Bug #11494


Wireguard interface sends ICMP Redirect when routing between two peers

Added by Blaine Palmer 10 months ago. Updated 9 months ago.

Target version:
Start date:
Due date:
% Done:


Estimated time:
Plus Target Version:
Release Notes:
Affected Version:
Affected Architecture:


When PFSense is used to route traffic between two WireGuard peers, it send ICMP Redirect when both peers are on the same WireGuard interface.

Example topology

Home PFSense
- WG0

Concentrator PFSense
- WG0

MacOS Client
- WG0

Abridge Concentrator WG Config

  1. Peer: ConcentratorToHome
    AllowedIPs =,
  1. Peer: ConcentratorToMacbook
    AllowedIPs =

In this scenario, with traffic being allowed from (MacBook Source) -> (Concentrator router #1) -> (Home router #2) -> (destination) you will see something like the following

64 bytes from icmp_seq=0 ttl=63 time=68.083 ms
92 bytes from Redirect Host(New addr:

I believe this is occurring because from Concentrator's perspective, the source and destination all route through the same interface. In most situations I think this behavior is correct, however in the case of WireGuard where multiple peers exist on the same interface I think ICMP Redirect should not be sent.

Actions #1

Updated by Blaine Palmer 10 months ago

Just for reference, it appears a similar issue was observed early in WireGuard's original development.

Similarly, it appears this similar code made it into the Linux Kernel implementation as well

if (dev_v4) {
/* At some point we might put this check near the ip_rt_send_ * redirect call of ip_forward in net/ipv4/ip_forward.c, similar * to the current secpath check.
IPV4_DEVCONF_ALL(dev_net(dev), SEND_REDIRECTS) = false;
Actions #2

Updated by Blaine Palmer 10 months ago

It would appear this may need to be corrected in the FreeBSD upstream.

Possibly relevant:
Line #119 (sysctl check)
Line #1040 (conditional logic)
Line 185 VNET_DECLARE(int, ipsendredirects);
Line 201 #define V_ipsendredirects VNET

I tried to poke around and look for other interfaces that were selectively disabling forward/redirect or other sysctl, but am not familiar enough with the code.
Thank you!

Actions #3

Updated by Blaine Palmer 10 months ago

One last interesting tidbit, similar assumptions causing issues with p2p interfaces in ipv6 which caused issues for WG pre-kernel implementation.

Actions #4

Updated by Jim Pingle 10 months ago

  • Priority changed from Normal to Low
  • Target version changed from 2.5.1 to CE-Next

This is likely a (mostly?) harmless side effect of how the routes in the routing table are added for WireGuard. Because they are added to the interface directly and not to a remote peer (due to how WireGuard operates), redirects are triggered.

This isn't limited to WireGuard. There is an old IPsec workaround to nudge firewall traffic into a policy-based IPsec tunnel which involves added a route using the firewall itself as a gateway. When using that kind of route, hosts on LAN also get the redirects.

If you know you have no need for ICMP redirects on any interface, you can also disable them completely by setting system tunables for:

Actions #5

Updated by Blaine Palmer 10 months ago

I've disabled redirect via the sysctl/tunable as suggested already.

Just to clarify this is for every incoming packet that is forwarded by FreeBSD/pfSense from one peer on a WireGuard interface to another peer on the same WireGuard interface.

I think the concern would be the overhead this would add unnecessarily. If I understand the FreeBSD ip_forward function properly, it will be sending a redirect for every non-fragmented packet forwarded.

I will test this and get back to you shortly with some packet statistics to confirm if this assertion is correct.

Actions #6

Updated by Jim Pingle 10 months ago

Another workaround is to do one peer per tunnel and a dynamic routing protocol like BGP, or routes using the remote peer tunnel address as the gateway (fill in the Peer WireGuard Address appropriately, leave Allowed IPs blank, add your own manual static routes)

If it's the routes targeting the interface itself to blame, that should take it out of the equation.

Actions #7

Updated by Blaine Palmer 10 months ago

I was able to confirm that there does not appear to be any rate limiting, the overhead isn't terrible though as the ICMP packets are so small. Getting statistics was a little more difficult on this new MacBook however and your proposed solution is the best I believe. This simplifies firewall rules and routing semantics.

Thank you

Actions #8

Updated by Jim Pingle 9 months ago

  • Target version changed from CE-Next to Future

Also available in: Atom PDF