wireguard-mpls

Wireguard, but compatible with mpls, no MTU overhead.

How

• Decapsulate MPLS header (decap_mpls) before sending traffic out of WG interface
• Encapsulate MPLS header (encap_mpls) after receiving traffic from WG interface
• Nearly all magical stuff is in magic.h
• Use back progagation of TTL from inner IP header to restore MPLS TTL
• Add MESSAGE_DATA_MPLS = 5 to avoid conflicts with some providers (e.g., cloudflare)
  • Also add MESSAGE_DATA_MPLS_MC = 6 as multicast MPLS
• select 0.0.0.0 and :: as peer for non-IP(mpls) traffic

(IP) ==[encap]==> (MPLS|IP) ==[wg-decap]==> (WG|IP) ==[wg-encap]==> (MPLS|IP) ==[decap]==> (IP)

╔═══════════════════════════════════════════════════════════════════════════════════╗
║                               WireGuard Header                                    ║
╠════════════════════════════════════════════════════════════════╦══════════════════╣
║ type                                                           ║ others           ║
║ 32 bit                                                         ║ (we don't care)  ║
╠══════════════════════════════════════════╦═════════════════════╬══════════════════╣
║ reserved zero 24 bit                     ║ type 8 bit          ║                  ║
║ Used for MPLS Label/Exp/BoS              ║ new type for MPLS   ║                  ║
╚══════════════════════════════════════════╩═════════════════════╩══════════════════╝

╔═══════════════════════════════════════════════════════════════════════════════════╗
║                               MPLS Header                                         ║
╠══════════════════════════════╦═══════╦═══════╦════════════════════════════════════╣
║ Label                        ║ Exp   ║ BoS   ║ TTL                                ║
║ 20 bit                       ║ 3 bit ║ 1 bit ║ 8 bit                              ║
╠══════════════════════════════╩═══════╩═══════╬════════════════════════════════════╣
║   Put in Wireguard reserved zeros (24 bit)   ║ Extracted from inner IP header TTL ║
╚══════════════════════════════════════════════╩════════════════════════════════════╝

╔═══════════════════════════════════════════════════════════════════════════════════╗
║                            sk_buff reserved room                                  ║
╠══════════════════════╦════════════════════════════════════════════════════════════╣
║ type                 ║ MPLS Data (label + exp + bos)                              ║
║ 8 bit                ║ 24 bit                                                     ║
╠══════════════════════╬════════════════════════════════════════════════════════════╣
║ 0 for MPLS_UC        ║ used in wg pkt processing                                  ║
║ 1 for MPLS_MC        ║ (clear before finish post-receive processing)              ║
╚══════════════════════╩════════════════════════════════════════════════════════════╝

Source

The origin code is copied from wireguard kernel module from linux-6.8.2.

Compile

• make
• use wireguard.ko

Usage

Setup wireguard tunnels as usual, route MPLS traffic into wg interfaces, and enjoy.

Limit

• MTU is reduced to 1416, may fix it soon.
• Only IPv4/IPv6 packets are allowed to be MPLS payload, may add fallback option to accept more protocols. (Now supported)
  • However, TTL would be propagated from inner IP headers to MPLS headers, inner protocols without TTL would cause MPLS TTL to always be 64.
• Can not disable MPLS function. Use sysctl for temporary solution.
• Non-IP traffic would be sent to peers with allowed-ips 0.0.0.0/0 and/or ::/0.
• There will be MTU issues. For 2 stacked labels, working MTU is reduced to 1408.
• Multicast MPLS is never tested!

TODO

• MPLS type encoding (send)
• MPLS type decoding (receive)
• MPLS packet to IP (send)
• IP packet to MPLS (receive)
• TTL back propagation (receive)
• MPLS handing (receive)
• Multi-layer MPLS headers handling
• non-IP inner protocol handling
• Adjust MPLS to use MTU 1420
• BUG: encap ip packet from wg1, decap mpls and send to wg2

Test

• make test-install: replace current wireguard with wireguard-mpls.
• make test-tunnel: set up a tunnel from host to netns with MPLS routing.
• make test-router: set up 6 netns and 5 tunnels with MPLS routing, run traceroutes.
• make test-stack: set up a tunnel from host to netns with MPLS routing with stacked labels.

Benchmark

WIP (From the results of known tests, there is little difference in performance)