[talk] VPNs: Choosing between OpenVPN and L2TP/IPsec

[Isaac (.ike) Levy]

Thanks for this thoughtful response Christos,

On 04/19/15 20:42, Christos Zoulas wrote:
> On Apr 19,  7:39pm, ike at blackskyresearch.net ("Isaac (.ike) Levy") wrote:
> -- Subject: Re: [talk] VPNs: Choosing between OpenVPN and L2TP/IPsec
> 
> | Christos, I don't mean to put you on the spot, but I figure you're a
> | great person to thoughtfully comment on the relative security of IPSec
> | itself these days?
> | 
> | Problems like these worry me,
> | http://www.mail-archive.com/cryptography@metzdowd.com/msg12325.html
> 
> I've read this article and I agree with all its points. While I agree,
> I don't see many alternatives out there -- you mentioned most of them
> (and nobody has published an exploit that I know of).

And we all sigh a big, loud, collective sigh.

I think we can pretty much all agree, VPN tech an area in computing
which desperately needs some attention.  Scrutiny and thoughtful cleanup
to the applied bits, at the very least.

> 
> | To me, it IPSec seems ripe for a very serious design flaw to come to
> | light in coming years- and at the least, all the fuss surrounding it-
> | and it's relative complexity- bothers me more.
> 
> What makes things worse is that the "roadwarrior" configuration
> everyone uses utilizes a common shared secret (which makes things
> easier to abuse and to break since now you only have to find the
> username and password) and also needs a wildcard match since we
> don't know a-priori the address of the client endpoint. This is
> explained here

Yep, it bothers me.  Even with a multi-factor auth hw token or something
to offset the common shared secret, the bar is still low.
(This reminds me of the earliest weaknesses in PPTP years ago- all about
weak auth...)

> (ENABLE_WILDCARD_MATCH explanation about shared
> secret problems):
> 
> I would not feel comfortable deploying that configuration to customer
> connections (unless I only had one customer), but this is what I use
> for home (where I am the only customer). Others don't seem to mind,
> and sell such tunnels for a few bucks a month... It is the ease of
> stealing ones credentials from those tunnels that compelled me
> to deploy my own.

understood.

> 
> I think that the OpenVPN solution is easier to deploy. 

Yes indeed, by an order of magnitude in complexity.

I just sat and did a side-by-side comparison of OpenVPN and L2TP/IPSsec
today:

L2TP/IPSEC was seriously a great deal more complex, including un-fun
gotchas like having to pull a gateway(ish) IP from outside the
distributable range.

Bits like the shared secret were not very transparent to troubleshoot,
(took me some time to figure out  I fat fingered a character).

I fat-fingered some OpenVPN bits too, and even the log messages made it
quick and simple to replace.
The OpenVPN bits were so much simpler in fact, that I went whole hog and
had enough time to setup the comparison using signed client
certificates- pretty hot IMHO.  Even the netblocks and routing was
cleaner to sew into a dev network.

> I chose to
> use IPSEC+L2TP mostly because I wanted to have a kernel supported,
> standards compliant tunnel solution on NetBSD working.

Well, so yeah- there's the rub.  OpenVPN still bothers me on several
levels- from ssl/tls, to the mere fact that the nice GUI clients for
users are 3rd party things.

Hrmph.

We'll be deciding our path tomorrow, I'll follow up here if any more fun
notes come out of it.

Best,
.ike



> 
> christos
>