Secure Communication Channel Establishment: TLS 1.3 (over TCP Fast Open) vs. QUIC

Shan Chen, Samuel Jero, Matthew Jagielski, Alexandra Boldyreva, Cristina Nita-Rotaru

TL;DR

This paper presents a novel, comprehensive security analysis framework for comparing secure communication protocols like TLS 1.3 and QUIC, going beyond cryptographic cores to include practical packet-level attacks. It reveals vulnerabilities in existing protocols, particularly concerning availability and header integrity, and highlights QUIC[TLS] as offering the strongest overall security.

What Problem Does This Paper Address?

Establishing secure communication channels over the internet is crucial, with protocols like TLS and QUIC aiming to do this efficiently. While TLS 1.3 and QUIC promise reduced latency (fewer round trips before data transmission), their security when layered with underlying transport protocols (like TCP Fast Open for TLS) hasn't been thoroughly compared. Existing security models often focus on the cryptographic primitives rather than the practical implications of how these protocols interact with transport layers and how they handle packet-level attacks that can impact availability and integrity. This paper aims to bridge this gap by providing a unified framework to rigorously compare the security and availability properties of these layered protocols.

Methods

The authors introduce a new formal security model called "multi-stage authenticated and confidential channel establishment" (msACCE). This model extends previous work to handle protocols with multiple stages of key exchange and secure channel phases, accommodating features like session resumption and resets. Crucially, it incorporates novel security notions beyond standard confidentiality and authentication, including:

  • IP-Spoofing Prevention: Ensuring that connections are indeed from the claimed IP address.
  • Key Exchange (KE) Header Integrity: Protecting the integrity of key exchange packet headers.
  • KE Payload Integrity: Protecting the integrity of key exchange packet payloads.
  • Secure Channel (SC) Header Integrity: Protecting the integrity of secure channel packet headers.
  • Reset Authentication: Ensuring that session resets are only initiated by legitimate parties.

These models allow for a "layered" security analysis, considering the interplay between transport and application layers. The authors then apply these models to analyze three specific protocol combinations: TLS 1.3 over TCP Fast Open (TFO), QUIC over UDP, and a new variant, QUIC[TLS] (which uses TLS 1.3's key exchange).

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Beyond Cryptography: The paper strongly emphasizes that focusing solely on the cryptographic core of protocols is insufficient. The interaction with transport layers (like TCP's unauthenticated headers) introduces significant vulnerabilities that are often overlooked.
  • Availability is Key: Many new attacks identified target the availability of the connection (e.g., causing delays or connection drops) rather than breaking confidentiality. These attacks are often harder to detect by clients.
  • TFO's Weakness: While TCP Fast Open (TFO) aims to speed up TCP connections, its unauthenticated headers make TLS 1.3 over TFO vulnerable to attacks that can undermine its performance benefits, such as TFO cookie removal and TCP reset attacks.
  • QUIC's Strengths: QUIC, by implementing its own transport functionalities over UDP, gains better control over packet integrity and authentication, leading to stronger security properties in several areas compared to TFO+TLS 1.3.
  • QUIC[TLS] as a Strong Contender: The combination of QUIC's transport layer with TLS 1.3's key exchange (QUIC[TLS]) emerges as the most robust, particularly in its guaranteed Reset Authentication.

Key Results

The paper provides a detailed security comparison summarized in a table (Table 2 in the paper). Key findings include:

  • IP-Spoofing Prevention: All three protocols (TFO+TLS 1.3, UDP+QUIC, UDP+QUIC[TLS]) achieve IP-spoofing prevention, with TFO+TLS 1.3 relying on TFO's cookie mechanism and UDP+QUIC/QUIC[TLS] on source-address tokens.
  • KE Header Integrity: None of the three protocols achieve KE Header Integrity, leading to attacks like TFO cookie removal.
  • KE Payload Integrity: TFO+TLS 1.3 achieves KE Payload Integrity, while UDP+QUIC and UDP+QUIC[TLS] do not, leading to attacks like "ServerReject Triggering."
  • SC Header Integrity: UDP+QUIC and UDP+QUIC[TLS] achieve SC Header Integrity, but TFO+TLS 1.3 does not due to unauthenticated TCP headers, exposing it to attacks like TCP flow control manipulation and acknowledgment injection.
  • Reset Authentication: Only UDP+QUIC[TLS] guarantees Reset Authentication, fixing a vulnerability present in TFO+TLS 1.3 (TCP Reset Attack) and UDP+QUIC (PublicReset Attack).

Limitations and Future Work

  • Model Scope: The msACCE model focuses on application and transport layers, and specifically IP-spoofing for network layer attacks. It does not cover attacks on lower network layers (link, physical) or more complex network-layer attacks beyond spoofing.
  • 0-RTT Security: As noted by prior work, 0-RTT data in TLS 1.3 and QUIC generally does not provide forward secrecy and is vulnerable to replay attacks. While the paper acknowledges constructions for this, they are computationally expensive.
  • Practical Impact of Theoretical Attacks: Some identified attacks (e.g., in KE Header Integrity for UDP+QUIC[TLS]) are noted as theoretical with no immediate practical impact, but highlight potential areas for future exploitation.
  • Future Work: The authors hope their models will aid protocol designers in future analyses and help practitioners understand protocol limitations. The paper implicitly suggests that future protocols should better integrate and authenticate transport functionalities to mitigate availability attacks.

Summary updated

BibTeX
@article{chen2019secure,
  author = {Shan Chen and Samuel Jero and Matthew Jagielski and Alexandra Boldyreva and Cristina Nita-Rotaru},
  title = {{Secure Communication Channel Establishment: TLS 1.3 (over TCP Fast Open) vs. QUIC}},
  journal = {IACR Cryptol. ePrint Arch.},
  year = {2019},
  url = {https://openreview.net/forum?id=gQso6vnRlE},
}

Loading related content...

Figure