Using TLS to Secure QUIC

TLS provides two endpoints with a way to establish a means of communication over an untrusted medium (that is, the Internet) that ensures that messages they exchange cannot be observed, modified, or forged.¶

Internally, TLS is a layered protocol, with the structure shown in Figure 1.¶

Each Handshake layer message (e.g., Handshake, Alerts, and Application Data) is carried as a series of typed TLS records by the Record layer. Records are individually cryptographically protected and then transmitted over a reliable transport (typically TCP) which provides sequencing and guaranteed delivery.¶

The TLS authenticated key exchange occurs between two endpoints: client and server. The client initiates the exchange and the server responds. If the key exchange completes successfully, both client and server will agree on a secret. TLS supports both pre-shared key (PSK) and Diffie-Hellman over either finite fields or elliptic curves ((EC)DHE) key exchanges. PSK is the basis for 0-RTT; the latter provides perfect forward secrecy (PFS) when the (EC)DHE keys are destroyed.¶

After completing the TLS handshake, the client will have learned and authenticated an identity for the server and the server is optionally able to learn and authenticate an identity for the client. TLS supports X.509 [RFC5280] certificate-based authentication for both server and client.¶

The TLS key exchange is resistant to tampering by attackers and it produces shared secrets that cannot be controlled by either participating peer.¶

• A full 1-RTT handshake in which the client is able to send Application Data after one round trip and the server immediately responds after receiving the first handshake message from the client.
• A 0-RTT handshake in which the client uses information it has previously learned about the server to send Application Data immediately. This Application Data can be replayed by an attacker so it MUST NOT carry a self-contained trigger for any non-idempotent action.

A simplified TLS handshake with 0-RTT application data is shown in Figure 2.¶

Figure 2 omits the EndOfEarlyData message, which is not used in QUIC; see Section 8.3. Likewise, neither ChangeCipherSpec nor KeyUpdate messages are used by QUIC. ChangeCipherSpec is redundant in TLS 1.3; see Section 8.4. QUIC has its own key update mechanism; see Section 6.¶

• Initial Keys
• Early Data (0-RTT) Keys
• Handshake Keys
• Application Data (1-RTT) Keys

Application Data may appear only in the Early Data and Application Data levels. Handshake and Alert messages may appear in any level.¶

The 0-RTT handshake is only possible if the client and server have previously communicated. In the 1-RTT handshake, the client is unable to send protected Application Data until it has received all of the Handshake messages sent by the server.¶

• Sending and receiving handshake messages
• Processing stored transport and application state from a resumed session and determining if it is valid to accept early data
• Rekeying (both transmit and receive)
• Handshake state updates

Additional functions might be needed to configure TLS.¶

This document describes how TLS 1.3 [TLS13] is used with QUIC.¶

In practice, the TLS handshake will negotiate a version of TLS to use. This could result in a newer version of TLS than 1.3 being negotiated if both endpoints support that version. This is acceptable provided that the features of TLS 1.3 that are used by QUIC are supported by the newer version.¶

A badly configured TLS implementation could negotiate TLS 1.2 or another older version of TLS. An endpoint MUST terminate the connection if a version of TLS older than 1.3 is negotiated.¶

The first Initial packet from a client contains the start or all of its first cryptographic handshake message, which for TLS is the ClientHello. Servers might need to parse the entire ClientHello (e.g., to access extensions such as Server Name Identification (SNI) or Application Layer Protocol Negotiation (ALPN)) in order to decide whether to accept the new incoming QUIC connection. If the ClientHello spans multiple Initial packets, such servers would need to buffer the first received fragments, which could consume excessive resources if the client’s address has not yet been validated. To avoid this, servers MAY use the Retry feature (see Section 8.1 of [QUIC-TRANSPORT]) to only buffer partial ClientHello messages from clients with a validated address.¶

QUIC packet and framing add at least 36 bytes of overhead to the ClientHello message. That overhead increases if the client chooses a connection ID without zero length. Overheads also do not include the token or a connection ID longer than 8 bytes, both of which might be required if a server sends a Retry packet.¶

A typical TLS ClientHello can easily fit into a 1200 byte packet. However, in addition to the overheads added by QUIC, there are several variables that could cause this limit to be exceeded. Large session tickets, multiple or large key shares, and long lists of supported ciphers, signature algorithms, versions, QUIC transport parameters, and other negotiable parameters and extensions could cause this message to grow.¶

For servers, in addition to connection IDs and tokens, the size of TLS session tickets can have an effect on a client’s ability to connect efficiently. Minimizing the size of these values increases the probability that clients can use them and still fit their ClientHello message in their first Initial packet.¶

The TLS implementation does not need to ensure that the ClientHello is sufficiently large. QUIC PADDING frames are added to increase the size of the packet as necessary.¶

The requirements for authentication depend on the application protocol that is in use. TLS provides server authentication and permits the server to request client authentication.¶

A client MUST authenticate the identity of the server. This typically involves verification that the identity of the server is included in a certificate and that the certificate is issued by a trusted entity (see for example [RFC2818]).¶

A server MAY request that the client authenticate during the handshake. A server MAY refuse a connection if the client is unable to authenticate when requested. The requirements for client authentication vary based on application protocol and deployment.¶

A server MUST NOT use post-handshake client authentication (as defined in Section 4.6.2 of [TLS13]), because the multiplexing offered by QUIC prevents clients from correlating the certificate request with the application-level event that triggered it (see [HTTP2-TLS13]). More specifically, servers MUST NOT send post-handshake TLS CertificateRequest messages and clients MUST treat receipt of such messages as a connection error of type PROTOCOL_VIOLATION.¶

