Post-quantum WireGuard

Andreas Hülsing, Kai-Chun Ning, Peter Schwabe, Florian Weber, Philip

R. Zimmermanna

October 25, 2021

a authors in alphabetic order

WireGuard

• Modern Virtual Private Network (VPN) protocol

• Presented by Donenfeld at NDSS 2017
• Merged into Linux kernel in 2020
• Only ≈4000 lines of code
• Runs over UDP

”Compared to horrors that are OpenVPN and IPSec, WireGuard is a work

of art”
—Linus Torvalds

1
“Cryptographically opinionated”

• No “crypto agility”
• Fixed suite of cryptographic primitives:
• X25519 as Diffie-Hellman routine
• ChaCha20-Poly1305 as AEAD
• Blake2s for hashing and keyed hashing
• HKDF for key derivation

2
“Cryptographically opinionated”

• No “crypto agility”
• Fixed suite of cryptographic primitives:
• X25519 as Diffie-Hellman routine
• ChaCha20-Poly1305 as AEAD
• Blake2s for hashing and keyed hashing
• HKDF for key derivation
• Focus today: the WireGuard handshake
• Authenticate parties to each other
• Establish a session key to encrypt payload data

2
“Cryptographically opinionated”

• No “crypto agility”
• Fixed suite of cryptographic primitives:
• X25519 as Diffie-Hellman routine
• ChaCha20-Poly1305 as AEAD
• Blake2s for hashing and keyed hashing
• HKDF for key derivation
• Focus today: the WireGuard handshake
• Authenticate parties to each other
• Establish a session key to encrypt payload data

2
Diffie-Hellman reminder

Alice Bob

A ← ga B ← gb

K ← Ba = (gb )a = gab K ← Ab = (ga )b = gab

3
Diffie-Hellman reminder

Attacker who can compute x given gx breaks Diffie-Hellman

This is known as Discrete-Logarithm Problem

3
4
“In the past, people have said, maybe it’s 50 years away, it’s a dream,
maybe it’ll happen sometime. I used to think it was 50. Now I’m thinking
like it’s 15 or a little more. It’s within reach. It’s within our lifetime. It’s
going to happen.”
—Mark Ketchen (IBM), Feb. 2012, about quantum computers

5
Post-quantum crypto

Definition
Post-quantum crypto is (asymmetric) crypto that resists attacks using
classical and quantum computers.

6
Post-quantum crypto

Definition
Post-quantum crypto is (asymmetric) crypto that resists attacks using
classical and quantum computers.

5 main directions
• Lattice-based crypto (PKE and Sigs)
• Code-based crypto (mainly PKE)
• Multivariate-based crypto (mainly Sigs)
• Hash-based signatures (only Sigs)
• Isogeny-based crypto (so far, mainly PKE)

6
NISTPQC

Overview tweeted by Jacob Alperin-Sheriff on Dec 4, 2017.

7
State of NIST PQC

• Nov 2017: 69 “complete and proper” submissions

8
State of NIST PQC

• Nov 2017: 69 “complete and proper” submissions

• Jan. 2019: Round-2 announcement planned at RWC

8
State of NIST PQC

• Nov 2017: 69 “complete and proper” submissions

• Jan. 2019: Round-2 announcement planned at RWC
• Feb. 2019: 26 round-2 candidates

8
State of NIST PQC

• Nov 2017: 69 “complete and proper” submissions

• Jan. 2019: Round-2 announcement planned at RWC
• Feb. 2019: 26 round-2 candidates
• Jun. 2020: Round-3 announcement planned

8
State of NIST PQC

• Nov 2017: 69 “complete and proper” submissions

• Jan. 2019: Round-2 announcement planned at RWC
• Feb. 2019: 26 round-2 candidates
• Jun. 2020: Round-3 announcement planned
• Jul. 2020: Round-3 announcement:
• 7 finalists
• 8 alternate schemes

8
What now?

• NIST is expected to announce winners in late 2021

• ≈ one year later get standards

9
What now?

• NIST is expected to announce winners in late 2021

• ≈ one year later get standards
• Replace existing crypto with new crypto

9
What now?

• NIST is expected to announce winners in late 2021

• ≈ one year later get standards
• Replace existing crypto with new crypto

Mission accomplished – The world is safe again!

