Re: [BUG REPORT] Unencrypted packets after SNAT, allthough IPSEC-Policies are present

[Steffen Klassert <steffen.klassert@secunet.com>]

On Wed, Sep 10, 2014 at 07:26:53PM +0200, Konstantinos Kolelis wrote:
> Hi all,
> 
> i' ve observed a problem with xfrm lookups, SNAT, blackhole route and
> missing SAs.
> The problem occures with all Kernels above 3.6.x and might has to do
> with the changes in
> ip4_blackhole_route() function in net/route.c.

Thanks for the report!

Is kernel v3.6 the first kernel with this issue? It seems that
we have this problem already longer, at least if my analysis
is correct.

> 
> Let say you have two network interfaces:
> eth0 with ip 172.16.0.10/24
> and
> eth1 with ip 192.168.0.1/24
> 
> and you have done the following configuration:
> 
> iptables -t nat -A POSTROUTING -s 192.168.0.0/24 -j SNAT --to-source
> 172.16.0.10
> 
> and
> 
> ip xfrm policy add dir out  src 172.16.0.10 dst 0.0.0.0/0 tmpl proto esp
> src 172.16.0.10 dst 172.31.0.10 mode tunnel
> 
> with the following routes:
> default via 172.16.0.1 dev eth0  proto static
> 172.16.0.0/24 dev eth0  proto kernel  scope link  src 172.16.0.10
> 192.168.0.0/24 dev eth1  proto kernel  scope link  src 192.168.0.1
> 
> If for what ever reason IPSEC-SAs can not be established, maybe because
> 172.31.0.10 is down,
> the traffic comming from 192.168.0.0/24 will leave unencrypted the
> external (eth0) interface.

I can reproduce it with SNAT and MASQUERADE. Looks like this was
introduced back in 2011 with git commit 2774c131 ("xfrm: Handle
blackhole route creation via afinfo.").

Before that commit, xfrm_lookup() and __xfrm_lookup() returned
an error if we have a matching policy but no states. The route
lookup functions used __xfrm_lookup() and generated a blackhole
route if __xfrm_lookup() returned -EREMOTE. All other functions
used xfrm_lookup() which returned -EAGAIN. This was treated as
as an error and the packet was dropped immediately.

After this commit all callers to xfrm_lookup() rely that
dst_output() is called afterwards. This seems to be not the
case, at least when postrouting nat is used.

Maybe we should go back to let only the route lookup functions
genarate a blackhole route. Everyone else should better drop
the packets immediately.

I'll try to do a patch.