QUIC can use the session resumption feature of TLS 1.3. It does this by carrying NewSessionTicket messages in CRYPTO frames after the handshake is complete. Session resumption is the basis of 0-RTT, but can be used without also enabling 0-RTT.¶

Endpoints that use session resumption might need to remember some information about the current connection when creating a resumed connection. TLS requires that some information be retained; see Section 4.6.1 of [TLS13]. QUIC itself does not depend on any state being retained when resuming a connection, unless 0-RTT is also used; see Section 4.6 and Section 7.4.1 of [QUIC-TRANSPORT]. Application protocols could depend on state that is retained between resumed connections.¶

Clients can store any state required for resumption along with the session ticket. Servers can use the session ticket to help carry state.¶

Session resumption allows servers to link activity on the original connection with the resumed connection, which might be a privacy issue for clients. Clients can choose not to enable resumption to avoid creating this correlation. Client SHOULD NOT reuse tickets as that allows entities other than the server to correlate connections; see Appendix C.4 of [TLS13].¶

To communicate their willingness to process 0-RTT data, servers send a NewSessionTicket message that contains the “early_data” extension with a max_early_data_size of 0xffffffff; the amount of data which the client can send in 0-RTT is controlled by the “initial_max_data” transport parameter supplied by the server. Servers MUST NOT send the “early_data” extension with a max_early_data_size set to any value other than 0xffffffff. A client MUST treat receipt of a NewSessionTicket that contains an “early_data” extension with any other value as a connection error of type PROTOCOL_VIOLATION.¶

A client that wishes to send 0-RTT packets uses the “early_data” extension in the ClientHello message of a subsequent handshake (see Section 4.2.10 of [TLS13]). It then sends the application data in 0-RTT packets.¶

A server accepts 0-RTT by sending an early_data extension in the EncryptedExtensions (see Section 4.2.10 of [TLS13]). The server then processes and acknowledges the 0-RTT packets that it receives.¶

A server rejects 0-RTT by sending the EncryptedExtensions without an early_data extension. A server will always reject 0-RTT if it sends a TLS HelloRetryRequest. When rejecting 0-RTT, a server MUST NOT process any 0-RTT packets, even if it could. When 0-RTT was rejected, a client SHOULD treat receipt of an acknowledgement for a 0-RTT packet as a connection error of type PROTOCOL_VIOLATION, if it is able to detect the condition.¶

When 0-RTT is rejected, all connection characteristics that the client assumed might be incorrect. This includes the choice of application protocol, transport parameters, and any application configuration. The client therefore MUST reset the state of all streams, including application state bound to those streams.¶

A client MAY attempt to send 0-RTT again if it receives a Retry or Version Negotiation packet. These packets do not signify rejection of 0-RTT.¶

When a server receives a ClientHello with the “early_data” extension, it has to decide whether to accept or reject early data from the client. Some of this decision is made by the TLS stack (e.g., checking that the cipher suite being resumed was included in the ClientHello; see Section 4.2.10 of [TLS13]). Even when the TLS stack has no reason to reject early data, the QUIC stack or the application protocol using QUIC might reject early data because the configuration of the transport or application associated with the resumed session is not compatible with the server’s current configuration.¶

QUIC requires additional transport state to be associated with a 0-RTT session ticket. One common way to implement this is using stateless session tickets and storing this state in the session ticket. Application protocols that use QUIC might have similar requirements regarding associating or storing state. This associated state is used for deciding whether early data must be rejected. For example, HTTP/3 ([QUIC-HTTP]) settings determine how early data from the client is interpreted. Other applications using QUIC could have different requirements for determining whether to accept or reject early data.¶

In TLS over TCP, the HelloRetryRequest feature (see Section 4.1.4 of [TLS13]) can be used to correct a client’s incorrect KeyShare extension as well as for a stateless round-trip check. From the perspective of QUIC, this just looks like additional messages carried in Initial packets. Although it is in principle possible to use this feature for address verification in QUIC, QUIC implementations SHOULD instead use the Retry feature (see Section 8.1 of [QUIC-TRANSPORT]). HelloRetryRequest is still used to request key shares.¶

If TLS experiences an error, it generates an appropriate alert as defined in Section 6 of [TLS13].¶

A TLS alert is converted into a QUIC connection error. The alert description is added to 0x100 to produce a QUIC error code from the range reserved for CRYPTO_ERROR. The resulting value is sent in a QUIC CONNECTION_CLOSE frame of type 0x1c.¶

The alert level of all TLS alerts is “fatal”; a TLS stack MUST NOT generate alerts at the “warning” level.¶

QUIC permits the use of a generic code in place of a specific error code; see Section 11 of [QUIC-TRANSPORT]. For TLS alerts, this includes replacing any alert with a generic alert, such as handshake_failure (0x128 in QUIC). Endpoints MAY use a generic error code to avoid possibly exposing confidential information.¶

After QUIC moves to a new encryption level, packet protection keys for previous encryption levels can be discarded. This occurs several times during the handshake, as well as when keys are updated; see Section 6.¶

Packet protection keys are not discarded immediately when new keys are available. If packets from a lower encryption level contain CRYPTO frames, frames that retransmit that data MUST be sent at the same encryption level. Similarly, an endpoint generates acknowledgements for packets at the same encryption level as the packet being acknowledged. Thus, it is possible that keys for a lower encryption level are needed for a short time after keys for a newer encryption level are available.¶