9
What now?

• NIST is expected to announce winners in late 2021

• ≈ one year later get standards
• Replace existing crypto with new crypto

Mission accomplished – The world is safe again!

. . . or is it?

9
The WireGuard handshake (basic idea: “4DH”)

Initiator has long-term DH key-pair (sski , spki )

Responder has long-term DH key-pair (sskr , spkr )
Initiator Responder

(eski , epki ) ← DH.Gen()

epk
−−−−−−−−−−−−−−−−−−−−−−−−−−i−−−−−−−−−−−−−−−−−−−−−−→

(eskr , epkr ) ← DH.Gen()

epk
←−−−−−−−−−−−−−−−−−−−−−−−−−r−−−−−−−−−−−−−−−−−−−−−−−

k1 ← DH.Shared(eski , spkr ) k2 ← DH.Shared(sskr , epki )

k2 ← DH.Shared(sski , epkr ) k3 ← DH.Shared(eskr , spki )
k3 ← DH.Shared(eski , epkr ) k4 ← DH.Shared(eskr , epki )
k4 ← DH.Shared(sski , spkr ) k1 ← DH.Shared(sskr , spki )

Derive session key from k1 , k2 , k3 , and k4

10
The WireGuard handshake (high-level)
Initiator Responder
1: (eski , epki ) ← DH.Gen()
2:
$
sidi ← {0, 1}32
3: ltk ← AEAD.Enc(κ3 , 0, spki , H3 )
4: now ← Timestamp()
5: time ← AEAD.Enc(κ4 , 0, H4 , now)
6: m1 ← MAC(H(lbl3 ∥ spkr ), type ∥ 03 ∥ sidi ∥ epki ∥ ltk ∥ time)
7: m2 ← MAC(cookie, type ∥ 03 ∥ sidi ∥ epki ∥ ltk ∥ time ∥ m1)
8: InitHello ← type ∥ 03 ∥ sidi ∥ epki ∥ ltk ∥ time ∥ m1 ∥ m2

InitHello
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→

9: (eskr , epkr ) ← DH.Gen()

10:
$
sidr ← {0, 1}32
11: zero ← AEAD.Enc(κ9 , 0, H9 , ∅)
12: m1 ← MAC(H(lbl3 ∥ spki ), type ∥ 03 ∥ sidr ∥ sidi ∥ epkr ∥ zero)
13: m2 ← MAC(cookie, type ∥ 03 ∥ sidr ∥ sidi ∥ epkr ∥ zero ∥ m1)
14: RespHello ← type ∥ 03 ∥ sidr ∥ sidi ∥ epkr ∥ zero ∥ m1 ∥ m2

RespHello
←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−

15: tki ← KDF1 (C9 , ∅)

16: tkr ← KDF2 (C9 , ∅)

AEAD.Enc(tki , ·, ∅, application data)

−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→ 11
Handshake security

• Key confidentiality
• Entity authentication

12
Handshake security

• Key confidentiality
• Entity authentication
• Key uniqueness
• Identity hiding
• Replay attack resistance
• Unknown key-share (UKS) attack resistance
• DoS attack resistance (early reject)

12
WireGuard security proofs

• Computational: Dowling and Paterson, 2018

• eCK-PFS-PSK
• Assumes additional key-confirmation message
• Missing: key uniqueness, identity hiding, DoS mitigation
• Symbolic: partially by Donenfeld and Milner, 2017
• Missing: perfect forward secrecy, replay attack resistance, DoS
mitigation

