Troubleshooting Blocked Log Entries due to Asymmetric Routing¶
Asymmetric routing happens when traffic between two nodes takes a different path in each direction (e.g. A->B->C, C->D->A), it can be a problem for TCP which has strict state tracking but often does not affect “stateless” protocols such as ICMP or UDP.
What happens in most cases is this:
- Client sends a TCP SYN packet, which arrives at pfSense and gets a state table entry
- pfSense sends back an ICMP redirect letting the client know to reach the target server via the alternate gateway
- Server sends back a TCP SYN+ACK packet by some other path that pfSense doesn’t see
- Client sends its ACK and further responses back by its other gateway that are not seen by pfSense
- After 30 seconds, pfSense removes its state table entry as the connection was never completed as observed by pfSense
- Some time later, the client’s ICMP redirect learned route expires and the client sends another packet back to pfSense
- Since this packet is not starting a new connection, the packet is dropped, and the client gets disconnected since it now has no way to reach the destination.
The Bypass firewall rules for traffic on the same interface option located under System > Advanced on the Firewall/NAT tab activates rules for traffic to/from the static route networks which are much more permissive when it comes to creating states for TCP traffic and allowing it to pass. The rules allow any TCP packets, regardless of their flags, to create a state and also have the “sloppy state” type set which performs a less strict state match.
The same rules may be created manually by adding one on the affected interface tab (e.g. LAN), and a second rule on the Floating tab using the same interface (LAN again) to match the traffic in the out direction. The rule must be set for a protocol of TCP, under TCP flags check Any Flags, and use a State Type of Sloppy State. The options for TCP flags and State Type can be found in the Advanced Features of the rules, under the normal options.
On occasion these issues can be caused by other factors that lead to asymmetric routing, such as issues with route-to or reply-to, both having to do with gateways on interface settings.
Defining gateways under System > Routing does not cause this, but rather these problems can come up when the gateway is improperly configured on the interface pages, Interfaces > WAN, Interfaces > LAN, and so on.
Gateway set when it should not be set¶
If a gateway is set on an internal interface, such as LAN, it can cause problematic behavior. Setting a gateway on an internal interface will tag that interface’s outbound rules with route-to, and inbound rules with reply-to which will cause packets to be forwarded to the defined gateway rather than following their natural path. For WANs this is typically a good thing! For LANs it is not. Among other ill effects, it can lead to a loop of sorts where packets bounce between the firewall and the defined gateway, eventually being blocked or dropped when their TTL expires.
Gateway not set when it should be set¶
A gateway should usually be set on a WAN or other external-type interface settings (MPLS, IP VPN, etc.) In these cases the reply-to and route-to behavior is desired and likely required. If it is missing the packets may be blocked or dropped as they attempt to leave the wrong interface. A packet could enter via the alternate WAN, but the reply would leave by the default gateway. Similar to the effect seen when improperly using an Interface Group for WAN interfaces.