An endpoint cannot discard keys for a given encryption level unless it has both received and acknowledged all CRYPTO frames for that encryption level and when all CRYPTO frames for that encryption level have been acknowledged by its peer. However, this does not guarantee that no further packets will need to be received or sent at that encryption level because a peer might not have received all the acknowledgements necessary to reach the same state.¶

Though an endpoint might retain older keys, new data MUST be sent at the highest currently-available encryption level. Only ACK frames and retransmissions of data in CRYPTO frames are sent at a previous encryption level. These packets MAY also include PADDING frames.¶

QUIC derives packet protection keys in the same way that TLS derives record protection keys.¶

Each encryption level has separate secret values for protection of packets sent in each direction. These traffic secrets are derived by TLS (see Section 7.1 of [TLS13]) and are used by QUIC for all encryption levels except the Initial encryption level. The secrets for the Initial encryption level are computed based on the client’s initial Destination Connection ID, as described in Section 5.2.¶

The keys used for packet protection are computed from the TLS secrets using the KDF provided by TLS. In TLS 1.3, the HKDF-Expand-Label function described in Section 7.1 of [TLS13] is used, using the hash function from the negotiated cipher suite. Other versions of TLS MUST provide a similar function in order to be used with QUIC.¶

The current encryption level secret and the label “quic key” are input to the KDF to produce the AEAD key; the label “quic iv” is used to derive the IV; see Section 5.3. The header protection key uses the “quic hp” label; see Section 5.4. Using these labels provides key separation between QUIC and TLS; see Section 9.6.¶

The KDF used for initial secrets is always the HKDF-Expand-Label function from TLS 1.3 (see Section 5.2).¶

The hash function for HKDF when deriving initial secrets and keys is SHA-256 [SHA].¶

The connection ID used with HKDF-Expand-Label is the Destination Connection ID in the Initial packet sent by the client. This will be a randomly-selected value unless the client creates the Initial packet after receiving a Retry packet, where the Destination Connection ID is selected by the server.¶

The value of initial_salt is a 20 byte sequence shown in the figure in hexadecimal notation. Future versions of QUIC SHOULD generate a new salt value, thus ensuring that the keys are different for each version of QUIC. This prevents a middlebox that only recognizes one version of QUIC from seeing or modifying the contents of packets from future versions.¶

The HKDF-Expand-Label function defined in TLS 1.3 MUST be used for Initial packets even where the TLS versions offered do not include TLS 1.3.¶

The secrets used for protecting Initial packets change when a server sends a Retry packet to use the connection ID value selected by the server. The secrets do not change when a client changes the Destination Connection ID it uses in response to an Initial packet from the server.¶

Note:

The Destination Connection ID is of arbitrary length, and it could be zero length if the server sends a Retry packet with a zero-length Source Connection ID field. In this case, the Initial keys provide no assurance to the client that the server received its packet; the client has to rely on the exchange that included the Retry packet for that property.

Appendix A contains test vectors for packet encryption.¶

The Authentication Encryption with Associated Data (AEAD) [AEAD] function used for QUIC packet protection is the AEAD that is negotiated for use with the TLS connection. For example, if TLS is using the TLS_AES_128_GCM_SHA256, the AEAD_AES_128_GCM function is used.¶

Packets are protected prior to applying header protection (Section 5.4). The unprotected packet header is part of the associated data (A). When removing packet protection, an endpoint first removes the header protection.¶

All QUIC packets other than Version Negotiation and Retry packets are protected with an AEAD algorithm [AEAD]. Prior to establishing a shared secret, packets are protected with AEAD_AES_128_GCM and a key derived from the Destination Connection ID in the client’s first Initial packet (see Section 5.2). This provides protection against off-path attackers and robustness against QUIC version unaware middleboxes, but not against on-path attackers.¶

QUIC can use any of the ciphersuites defined in [TLS13] with the exception of TLS_AES_128_CCM_8_SHA256. A ciphersuite MUST NOT be negotiated unless a header protection scheme is defined for the ciphersuite. This document defines a header protection scheme for all ciphersuites defined in [TLS13] aside from TLS_AES_128_CCM_8_SHA256. These ciphersuites have a 16-byte authentication tag and produce an output 16 bytes larger than their input.¶

Note:

An endpoint MUST NOT reject a ClientHello that offers a ciphersuite that it does not support, or it would be impossible to deploy a new ciphersuite. This also applies to TLS_AES_128_CCM_8_SHA256.

The key and IV for the packet are computed as described in Section 5.1. The nonce, N, is formed by combining the packet protection IV with the packet number. The 62 bits of the reconstructed QUIC packet number in network byte order are left-padded with zeros to the size of the IV. The exclusive OR of the padded packet number and the IV forms the AEAD nonce.¶

The associated data, A, for the AEAD is the contents of the QUIC header, starting from the flags byte in either the short or long header, up to and including the unprotected packet number.¶

The input plaintext, P, for the AEAD is the payload of the QUIC packet, as described in [QUIC-TRANSPORT].¶

The output ciphertext, C, of the AEAD is transmitted in place of P.¶

Some AEAD functions have limits for how many packets can be encrypted under the same key and IV (see for example [AEBounds]). This might be lower than the packet number limit. An endpoint MUST initiate a key update (Section 6) prior to exceeding any limit set for the AEAD that is in use.¶

Parts of QUIC packet headers, in particular the Packet Number field, are protected using a key that is derived separate to the packet protection key and IV. The key derived using the “quic hp” label is used to provide confidentiality protection for those fields that are not exposed to on-path elements.¶

This protection applies to the least-significant bits of the first byte, plus the Packet Number field. The four least-significant bits of the first byte are protected for packets with long headers; the five least significant bits of the first byte are protected for packets with short headers. For both header forms, this covers the reserved bits and the Packet Number Length field; the Key Phase bit is also protected for packets with a short header.¶

