[Openswan Users] Firewall rules for openswan behind NAT

Wed Nov 20 19:11:31 UTC 2013

You've addressed some or most of what's in the following bird's-eye view; it 
shouldn't hurt to review in case you've overlooked something.

You need to forward UDP ports 500 and 4500 on each firewall to the local 
openswan box (limit it to coming from the remote IP/openswan box for added 
security), and allow UDP ports 500 and 4500 out from the local openswan box 
(preferably limit it to going to the remote openswan box for added security). 
This will allow either end to start the VPN. Each firewall should drop packets 
destined for the remote LAN(s) since they don't know how to reach them. 
Logging them with a specific prefix would make it easy to discover 
malconfigured nodes.

I don't believe protocols 50 and 51 are used when NAT-T is used.

Each internal node must have an explicit route to the remote LAN via the local 
openswan box's IP.

Each openswan box needs route(s) to the local internal LAN(s) and a default 
route via the local firewall's internal IP address. Openswan takes care of 
adding/removing routes to remote LANs when it brings VPNs up and down.

Do you have a 'config setup' in each ipsec.conf, possibly similar to:
        protostack=klips     # or netkey or mast
        interfaces=%defaultroute
        klipsdebug=none
        plutodebug=none
        plutowait=no
        uniqueids=yes
        nat_traversal=yes
You might need a virtual_private declaration describing your internal LANs, 
but probably not, since you aren't using a road warrior setup.

In each conn spec, you may need to specify ike= and esp=; that is, tell 
openswan which encrytpion to use. I don't know what happens if you don't 
specify any encryption methods. (Does openswan then transmit in the clear?)

I haven't played with NAT Traversal lately; I forget how to configure a conn 
to make NAT-T work.

