[strongSwan] day in the life of an Encapsulating Security Payload(ESP) packet in Linux

[Martin T]

Hi,

I have a simple site-to-site IPSec VPN where "server-A" is connected
to a "firewall-A" over an IPSec tunnel. In front of "server-A" there
is a switch with has a 1500 byte MTU interface facing the server.
Sometimes clients behind "firewall-A" send large packets to "server-A"
and server replies with ICMP "unreachable; frag needed" messages:

11:19:22.309296 IP 10.10.10.135 > 192.168.100.4: ICMP 10.10.10.135
unreachable - need to frag (mtu 1438), length 36

10.10.10.135 is the IP address on "server-A" eth0 interface and
192.168.100.4 is the IP address of the end-client behind "firewall-A".

Am I correct that ICMP "unreachable; frag needed" messages are sent
only in case (server acting as) a router wants to route a package to
another interface, but this interface has a smaller MTU than the
package and router is not allowed to fragment this package because DF
flag is set? If yes, then "server-A" does not do any routing. It's
"left side" Strongswan configuration is following:

        left=10.10.10.135
        leftsourceip=10.10.10.135
        leftsubnet=10.10.10.135/32
        leftnexthop=%defaultroute


When I check the packet flow diagram in
Wikipedia(https://upload.wikimedia.org/wikipedia/commons/3/37/Netfilter-packet-flow.svg),
then ESP packet should traverse "routing decision" step twice- first
when the packet is encapsulated and second time when it is already
decapsulated. However, in both occasions the destination IP address is
10.10.10.135, which is a local physical interface(eth0) in "server-A":

server-A# ip route get 10.10.10.135
local 10.10.10.135 dev lo  src 10.10.10.135
    cache <local>
server-A#

..and thus the routing is done via lo interface which has a 64KB MTU.

Could anybody explain the day in the life of an ESP packet in Linux
and why does it sent out those ICMP "unreachable; frag needed"
messages?


thanks,
Martin