The same header protection key is used for the duration of the connection, with the value not changing after a key update (see Section 6). This allows header protection to be used to protect the key phase.¶

This process does not apply to Retry or Version Negotiation packets, which do not contain a protected payload or any of the fields that are protected by this process.¶

Once an endpoint successfully receives a packet with a given packet number, it MUST discard all packets in the same packet number space with higher packet numbers if they cannot be successfully unprotected with either the same key, or - if there is a key update - the next packet protection key (see Section 6). Similarly, a packet that appears to trigger a key update, but cannot be unprotected successfully MUST be discarded.¶

Failure to unprotect a packet does not necessarily indicate the existence of a protocol error in a peer or an attack. The truncated packet number encoding used in QUIC can cause packet numbers to be decoded incorrectly if they are delayed significantly.¶

If 0-RTT keys are available (see Section 4.6), the lack of replay protection means that restrictions on their use are necessary to avoid replay attacks on the protocol.¶

A client MUST only use 0-RTT keys to protect data that is idempotent. A client MAY wish to apply additional restrictions on what data it sends prior to the completion of the TLS handshake. A client otherwise treats 0-RTT keys as equivalent to 1-RTT keys, except that it MUST NOT send ACKs with 0-RTT keys.¶

A client that receives an indication that its 0-RTT data has been accepted by a server can send 0-RTT data until it receives all of the server’s handshake messages. A client SHOULD stop sending 0-RTT data if it receives an indication that 0-RTT data has been rejected.¶

A server MUST NOT use 0-RTT keys to protect packets; it uses 1-RTT keys to protect acknowledgements of 0-RTT packets. A client MUST NOT attempt to decrypt 0-RTT packets it receives and instead MUST discard them.¶

Once a client has installed 1-RTT keys, it MUST NOT send any more 0-RTT packets.¶

Note:

0-RTT data can be acknowledged by the server as it receives it, but any packets containing acknowledgments of 0-RTT data cannot have packet protection removed by the client until the TLS handshake is complete. The 1-RTT keys necessary to remove packet protection cannot be derived until the client receives all server handshake messages.

Due to reordering and loss, protected packets might be received by an endpoint before the final TLS handshake messages are received. A client will be unable to decrypt 1-RTT packets from the server, whereas a server will be able to decrypt 1-RTT packets from the client. Endpoints in either role MUST NOT decrypt 1-RTT packets from their peer prior to completing the handshake.¶

• The client is not authenticated, unless the server has chosen to use a pre-shared key and validated the client’s pre-shared key binder; see Section 4.2.11 of [TLS13].
• The client has not demonstrated liveness, unless a RETRY packet was used.
• Any received 0-RTT data that the server responds to might be due to a replay attack.

Therefore, the server’s use of 1-RTT keys MUST be limited to sending data before the handshake is complete. A server MUST NOT process incoming 1-RTT protected packets before the TLS handshake is complete. Because sending acknowledgments indicates that all frames in a packet have been processed, a server cannot send acknowledgments for 1-RTT packets until the TLS handshake is complete. Received packets protected with 1-RTT keys MAY be stored and later decrypted and used once the handshake is complete.¶

Note:

TLS implementations might provide all 1-RTT secrets prior to handshake completion. Even where QUIC implementations have 1-RTT read keys, those keys cannot be used prior to completing the handshake.

The requirement for the server to wait for the client Finished message creates a dependency on that message being delivered. A client can avoid the potential for head-of-line blocking that this implies by sending its 1-RTT packets coalesced with a handshake packet containing a copy of the CRYPTO frame that carries the Finished message, until one of the handshake packets is acknowledged. This enables immediate server processing for those packets.¶

A server could receive packets protected with 0-RTT keys prior to receiving a TLS ClientHello. The server MAY retain these packets for later decryption in anticipation of receiving a ClientHello.¶

Retry packets (see the Retry Packet section of [QUIC-TRANSPORT]) carry a Retry Integrity Tag that provides two properties: it allows discarding packets that have accidentally been corrupted by the network, and it diminishes off-path attackers’ ability to send valid Retry packets.¶

The secret key and the nonce are values derived by calling HKDF-Expand-Label using 0x8b0d37eb8535022ebc8d76a207d80df22646ec06dc809642c30a8baa2baaff4c as the secret, with labels being “quic key” and “quic iv” (Section 5.1).¶

ODCID Length:

The ODCID Len contains the length in bytes of the Original Destination Connection ID field that follows it, encoded as an 8-bit unsigned integer.

Original Destination Connection ID:

The Original Destination Connection ID contains the value of the Destination Connection ID from the Initial packet that this Retry is in response to. The length of this field is given in ODCID Len. The presence of this field mitigates an off-path attacker’s ability to inject a Retry packet.

Endpoints maintain separate read and write secrets for packet protection. An endpoint initiates a key update by updating its packet protection write secret and using that to protect new packets. The endpoint creates a new write secret from the existing write secret as performed in Section 7.2 of [TLS13]. This uses the KDF function provided by TLS with a label of “quic ku”. The corresponding key and IV are created from that secret as defined in Section 5.1. The header protection key is not updated.¶

The endpoint toggles the value of the Key Phase bit and uses the updated key and IV to protect all subsequent packets.¶

An endpoint MUST NOT initiate a key update prior to having confirmed the handshake (Section 4.1.2). An endpoint MUST NOT initiate a subsequent key update prior unless it has received an acknowledgment for a packet that was sent protected with keys from the current key phase. This ensures that keys are available to both peers before another key update can be initiated. This can be implemented by tracking the lowest packet number sent with each key phase, and the highest acknowledged packet number in the 1-RTT space: once the latter is higher than or equal to the former, another key update can be initiated.¶

Note:

Keys of packets other than the 1-RTT packets are never updated; their keys are derived solely from the TLS handshake state.