N

On Wednesday, November 20, 2013 12:28:35 PM Fred Weston wrote:
> I cannot ping LAN-LAN without the tunnel.
> 
> In site A I have 10.0.0.0/16 and in site B I have 10.1.0.0/16.  In each
> site the routing table has an entry for the opposite site’s IP space which
> is pointed at the local openswan box.
> 
> I think we are confusing firewall and tunnel terminology.  The firewall I
> am speaking of is builtin to AWS and controls ingress and egress traffic
> from the openswan boxes.  It is something that is part of the AWS network
> stack and the only thing I can do to it is change its ruleset.
> 
> The traffic between sites is traversing the openswan tunnel, however when
> it’s doing so it isn’t being encrypted, so for instance when I send RDP
> traffic across the tunnel, the AWS firewall sees RDP traffic and doesn’t
> let it through.  If the tunnel were encrypting traffic, all the AWS
> firewall should see is UDP traffic coming into port 500 and it should have
> no idea what that traffic is, but it’ll allow it because I’ve told it to
> permit udp/500 inbound.
> 
> In the diagram below I’ve notated where the AWS firewall at each site is
> inspecting traffic.  The issue is that the traffic coming across the
> tunnel is in the clear.
> 
> [cid:image002.jpg at 01CEE5EC.06F0A5C0]
> 
> From: Nick Howitt [mailto:n1ck.h0w1tt at gmail.com]
> Sent: Wednesday, November 20, 2013 3:36 AM
> To: Fred Weston
> Cc: users at lists.openswan.org
> Subject: Re: [Openswan Users] Firewall rules for openswan behind NAT
> 
> 
> I am curious that your are even pinging LAN-LAN outside the tunnel. You
> have private subnets and the internet would not know where to find a
> 10.x.y.z address. Can you do a tracert/traceroute end to end with and
> without the firewall?
> 
> On your gateways to you have routes set up to your far LAN's via your
> Openswan devices?
> 
> Can you ping from 10.0.0.82 to 10.1.0.67 or vice-versa with the firewall up
> and down?
> 
> On 2013-11-19 17:14, Fred Weston wrote:
> The tunnel comes up either way but I can't ping unless I permit icmp from
> the Internet into openswan.  I'm pinging from another device behind
> openswan.
> 
> Thanks ,
> FW
> 
> On Nov 19, 2013, at 11:40 AM, "Nick Howitt"
> <n1ck.h0w1tt at gmail.com<mailto:n1ck.h0w1tt at gmail.com>> wrote:
> 
> OK that shows the tunnel is up. Is that with or without the firewall (and
> btw it is using NAT-T so you should not need protocols 50 and 51 through
> your firewall).
> 
> When you are pinging end to end, is that from the openswan device or from
> another LAN device?
> 
> On 2013-11-19 14:03, Fred Weston wrote:
> This is what I see in the log; it looks like it’s encrypting traffic but
> that doesn’t seem to be the case based upon the behavior I’m seeing.  If
> it is encrypting then the firewall in front of openswan should have no
> effect on the traffic I can pass over the tunnel as long as the tunnel is
> up.
> 
> Nov 19 14:00:55 ip-10-0-0-82 pluto[12517]: "vpc1-to-vpc2" #1:
> NAT-Traversal: Result using RFC 3947 (NAT-Traversal): both are NATed Nov
> 19 14:00:55 ip-10-0-0-82 pluto[12517]: "vpc1-to-vpc2" #1: transition from
> state STATE_MAIN_I2 to state STATE_MAIN_I3 Nov 19 14:00:55 ip-10-0-0-82
> pluto[12517]: "vpc1-to-vpc2" #1: STATE_MAIN_I3: sent MI3, expecting MR3
> Nov 19 14:00:55 ip-10-0-0-82 pluto[12517]: "vpc1-to-vpc2" #1: received
> Vendor ID payload [CAN-IKEv2] Nov 19 14:00:55 ip-10-0-0-82 pluto[12517]:
> "vpc1-to-vpc2" #1: Main mode peer ID is ID_IPV4_ADDR: '50.18.211.121' Nov
> 19 14:00:55 ip-10-0-0-82 pluto[12517]: "vpc1-to-vpc2" #1: transition from
> state STATE_MAIN_I3 to state STATE_MAIN_I4 Nov 19 14:00:55 ip-10-0-0-82
> pluto[12517]: "vpc1-to-vpc2" #1: STATE_MAIN_I4: ISAKMP SA established
> {auth=OAKLEY_PRESHARED_KEY cipher=aes_128 prf=oakley_sha group=modp2048}
> Nov 19 14:00:55 ip-10-0-0-82 pluto[12517]: "vpc1-to-vpc2" #2: initiating
> Quick Mode PSK+ENCRYPT+TUNNEL+PFS+UP+IKEv2ALLOW+SAREFTRACK {using isakmp#1
> msgid:463435ec proposal=defaults pfsgroup=OAKLEY_GROUP_MODP2048} Nov 19
> 14:00:56 ip-10-0-0-82 pluto[12517]: "vpc1-to-vpc2" #2: transition from
> state STATE_QUICK_I1 to state STATE_QUICK_I2 Nov 19 14:00:56 ip-10-0-0-82