13
Post-quantum security of WireGuard

• The optional PSK provides confidentiality against quantum

attacks.
• Assumption: PSK cannot be recovered by quantum attackers
• Post-quantum cryptography: Donenfeld claimed ’not practical for
use here’
• Applebaum, Martindale, Wu, 2019:
• Tweak to WireGuard protocol
• Send H(pk) instead of pk
• Quantum attacker does not easily get pk
• Resistance against mass-surveillance attackers

14
PQ-WireGuard – our goals

• Post-quantum confidentiality and authentication

• NIST security level 3 (≈AES-192)
• Retain all security properties of WireGuard
• Efficient 1-round-trip handshake

15
PQ-WireGuard – our goals

• Post-quantum confidentiality and authentication

• NIST security level 3 (≈AES-192)
• Retain all security properties of WireGuard
• Efficient 1-round-trip handshake
• No fragmentation
• Remember: WireGuard uses UDP
• Lost packets, filtering ⇒ more complex state machine
• Packet-size constraint:
• IPv6 guarantee: no fragmentation of packets ≤ 1280 bytes
• Fit WireGuard messages into 1232 bytes

15
PQ-WireGuard – the idea

1. Replace DH with key-encapsulation mechanisms (KEMs)

2. Instantiate with PQ KEMs achieving desired security

16
Diffie-Hellman

Alice Bob

A ← ga B ← gb

K ← Ba = (gb )a = gab K ← Ab = (ga )b = gab

17
Diffie-Hellman

Alice Bob

A ← ga B ← gb

K ← Ba = (gb )a = gab K ← Ab = (ga )b = gab

17
KEMs: as close as you’ll get to DH

Initiator Responder

(pk, sk) ← KEM.Gen

pk

(ct, K) ← KEM.Enc(pk)

ct

K ← KEM.Dec(ct, sk)

18
KEMs: as close as you’ll get to DH∗

Initiator Responder

(pk, sk) ← KEM.Gen

pk

(ct, K) ← KEM.Enc(pk)

ct

K ← KEM.Dec(ct, sk)

∗ Except with CSIDH (Castryck, Lange, Martindale, Renes, Panny, 2018)

18
What can KEMs (not) do?

Initiator Responder

(eski , epki ) ← DH.Gen()

epk
−−−−−−−−−−−−−−−−−−−−−−−−−−i−−−−−−−−−−−−−−−−−−−−−−→

(eskr , epkr ) ← DH.Gen()

epk
←−−−−−−−−−−−−−−−−−−−−−−−−−r−−−−−−−−−−−−−−−−−−−−−−−

k1 ← DH.Shared(eski , spkr ) k1 ← DH.Shared(sskr , epki )

k2 ← DH.Shared(sski , epkr ) k2 ← DH.Shared(eskr , spki )
k3 ← DH.Shared(eski , epkr ) k3 ← DH.Shared(eskr , epki )
k4 ← DH.Shared(sski , spkr ) k4 ← DH.Shared(sskr , spki )

19
A first approach with KEMs

Initiator Responder

(eski , epki ) ← CPAKEM.Gen()

r1 ← {0, 1}λ , (c1 , k1 ) ← CCAKEM.Enc(spkr , r1 )
$

epk , c1
−−−−−−−−−−−−−−−−−−−−−−−−−i −−−−−−−−−−−−−−−−−−−−−−−→

r2 ← {0, 1}λ , (c2 , k2 ) ← CCAKEM.Enc(spki , r2 )

$

r3 ← {0, 1}λ , (c3 , k3 ) ← CPAKEM.Enc(epki , r3 )

$

c2 , c3
←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−

k1 ← CCAKEM.Dec(sskr , c1 )
k2 ← CCAKEM.Dec(sski , c2 )
k3 ← CPAKEM.Dec(eski , c3 )

20
What are we lacking?

DoS resistance
• First initiator message is unauthenticated
• Solution: Use (optional) pre-shared key for early rejects

21
What are we lacking?

DoS resistance
• First initiator message is unauthenticated
• Solution: Use (optional) pre-shared key for early rejects

“MEX” resistance
• Some security also if all RNGs are insecure
• Static-static DH for confidentiality from long-term keys
• Solution: Use “NAXOS trick”