The endpoint that initiates a key update also updates the keys that it uses for receiving packets. These keys will be needed to process packets the peer sends after updating.¶

An endpoint SHOULD retain old keys so that packets sent by its peer prior to receiving the key update can be processed. Discarding old keys too early can cause delayed packets to be discarded. Discarding packets will be interpreted as packet loss by the peer and could adversely affect performance.¶

A peer is permitted to initiate a key update after receiving an acknowledgement of a packet in the current key phase. An endpoint detects a key update when processing a packet with a key phase that differs from the value last used to protect the last packet it sent. To process this packet, the endpoint uses the next packet protection key and IV. See Section 6.3 for considerations about generating these keys.¶

If a packet is successfully processed using the next key and IV, then the peer has initiated a key update. The endpoint MUST update its send keys to the corresponding key phase in response, as described in Section 6.1. Sending keys MUST be updated before sending an acknowledgement for the packet that was received with updated keys. By acknowledging the packet that triggered the key update in a packet protected with the updated keys, the endpoint signals that the key update is complete.¶

An endpoint can defer sending the packet or acknowledgement according to its normal packet sending behaviour; it is not necessary to immediately generate a packet in response to a key update. The next packet sent by the endpoint will use the updated keys. The next packet that contains an acknowledgement will cause the key update to be completed. If an endpoint detects a second update before it has sent any packets with updated keys containing an acknowledgement for the packet that initiated the key update, it indicates that its peer has updated keys twice without awaiting confirmation. An endpoint MAY treat consecutive key updates as a connection error of type KEY_UPDATE_ERROR.¶

An endpoint that receives an acknowledgement that is carried in a packet protected with old keys where any acknowledged packet was protected with newer keys MAY treat that as a connection error of type KEY_UPDATE_ERROR. This indicates that a peer has received and acknowledged a packet that initiates a key update, but has not updated keys in response.¶

Endpoints responding to an apparent key update MUST NOT generate a timing side-channel signal that might indicate that the Key Phase bit was invalid (see Section 9.4). Endpoints can use dummy packet protection keys in place of discarded keys when key updates are not yet permitted. Using dummy keys will generate no variation in the timing signal produced by attempting to remove packet protection, and results in all packets with an invalid Key Phase bit being rejected.¶

The process of creating new packet protection keys for receiving packets could reveal that a key update has occurred. An endpoint MAY perform this process as part of packet processing, but this creates a timing signal that can be used by an attacker to learn when key updates happen and thus the value of the Key Phase bit in certain packets. Endpoints MAY instead defer the creation of the next set of receive packet protection keys until some time after a key update completes, up to three times the PTO; see Section 6.5.¶

Once generated, the next set of packet protection keys SHOULD be retained, even if the packet that was received was subsequently discarded. Packets containing apparent key updates are easy to forge and - while the process of key update does not require significant effort - triggering this process could be used by an attacker for DoS.¶

For this reason, endpoints MUST be able to retain two sets of packet protection keys for receiving packets: the current and the next. Retaining the previous keys in addition to these might improve performance, but this is not essential.¶

An endpoint always sends packets that are protected with the newest keys. Keys used for packet protection can be discarded immediately after switching to newer keys.¶

Packets with higher packet numbers MUST be protected with either the same or newer packet protection keys than packets with lower packet numbers. An endpoint that successfully removes protection with old keys when newer keys were used for packets with lower packet numbers MUST treat this as a connection error of type KEY_UPDATE_ERROR.¶

For receiving packets during a key update, packets protected with older keys might arrive if they were delayed by the network. Retaining old packet protection keys allows these packets to be successfully processed.¶

As packets protected with keys from the next key phase use the same Key Phase value as those protected with keys from the previous key phase, it can be necessary to distinguish between the two. This can be done using packet numbers. A recovered packet number that is lower than any packet number from the current key phase uses the previous packet protection keys; a recovered packet number that is higher than any packet number from the current key phase requires the use of the next packet protection keys.¶

Some care is necessary to ensure that any process for selecting between previous, current, and next packet protection keys does not expose a timing side channel that might reveal which keys were used to remove packet protection. See Section 9.5 for more information.¶

Alternatively, endpoints can retain only two sets of packet protection keys, swapping previous for next after enough time has passed to allow for reordering in the network. In this case, the Key Phase bit alone can be used to select keys.¶

An endpoint MAY allow a period of approximately the Probe Timeout (PTO; see [QUIC-RECOVERY]) after a key update before it creates the next set of packet protection keys. These updated keys MAY replace the previous keys at that time. With the caveat that PTO is a subjective measure - that is, a peer could have a different view of the RTT - this time is expected to be long enough that any reordered packets would be declared lost by a peer even if they were acknowledged and short enough to allow for subsequent key updates.¶

Endpoints need to allow for the possibility that a peer might not be able to decrypt packets that initiate a key update during the period when it retains old keys. Endpoints SHOULD wait three times the PTO before initiating a key update after receiving an acknowledgment that confirms that the previous key update was received. Failing to allow sufficient time could lead to packets being discarded.¶

An endpoint SHOULD retain old read keys for no more than three times the PTO. After this period, old read keys and their corresponding secrets SHOULD be discarded.¶