> pluto[12517]: "vpc1-to-vpc2" #2: STATE_QUICK_I2: sent QI2, IPsec SA
> established tunnel mode {ESP=>0x54dd12fe <0x2bb3e074
> xfrm=AES_128-HMAC_SHA1 NATOA=none NATD=50.18.211.121:4500 DPD=none}
> 
> From: Nick Howitt [mailto:n1ck.h0w1tt at gmail.com]
> Sent: Tuesday, November 19, 2013 8:21 AM
> To: Fred Weston
> Cc: users at lists.openswan.org<mailto:users at lists.openswan.org>
> Subject: RE: [Openswan Users] Firewall rules for openswan behind NAT
> 
> 
> What are you getting in /var/log/secure - just the bit where the tunnel is
> negotiating, not the bit where ipsec loads?
> 
> Also what do you have in the "config setup" sections of your conf files?
> 
> On 2013-11-19 13:14, Fred Weston wrote:
> So here’s something interesting…this morning just for the heck of it, I
> added ICMP to the permit list and that immediately got ping working.
> 
> Since the tunnel shouldn’t require ICMP, that got me thinking that the
> traffic isn’t actually being encrypted.  I verified that by trying to
> remote desktop to a host on the far side of the tunnel.  It didn’t work
> when I have the firewall rules set to only allow the few ports/protocols
> the tunnel should need, but as soon as I changed the ruleset to permit all
> traffic RDP worked, so it seems the problem is actually that the tunnel
> isn’t encrypting the traffic.
> 
> I’m not quite sure why this is.
> 
> Here are the configs from each side, can someone comment as to what I need
> to add to get the traffic to be encrypted?
> 
> conn vpc1-to-vpc2
>         type=tunnel
>         authby=secret
>         left=%defaultroute
>         leftid=107.21.17.86
>         leftnexthop=%defaultroute
>         leftsubnet=10.0.0.0/16
>         leftsourceip=10.0.0.82
>         right=50.18.211.121
>         rightsubnet=10.1.0.0/16
>         pfs=yes
>         auto=start
>         phase2=esp
> 
> conn vpc2-to-vpc1
>         type=tunnel
>         authby=secret
>         left=%defaultroute
>         leftid=50.18.211.121
>         leftnexthop=%defaultroute
>         leftsubnet=10.1.0.0/16
>         leftsourceip=10.1.0.67
>         right=107.21.17.86
>         rightsubnet=10.0.0.0/16
>         pfs=yes
>         auto=start
>         phase2=esp
> 
> 
> 
> From: Nick Howitt [mailto:n1ck.h0w1tt at gmail.com]
> Sent: Tuesday, November 19, 2013 5:19 AM
> To: Fred Weston
> Cc: users at lists.openswan.org<mailto:users at lists.openswan.org>
> Subject: RE: [Openswan Users] Firewall rules for openswan behind NAT
> 
> 
> Yes, I was picturing firewalling in the hosts.
> 
> Have a look in the logs and see if Openswan is connecting with or without
> NAT-T when your firewall is not up. Then, with your firewalling, try
> forceencaps=yes in the conn and nat_traversal=yes in config setup.
> 
> For minor tunnel info I use "service ipsec status", but it depends on your
> distro and the information is almost useless. You can also have a look at
> "ip xfrm" and "ip route".
> 
> Nick
> 
> On 2013-11-19 10:03, Fred Weston wrote:
> I think you’re picturing the firewalling taking place on the openswan
> hosts, which isn’t the case.  There isn’t a firewall on either openswan
> box (other than any standard firewall that may be enabled by default). 
> The firewall rules I am manipulating are in the network / NAT device in
> front of openswan.  Since the tunnel works when I permit all traffic to
> openswan, that would seem to discount the possibility of any firewall
> issues on the hosts themselves.
> 
> I’ll take a look at the logs to see if they show anything interesting.  Is
> there a utility that will show tunnel status?
> 
> From: Nick Howitt [mailto:n1ck.h0w1tt at gmail.com]
> Sent: Tuesday, November 19, 2013 3:44 AM
> To: Fred Weston
> Cc: users at lists.openswan.org<mailto:users at lists.openswan.org>
> Subject: RE: [Openswan Users] Firewall rules for openswan behind NAT
> 
> 
> For your rules, I was hoping for something like the output to "iptables -L
> -n -v" and "iptables -t nat -L -n -v" rather than a description of the
> rules.
> 
> Forceencaps is unlikely but can be useful
> 
> Openswan/ipsec logs are typically found in /var/log/secure depending on