21
What are we lacking?

DoS resistance
• First initiator message is unauthenticated
• Solution: Use (optional) pre-shared key for early rejects

“MEX” resistance
• Some security also if all RNGs are insecure
• Static-static DH for confidentiality from long-term keys
• Solution: Use “NAXOS trick”

UKS-attack resistance
• WireGuard does not hash public keys into session key
• UKS resistance derived from static-static DH
• Solution: Use default PSK as H(spki ⊕ spkr )

21
(Most of) the PQ-WireGuard handshake
Initiator Responder
1: (eski , epki ) ← CPAKEM.Gen()
2: sidi ← {0, 1}32
$

3: ri ← {0, 1}λ
4: (ct1, shk1) ← CCAKEM.Enc(spkr , KDF1 (σi , ri ))
5: ltk ← AEAD.Enc(κ3 , 0, H(spki ), H3 )
6: now ← Timestamp()
7: time ← AEAD.Enc(κ4 , 0, H4 , now)
8: m1 ← MAC(H(lbl3 ∥ spkr ), type ∥ 03 ∥ sidi ∥ epki ∥ ct1 ∥ ltk ∥ time)
9: m2 ← MAC(cookie, type ∥ 03 ∥ sidi ∥ epki ∥ ct1 ∥ ltk ∥ time ∥ m1)
10: InitHello ← type ∥ 03 ∥ sidi ∥ epki ∥ ct1 ∥ ltk ∥ time ∥ m1 ∥ m2
InitHello
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→

11: e, rr ← {0, 1}λ × {0, 1}λ

12: (ct2, shk2) ← CPAKEM.Enc(epki , e)
13: (ct3, shk3) ← CCAKEM.Enc(spki , KDF1 (σr , rr ))
14:
$
sidr ← {0, 1}32
15: zero ← AEAD.Enc(κ9 , 0, H9 , ∅)
16: m1 ← MAC(H(lbl3 ∥ spki ), type ∥ 03 ∥ sidr ∥ sidi ∥ ct2 ∥ ct3 ∥ zero)
17: m2 ← MAC(cookie, type ∥ 03 ∥ sidr ∥ sidi ∥ ct2 ∥ ct3 ∥ zero ∥ m1)
18: RespHello ← type ∥ 03 ∥ sidr ∥ sidi ∥ ct2 ∥ ct3 ∥ zero ∥ m1 ∥ m2
RespHello
←−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
22
Adding explicit key confirmation

Initiator Responder
19: conf ← AEAD.Enc(κ10 , 0, H10 , ∅)
20: m1 ← MAC(H(lbl3 ∥ spkr ), type ∥ 03 ∥ sidi ∥ sidr ∥ conf)
21: m2 ← MAC(cookie, type ∥ 03 ∥ sidi ∥ sidr ∥ conf ∥ m1)
22: InitConf ← type ∥ 03 ∥ sidi ∥ sidr ∥ conf ∥ m1 ∥ m2
InitConf
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→

23: tki ← KDF1 (C10 , ∅)

24: tkr ← KDF2 (C10 , ∅)

• Allows proofs to separate handshake from data transmission

• eCK-PFS-PSK proof applies to actual protocol

23
PQ-WireGuard security proofs

• Computational:
• Based on Dowling and Paterson (2018)
• Proof in the eCK-PFS-PSK model
• Standard model proof
• Symbolic:
• Based on Donenfeld and Milner (2017)
• Uses the Tamarin prover
• Cover all desired security properties

24
Instantiation

• Long-term IND-CCA-secure KEM: Classic McEliece

• Smallest ciphertext of all NIST PQC candidates
• Public-key size does not matter
• Key-generation time does not matter

25
Instantiation

• Long-term IND-CCA-secure KEM: Classic McEliece

• Smallest ciphertext of all NIST PQC candidates
• Public-key size does not matter
• Key-generation time does not matter
• Ephemeral IND-CPA-secure KEM requirements:
• NIST PQC round-2 candidate at level 3
• High-speed constant-time implementation
• Pick “conservative” primitives
• No patent claims by submitters
• No tweaks that lower security