Key updates MUST be initiated before usage limits on packet protection keys are exceeded. For the cipher suites mentioned in this document, the limits in Section 5.5 of [TLS13] apply. [TLS13] does not specify a limit for AEAD_AES_128_CCM, but the analysis in Appendix B shows that a limit of 2^23 packets can be used to obtain the same confidentiality protection as the limits specified in TLS.¶

The usage limits defined in TLS 1.3 exist for protection against attacks on confidentiality and apply to successful applications of AEAD protection. The integrity protections in authenticated encryption also depend on limiting the number of attempts to forge packets. TLS achieves this by closing connections after any record fails an authentication check. In comparison, QUIC ignores any packet that cannot be authenticated, allowing multiple forgery attempts.¶

Endpoints MUST count the number of received packets that fail authentication for each set of keys. If the number of packets that fail authentication with the same key exceeds a limit that is specific to the AEAD in use, the endpoint MUST stop using those keys. Endpoints MUST initiate a key update before reaching this limit. If a key update is not possible, the endpoint MUST immediately close the connection. Applying a limit reduces the probability that an attacker is able to successfully forge a packet; see [AEBounds] and [ROBUST].¶

Note:

Due to the way that header protection protects the Key Phase, packets that are discarded are likely to have an even distribution of both Key Phase values. This means that packets that fail authentication will often use the packet protection keys from the next key phase. It is therefore necessary to also track the number of packets that fail authentication with the next set of packet protection keys. To avoid exhaustion of both sets of keys, it might be necessary to initiate two key updates in succession.

For AEAD_AES_128_GCM, AEAD_AES_256_GCM, and AEAD_CHACHA20_POLY1305, the limit on the number of packets that fail authentication is 2^36. Note that the analysis in [AEBounds] supports a higher limit for the AEAD_AES_128_GCM and AEAD_AES_256_GCM, but this specification recommends a lower limit. For AEAD_AES_128_CCM, the limit on the number of packets that fail authentication is 2^23.5; see Appendix B.¶

Note:

These limits were originally calculated using assumptions about the limits on TLS record size. The maximum size of a TLS record is 2^14 bytes. In comparison, QUIC packets can be up to 2^16 bytes. However, it is expected that QUIC packets will generally be smaller than TLS records. Where packets might be larger than 2^14 bytes in length, smaller limits might be needed.

Any TLS cipher suite that is specified for use with QUIC MUST define limits on the use of the associated AEAD function that preserves margins for confidentiality and integrity. That is, limits MUST be specified for the number of packets that can be authenticated and for the number packets that can fail authentication. Providing a reference to any analysis upon which values are based - and any assumptions used in that analysis - allows limits to be adapted to varying usage conditions.¶

The KEY_UPDATE_ERROR error code (0xE) is used to signal errors related to key updates.¶

QUIC requires that the cryptographic handshake provide authenticated protocol negotiation. TLS uses Application Layer Protocol Negotiation (ALPN) [ALPN] to select an application protocol. Unless another mechanism is used for agreeing on an application protocol, endpoints MUST use ALPN for this purpose.¶

When using ALPN, endpoints MUST immediately close a connection (see Section 10.3 of [QUIC-TRANSPORT]) with a no_application_protocol TLS alert (QUIC error code 0x178; see Section 4.10) if an application protocol is not negotiated. While [ALPN] only specifies that servers use this alert, QUIC clients MUST use error 0x178 to terminate a connection when ALPN negotiation fails.¶

An application protocol MAY restrict the QUIC versions that it can operate over. Servers MUST select an application protocol compatible with the QUIC version that the client has selected. The server MUST treat the inability to select a compatible application protocol as a connection error of type 0x178 (no_application_protocol). Similarly, a client MUST treat the selection of an incompatible application protocol by a server as a connection error of type 0x178.¶

QUIC transport parameters are carried in a TLS extension. Different versions of QUIC might define a different method for negotiating transport configuration.¶

Including transport parameters in the TLS handshake provides integrity protection for these values.¶

The extension_data field of the quic_transport_parameters extension contains a value that is defined by the version of QUIC that is in use.¶

The quic_transport_parameters extension is carried in the ClientHello and the EncryptedExtensions messages during the handshake. Endpoints MUST send the quic_transport_parameters extension; endpoints that receive ClientHello or EncryptedExtensions messages without the quic_transport_parameters extension MUST close the connection with an error of type 0x16d (equivalent to a fatal TLS missing_extension alert, see Section 4.10).¶

While the transport parameters are technically available prior to the completion of the handshake, they cannot be fully trusted until the handshake completes, and reliance on them should be minimized. However, any tampering with the parameters will cause the handshake to fail.¶

Endpoints MUST NOT send this extension in a TLS connection that does not use QUIC (such as the use of TLS with TCP defined in [TLS13]). A fatal unsupported_extension alert MUST be sent by an implementation that supports this extension if the extension is received when the transport is not QUIC.¶

The TLS EndOfEarlyData message is not used with QUIC. QUIC does not rely on this message to mark the end of 0-RTT data or to signal the change to Handshake keys.¶

Clients MUST NOT send the EndOfEarlyData message. A server MUST treat receipt of a CRYPTO frame in a 0-RTT packet as a connection error of type PROTOCOL_VIOLATION.¶

As a result, EndOfEarlyData does not appear in the TLS handshake transcript.¶

Appendix D.4 of [TLS13] describes an alteration to the TLS 1.3 handshake as a workaround for bugs in some middleboxes. The TLS 1.3 middlebox compatibility mode involves setting the legacy_session_id field to a 32-byte value in the ClientHello and ServerHello, then sending a change_cipher_spec record. Both field and record carry no semantic content and are ignored.¶