> your system. If you have dpd enabled you should see constant tunnel
> renegotiation if the tunnel has gone down. You'll see nothing odd if the
> tunnel is up but no traffic is passing. When the tunnel is up you should
> see in the logs something like "IPSec SA Established". Available status
> information is not particulrly helpful.
> 
> Can you also try adding a firewall rule something like:
> 
> iptables -t nat -I POSTROUTING -m policy --dir out --pol ipsec -j ACCEPT
> 
> Either that or somthing in the post routing chain which allows traffic
> between the local and remote subnets, but this rule is more flexible as
> you don't need to specify the subnets.
> 
> Nick
> 
> On 2013-11-18 23:28, Fred Weston wrote:
> 
> From:
> users-bounces at lists.openswan.org<mailto:users-bounces at lists.openswan.org>
> [mailto:users-bounces at lists.openswan.org] On Behalf Of Nick Howitt
> 
> Can you post the exact rules you are using?
> 
> I included those in my original message.
> *:* > UDP 500
> *:* > UDP 4500
> * > IP Protocol 50
> * > IP Protocol 51
> 
> Also have you tried forcing encapsulation with forceencaps=yes in your
> conns?
> 
> No, I haven’t tried that.
> 
> When you say "things stop working" does the tunnel come down, or does
> traffic just fail to pass?
> 
> I’m not sure how to tell the difference, my test methodology was to ping a
> host on the far side of the tunnel and when I change the firewall rules
> from wide open to those above the ping starts timing out.  How can I tell
> what state the tunnel is in?
> 
> Regards,
> 
> Nick
> 
> On 2013-11-17 17:13, Fred Weston wrote:
> Does anyone else have any suggestions?  I would like to implement this in
> production but I am hesitant to do so when the only way I can get it to
> work is permit all traffic from the Internet into the openswan boxes.
> 
> From:
> users-bounces at lists.openswan.org<mailto:users-bounces at lists.openswan.org>
> [mailto:users-bounces at lists.openswan.org] On Behalf Of Fred Weston Sent:
> Wednesday, November 13, 2013 11:49 AM
> To: Leto
> Cc: users at lists.openswan.org<mailto:users at lists.openswan.org>
> Subject: Re: [Openswan Users] Firewall rules for openswan behind NAT
> 
> Let me clarify – when I reference ports/protocols that I’m allowing
> inbound, I’m allowing it from the opposite host and not specifying a
> source port.
> 
> Thanks,
> FW
> 
> 
> 
> 
> 
> From: Leto [mailto:letoams at gmail.com]
> Sent: Wednesday, November 13, 2013 11:27 AM
> To: Fred Weston
> Cc: users at lists.openswan.org<mailto:users at lists.openswan.org>
> Subject: Re: [Openswan Users] Firewall rules for openswan behind NAT
> 
> 
> 
> sent from a tiny device
> 
> On 2013-11-13, at 10:44, Fred Weston
> <fred.weston at lpga.com<mailto:fred.weston at lpga.com>> wrote: Hello All,
> 
> I’m using OpenSwan with AWS to link two private VPC networks in different
> regions.
> 
> I’m having trouble getting my firewall ACLs right.  Everything works if I
> permit all traffic to the OpenSwan boxes, however when I try to get more
> restrictive and permit only the necessary ports things stop working.
> 
> One side has all traffic permitted inbound for the time being and I’m
> making ACL changes trying to restrict traffic to certain ports/protocols
> on the other side.
> 
> Both endpoints are behind 1:1 NAT.  Everything is permitted outbound on
> both sides.
> 
> From reading online, I understand that the following ports and protocols
> should be all I need:
> 
> UDP 500
> UDP 4500
> IP Protocol 50
> IP Protocol 51
> 
> I tried the above and had no luck.  As soon as I change from permitting all
> inbound to permitting only the above list the tunnel comes down.
> 
> You should really allow icmp.
> 
> Note that you need to accept from a random high port to dest udp 4500, not
> just 4500 <-> 4500. Same for 500
> 
> 
> 
> 
> I also tried permitting tcp/1721 and tcp/1723 and IP Protocol 47.
> 
> I am using AWS ‘security groups’ to control filtering and according to the
> docs (and my observations) security groups are stateful, so I am not sure
> why this isn’t working.
> 
> Can anyone offer any suggestions?
> 
> 
> Thanks,
> 
> Fred Weston