25
Instantiation

• Long-term IND-CCA-secure KEM: Classic McEliece

• Smallest ciphertext of all NIST PQC candidates
• Public-key size does not matter
• Key-generation time does not matter
• Ephemeral IND-CPA-secure KEM requirements:
• NIST PQC round-2 candidate at level 3
• High-speed constant-time implementation
• Pick “conservative” primitives
• No patent claims by submitters
• No tweaks that lower security
• Fit into unfragmented IPv6 packet:
• public key of ≤928 bytes
• ciphertext of ≤984 bytes

25
Dagger

• Only three NIST round-2 candidates within size constraints:

• SIKE – not high-speed
• ROLLO – not conservative
• Round5 – patent encumbered

26
Dagger

• Only three NIST round-2 candidates within size constraints:

• SIKE – not high-speed
• ROLLO – not conservative
• Round5 – patent encumbered
• Idea: Tweak lattice-based KEM:
• More public-key and ciphertext compression
• Increase hardness of lattice problems
• Increase failure probability (no issue for CPA sec.)

26
Dagger

• Only three NIST round-2 candidates within size constraints:

• SIKE – not high-speed
• ROLLO – not conservative
• Round5 – patent encumbered
• Idea: Tweak lattice-based KEM:
• More public-key and ciphertext compression
• Increase hardness of lattice problems
• Increase failure probability (no issue for CPA sec.)
• Tweaked (smaller, more lightweight) Saber: Dagger

26
Implementation and Evaluation

• Implement as Linux kernel module

• Use existing high-speed constant-time software for McEliece and
Dagger (Saber)

27
Implementation and Evaluation

• Implement as Linux kernel module

• Use existing high-speed constant-time software for McEliece and
Dagger (Saber)
• Metrics for comparison:
• Amount of traffic
• Number of packets
• Handshake latency

27
Implementation and Evaluation

• Implement as Linux kernel module

• Use existing high-speed constant-time software for McEliece and
Dagger (Saber)
• Metrics for comparison:
• Amount of traffic
• Number of packets
• Handshake latency
• Use virtual 10Gbps Ethernet link between two VMs
• Both IPv4 and IPv6: similar results
• Compare with WireGuard, OpenVPN, IPsec, PQCrypto-VPN

27
Results

Packet Traffic Client Time Server Time

VPN Software
Number (bytes) (milliseconds) (milliseconds)
WireGuard 3 458 0.592 0.480
(0) (0) (0.399) (0.389)
PQ-WireGuard 3 2654 1.015 0.786
(0) (0) (0.618) (0.621)
IPsec 6 4299 17.188 11.912
(RSA-2048) (0) (0) (0.712) (0.535)
IPsec 4 2281 5.226 2.822
(Curve25519) (0) (0) (0.575) (0.436)
OpenVPN 21.003 7955.409 1148.733 1142.650
(RSA-2048) (0.055) (7.319) (250.513) (243.184)
OpenVPN 19.005 5788.610 1139.140 1133.944
(NIST P-256) (0.007) (9.423) (247.659) (240.691)
OpenVPN-NL 19.005 6065.700 1162.649 1151.790
(RSA-2048) (0.072) (9.665) (261.078) (246.363)
OpenVPN-NL 19.001 6061.138 1159.627 1153.949
(NIST P-256) (0.003) (4.304) (252.989) (247.470)
PQ-OpenVPN 63.006 35608.817 1160.922 1155.713
(Frodo-752 [BCD+16]) (0.078) (10.324) (259.246) (245.614)
PQ-OpenVPN 23.005 8996.684 1277.172 1269.074
(SIDHp503 ) (0.072) (9.449) (251.461) (257.427)

28
More online

Paper:
https://cryptojedi.org/papers/#pqwireguard

Code:
https://cryptojedi.org/crypto/#pqwireguard

29