This mode has no use in QUIC as it only applies to middleboxes that interfere with TLS over TCP. QUIC also provides no means to carry a change_cipher_spec record. A client MUST NOT request the use of the TLS 1.3 compatibility mode. A server SHOULD treat the receipt of a TLS ClientHello with a non-empty legacy_session_id field as a connection error of type PROTOCOL_VIOLATION.¶

Use of TLS session tickets allows servers and possibly other entities to correlate connections made by the same client; see Section 4.5 for details.¶

As described in Section 8 of [TLS13], use of TLS early data comes with an exposure to replay attack. The use of 0-RTT in QUIC is similarly vulnerable to replay attack.¶

Endpoints MUST implement and use the replay protections described in [TLS13], however it is recognized that these protections are imperfect. Therefore, additional consideration of the risk of replay is needed.¶

QUIC is not vulnerable to replay attack, except via the application protocol information it might carry. The management of QUIC protocol state based on the frame types defined in [QUIC-TRANSPORT] is not vulnerable to replay. Processing of QUIC frames is idempotent and cannot result in invalid connection states if frames are replayed, reordered or lost. QUIC connections do not produce effects that last beyond the lifetime of the connection, except for those produced by the application protocol that QUIC serves.¶

Note:

TLS session tickets and address validation tokens are used to carry QUIC configuration information between connections. These MUST NOT be used to carry application semantics. The potential for reuse of these tokens means that they require stronger protections against replay.

A server that accepts 0-RTT on a connection incurs a higher cost than accepting a connection without 0-RTT. This includes higher processing and computation costs. Servers need to consider the probability of replay and all associated costs when accepting 0-RTT.¶

Ultimately, the responsibility for managing the risks of replay attacks with 0-RTT lies with an application protocol. An application protocol that uses QUIC MUST describe how the protocol uses 0-RTT and the measures that are employed to protect against replay attack. An analysis of replay risk needs to consider all QUIC protocol features that carry application semantics.¶

Disabling 0-RTT entirely is the most effective defense against replay attack.¶

QUIC extensions MUST describe how replay attacks affect their operation, or prohibit their use in 0-RTT. Application protocols MUST either prohibit the use of extensions that carry application semantics in 0-RTT or provide replay mitigation strategies.¶

A small ClientHello that results in a large block of handshake messages from a server can be used in packet reflection attacks to amplify the traffic generated by an attacker.¶

QUIC includes three defenses against this attack. First, the packet containing a ClientHello MUST be padded to a minimum size. Second, if responding to an unverified source address, the server is forbidden to send more than three UDP datagrams in its first flight (see Section 8.1 of [QUIC-TRANSPORT]). Finally, because acknowledgements of Handshake packets are authenticated, a blind attacker cannot forge them. Put together, these defenses limit the level of amplification.¶

The header protection variants in this document use a pseudorandom permutation (PRP) in place of a generic PRF. However, since all PRPs are also PRFs [IMC], these variants do not deviate from the HN1 construction.¶

As hp_key is distinct from the packet protection key, it follows that header protection achieves AE2 security as defined in [NAN] and therefore guarantees privacy of field, the protected packet header. Future header protection variants based on this construction MUST use a PRF to ensure equivalent security guarantees.¶

Use of the same key and ciphertext sample more than once risks compromising header protection. Protecting two different headers with the same key and ciphertext sample reveals the exclusive OR of the protected fields. Assuming that the AEAD acts as a PRF, if L bits are sampled, the odds of two ciphertext samples being identical approach 2^(-L/2), that is, the birthday bound. For the algorithms described in this document, that probability is one in 2^64.¶

To prevent an attacker from modifying packet headers, the header is transitively authenticated using packet protection; the entire packet header is part of the authenticated additional data. Protected fields that are falsified or modified can only be detected once the packet protection is removed.¶

An attacker could guess values for packet numbers or Key Phase and have an endpoint confirm guesses through timing side channels. Similarly, guesses for the packet number length can be trialed and exposed. If the recipient of a packet discards packets with duplicate packet numbers without attempting to remove packet protection they could reveal through timing side-channels that the packet number matches a received packet. For authentication to be free from side-channels, the entire process of header protection removal, packet number recovery, and packet protection removal MUST be applied together without timing and other side-channels.¶

For the sending of packets, construction and protection of packet payloads and packet numbers MUST be free from side-channels that would reveal the packet number or its encoded size.¶

During a key update, the time taken to generate new keys could reveal through timing side-channels that a key update has occurred. Alternatively, where an attacker injects packets this side-channel could reveal the value of the Key Phase on injected packets. After receiving a key update, an endpoint SHOULD generate and save the next set of receive packet protection keys, as described in Section 6.3. By generating new keys before a key update is received, receipt of packets will not create timing signals that leak the value of the Key Phase.¶

This depends on not doing this key generation during packet processing and it can require that endpoints maintain three sets of packet protection keys for receiving: for the previous key phase, for the current key phase, and for the next key phase. Endpoints can instead choose to defer generation of the next receive packet protection keys until they discard old keys so that only two sets of receive keys need to be retained at any point in time.¶

In using TLS, the central key schedule of TLS is used. As a result of the TLS handshake messages being integrated into the calculation of secrets, the inclusion of the QUIC transport parameters extension ensures that handshake and 1-RTT keys are not the same as those that might be produced by a server running TLS over TCP. To avoid the possibility of cross-protocol key synchronization, additional measures are provided to improve key separation.¶

The QUIC packet protection keys and IVs are derived using a different label than the equivalent keys in TLS.¶

