re-establish the connection) at 16:54.
> 
> 
> Q1: Is there a known issue with the TAP driver that's included with
> 1.5 beta 6 that could cause this?

It's quite possible.  I would try to test on beta9 (will be released shortly).

I will certainly work with you to track down bugs.

> Q2: has a timeframe been set for a non-beta release?
> I'm not trying to press, but I'll be needing stable, reliable tunnels
> before the end of the year, sooner if possible.
> If that falls outside of the expected timeframe, I'll have to look for
> another solution.

My threshold for releasing 1.5 is that a beta or release candidate has been
out for 2 or 3 weeks and shows all of the qualities of stability of OpenVPN
1.4.3 running on non-Windows platforms.

While it's hard to say exactly when that will happen, I would expect that it's
probably more than one month away but less than four.

One thing that I could use right now to accelerate the process is direct
access to a Win2K machine.  I rarely hear of problems running the TAP-Win32
driver on XP -- because that's what my development laptop is running, and I
test on XP fairly extensively before I release stuff.  The problem is that
Win2K is different enough from XP that a lot of things that work fine on XP
(such as ethernet bridging) simply don't work on Win2K.  So when there are
problems, I tend to only hear about them by email, and I can tell you that
debugging by email is at least 10 times slower than debugging when you have
direct access to a machine.

> Some config details:
> 
> - TAP subnet 10.254.254.4/30 (server 10.254.254.5, client
> 10.254.254.6)
> [the other tunnels currently in use are 10.254.254.0/30 and
> 10.254.254.8/30]
> 
> - ports used: high (above 50000), one per tunnel.
> 
> - a different keyfile per tunnel (adequately protected against prying
> eyes, even on the local machine).
> 
> - MTU on the TAP adapter = 1052 (1024 + packet header)
> 
> - server and client sides are _both_ set to auto-restart and to send
> keepalives.
> Originally the server side didn't auto-restart, I added this after it
> happened a number of times to get a log entry when the tunnel is
> broken.  I set it to double the restart delay of the client side, so
> they don't both restart at the same moment.
> 
> - both Win2000 servers run RRAS with static routes set to the other
> side's LAN through the tunnel.
> One network is 192.168.0.0/24, the other 10.13.9.0/24.
> On the tunnel server side (192.168), packet filters are installed that
> allow only the right incoming packets to keep the clients off of a
> second local network they have no business on, and to keep them out of
> each other's networks.  Allowed packets are: the LAN they have access
> to, the tunnel's own subnet, and the server's address (/32) on the
> network they have no access to (in case something on it ever tells a
> client that's its IP address).
> 
> 
> Config file, server side (comments stripped):
> The resolv-retry entries do nothing, because of fixed IP addressing.
> 
> port 5xxxx
> tun-mtu 1052
> dev tap
> dev-node OpenVPN-5xxxx
> secret key-5xxxx.txt
> ping-restart 120
> ping-timer-rem
> persist-tun
> persist-key
> resolv-retry 86400
> ping 10
> comp-lzo
> verb 3
> mute 10
> up updownlog.cmd
> down updownlog.cmd
> up-restart
> 
> The config file on the client size is identical, except for:
> 
> remote x.y.z.a
> ping-restart 60
> 
> 
> 
> 
> -------------------------------------------------------
> This sf.net email is sponsored by:ThinkGeek
> Welcome to geek heaven.
> http://thinkgeek.com/sf
> _______________________________________________
> Openvpn-users mailing list
> Openvpn-users@xxxxxxxxxxxxxxxxxxxxx
> https://lists.sourceforge.net/lists/listinfo/openvpn-users
> 



-- 




____________________________________________
Openvpn-users mailing list
Openvpn-users@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/openvpn-users