To preserve this separation, a new version of QUIC SHOULD define new labels for key derivation for packet protection key and IV, plus the header protection keys. This version of QUIC uses the string “quic”. Other versions can use a version-specific label in place of that string.¶

The initial secrets use a key that is specific to the negotiated QUIC version. New QUIC versions SHOULD define a new salt value used in calculating initial secrets.¶

client in:

00200f746c73313320636c69656e7420696e00

server in:

00200f746c7331332073657276657220696e00

quic key:

00100e746c7331332071756963206b657900

quic iv:

000c0d746c733133207175696320697600

quic hp:

00100d746c733133207175696320687000

This example shows some of the steps required to protect a packet with a short header. This example uses AEAD_CHACHA20_POLY1305.¶

In this example, TLS produces an application write secret from which a server uses HKDF-Expand-Label to produce four values: a key, an IV, a header protection key, and the secret that will be used after keys are updated (this last value is not used further in this example).¶

The following shows the steps involved in protecting a minimal packet with an empty Destination Connection ID. This packet contains a single PING frame (that is, a payload of just 0x01) and has a packet number of 654360564. In this example, using a packet number of length 3 (that is, 49140 is encoded) avoids having to pad the payload of the packet; PADDING frames would be needed if the packet number is encoded on fewer octets.¶

The resulting ciphertext is the minimum size possible. One byte is skipped to produce the sample for header protection.¶

The protected packet is the smallest possible packet size of 21 bytes.¶

That is, endpoints cannot protect more than 2^23 packets with the same set of keys without causing an attacker to gain an larger advantage than the target of 2^-60.¶

Using the previously-established value of 2^23 for q and rounding, this leads to an upper limit on v of 2^23.5. That is, endpoints cannot attempt to authenticate more than 2^23.5 packets with the same set of keys without causing an attacker to gain an larger advantage than the target of 2^-57.¶

• Defined limits on the number of packets that can be protected with a single key and limits on the number of packets that can fail authentication (#3619, #3620)

• Allowed CONNECTION_CLOSE in any packet number space, with restrictions on use of the application-specific variant (#3430, #3435, #3440)
• Prohibit the use of the compatibility mode from TLS 1.3 (#3594, #3595)

• No changes

• No changes

• Rewrite key updates (#3050)
  • Allow but don’t recommend deferring key updates (#2792, #3263)
  • More completely define received behavior (#2791)
  • Define the label used with HKDF-Expand-Label (#3054)

• Key update text update (#3050):
  • Recommend constant-time key replacement (#2792)
  • Provide explicit labels for key update key derivation (#3054)
• Allow first Initial from a client to span multiple packets (#2928, #3045)
• PING can be sent at any encryption level (#3034, #3035)

• Update the salt used for Initial secrets (#2887, #2980)

• No changes

• Mandate the use of the QUIC transport parameters extension (#2528, #2560)
• Define handshake completion and confirmation; define clearer rules when it encryption keys should be discarded (#2214, #2267, #2673)

• Increased the set of permissible frames in 0-RTT (#2344, #2355)
• Transport parameter extension is mandatory (#2528, #2560)

• Endpoints discard initial keys as soon as handshake keys are available (#1951, #2045)
• Use of ALPN or equivalent is mandatory (#2263, #2284)

• Update the salt used for Initial secrets (#1970)
• Clarify that TLS_AES_128_CCM_8_SHA256 isn’t supported (#2019)
• Change header protection
  • Sample from a fixed offset (#1575, #2030)
  • Cover part of the first byte, including the key phase (#1322, #2006)
• TLS provides an AEAD and KDF function (#2046)
  • Clarify that the TLS KDF is used with TLS (#1997)
  • Change the labels for calculation of QUIC keys (#1845, #1971, #1991)
• Initial keys are discarded once Handshake keys are available (#1951, #2045)

• Updated to TLS 1.3 final (#1660)

• Changes to integration of the TLS handshake (#829, #1018, #1094, #1165, #1190, #1233, #1242, #1252, #1450)
  • The cryptographic handshake uses CRYPTO frames, not stream 0
  • QUIC packet protection is used in place of TLS record protection
  • Separate QUIC packet number spaces are used for the handshake
  • Changed Retry to be independent of the cryptographic handshake
  • Limit the use of HelloRetryRequest to address TLS needs (like key shares)
• Changed codepoint of TLS extension (#1395, #1402)

• Encrypted packet numbers.

• No significant changes.

• Cleaned up key schedule and updated the salt used for handshake packet protection (#1077)

• Specify value for max_early_data_size to enable 0-RTT (#942)
• Update key derivation function (#1003, #1004)

• Handshake errors can be reported with CONNECTION_CLOSE (#608, #891)

No significant changes.¶

• Update labels used in HKDF-Expand-Label to match TLS 1.3 (#642)

No significant changes.¶

• Updates to match changes in transport draft

• Use TLS alerts to signal TLS errors (#272, #374)
• Require ClientHello to fit in a single packet (#338)
• The second client handshake flight is now sent in the clear (#262, #337)
• The QUIC header is included as AEAD Associated Data (#226, #243, #302)
• Add interface necessary for client address validation (#275)
• Define peer authentication (#140)
• Require at least TLS 1.3 (#138)
• Define transport parameters as a TLS extension (#122)
• Define handling for protected packets before the handshake completes (#39)
• Decouple QUIC version and ALPN (#12)

• Changed bit used to signal key phase
• Updated key phase markings during the handshake
• Added TLS interface requirements section
• Moved to use of TLS exporters for key derivation
• Moved TLS error code definitions into this document

• Adopted as base for draft-ietf-quic-tls
• Updated authors/editors list
• Added status note