| Internet-Draft | QUIC Manageability | January 2021 |
| Kuehlewind & Trammell | Expires 26 July 2021 | [Page] |
This document discusses manageability of the QUIC transport protocol, focusing on caveats impacting network operations involving QUIC traffic. Its intended audience is network operators, as well as content providers that rely on the use of QUIC-aware middleboxes, e.g. for load balancing.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 26 July 2021.¶
Copyright (c) 2021 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.¶
QUIC [QUIC-TRANSPORT] is a new transport protocol encapsulated in UDP and encrypted by default. QUIC integrates TLS [QUIC-TLS] to encrypt all payload data and most control information. The design focused on support of semantics for HTTP, which required changes to HTTP known as HTTP/3 [QUIC-HTTP].¶
Given that QUIC is an end-to-end transport protocol, all information in the protocol header, even that which can be inspected, is not meant to be mutable by the network, and is therefore integrity-protected. While less information is visible to the network than for TCP, integrity protection can also simplify troubleshooting, because none of the nodes on the network path can modify the transport layer information.¶
This document provides guidance for network operations that manage QUIC traffic. This includes guidance on how to interpret and utilize information that is exposed by QUIC to the network, requirements and assumptions that the QUIC design with respect to network treatment, and a description of how common network management practices will be impacted by QUIC.¶
Since QUIC's wire image [WIRE-IMAGE] is integrity protected and not modifiable on path, in-network operations are not possible without terminating the QUIC connection, for instance using a back-to-back proxy. Proxy operations are not in scope for this document. A proxy can either explicit identify itself as providing a proxy service, or may share the TLS credentials to authenticate as the server and (in some cases) client acting as a front-facing instance for the endpoint itself.¶
Network management is not a one-size-fits-all endeavour: practices considered necessary or even mandatory within enterprise networks with certain compliance requirements, for example, would be impermissible on other networks without those requirements. This document therefore does not make any specific recommendations as to which practices should or should not be applied; for each practice, it describes what is and is not possible with the QUIC transport protocol as defined.¶
In this section, we discuss those aspects of the QUIC transport protocol that have an impact on the design and operation of devices that forward QUIC packets. Here, we are concerned primarily with the unencrypted part of QUIC's wire image [WIRE-IMAGE], which we define as the information available in the packet header in each QUIC packet, and the dynamics of that information. Since QUIC is a versioned protocol, the wire image of the header format can also change from version to version. However, the field that identifies the QUIC version in some packets, and the format of the Version Negotiation Packet, are both inspectable and invariant [QUIC-INVARIANTS].¶
This document describes version 1 of the QUIC protocol, whose wire image is fully defined in [QUIC-TRANSPORT] and [QUIC-TLS]. Features of the wire image described herein may change in future versions of the protocol, except when specified as an invariant [QUIC-INVARIANTS], and cannot be used to identify QUIC as a protocol or to infer the behavior of future versions of QUIC.¶
Section 9.1 provides non-normative guidance on the identification of QUIC version 1 packets compared to some pre-standard versions.¶
QUIC packets may have either a long header, or a short header. The first bit of the QUIC header is the Header Form bit, and indicates which type of header is present. The purpose of this bit is invariant across QUIC versions.¶
The long header exposes more information. It is used during connection establishment, including version negotiation, retry, and 0-RTT data. It contains a version number, as well as source and destination connection IDs for grouping packets belonging to the same flow. The definition and location of these fields in the QUIC long header are invariant for future versions of QUIC, although future versions of QUIC may provide additional fields in the long header [QUIC-INVARIANTS].¶
Short headers are used after connection establishment, and contain only an optional destination connection ID and the spin bit for RTT measurement.¶
The following information is exposed in QUIC packet headers:¶
Retry (Section 17.2.5 of [QUIC-TRANSPORT]) and Version Negotiation (Section 17.2.1 of [QUIC-TRANSPORT]) packets are not encrypted or obfuscated in any way. For other kinds of packets, other information in the packet headers is cryptographically obfuscated:¶
Multiple QUIC packets may be coalesced into a UDP datagram, with a datagram carrying one or more long header packets followed by zero or one short header packets. When packets are coalesced, the Length fields in the long headers are used to separate QUIC packets; see Section 12.2 of [QUIC-TRANSPORT]. The length header field is variable length, and its position in the header is also variable depending on the length of the source and destination connection ID; see Section 17.2 of [QUIC-TRANSPORT].¶
Applications that have a mapping for TCP as well as QUIC are expected to use the same port number for both services. However, as with TCP-based services, especially when application layer information is encrypted, there is no guarantee that a specific application will use the registered port, or the used port is carrying traffic belonging to the respective registered service. For example, [QUIC-HTTP] specifies the use of Alt-Svc for discovery of HTTP/3 services on other ports.¶
Further, as QUIC has a connection ID, it is also possible to maintain multiple QUIC connections over one 5-tuple. However, if the connection ID is not present in the packet header, all packets of the 5-tuple belong to the same QUIC connection.¶
New QUIC connections are established using a handshake, which is distinguishable on the wire and contains some information that can be passively observed.¶
To illustrate the information visible in the QUIC wire image during the handshake, we first show the general communication pattern visible in the UDP datagrams containing the QUIC handshake, then examine each of the datagrams in detail.¶
In the nominal case, the QUIC handshake can be recognized on the wire through at least four datagrams we'll call "QUIC Client Hello", "QUIC Server Hello", and "Initial Completion", and "Handshake Completion", for purposes of this illustration, as shown in Figure 1.¶
Packets in the handshake belong to three separate cryptographic and transport contexts ("Initial", which contains observable payload, and "Handshake" and "1-RTT", which do not). QUIC packets in separate contexts during the handshake are generally coalesced (see Section 2.2) in order to reduce the number of UDP datagrams sent during the handshake.¶
As shown here, the client can send 0-RTT data as soon as it has sent its Client Hello, and the server can send 1-RTT data as soon as it has sent its Server Hello.¶
Client Server | | +----QUIC Client Hello-------------------->| +----(zero or more 0RTT)------------------>| | | |<--------------------QUIC Server Hello----+ |<---------(1RTT encrypted data starts)----+ | | +----Initial Completion------------------->| +----(1RTT encrypted data starts)--------->| | | |<-----------------Handshake Completion----+ | |
A typical handshake starts with the client sending of a QUIC Client Hello datagram as shown in Figure 2, which elicits a QUIC Server Hello datagram as shown in Figure 3 typically containing three packets: an Initial packet with the Server Hello, a Handshake packet with the rest of the server's side of the TLS handshake, and initial 1-RTT data, if present.¶
The Initial Completion datagram contains at least one Handshake packet and some also include an Initial packet.¶
Datagrams that contain a QUIC Initial Packet (Client Hello, Server Hello, and some Initial Completion) must be at least 1200 octets long. This protects against amplification attacks and verifies that the network path meets minimum Maximum Transmission Unit (MTU) requirements. This is usually accomplished with either the addition of PADDING frames to the Initial packet, or coalescing of the Initial Packet with packets from other encryption contexts.¶
The content of QUIC Initial packets are encrypted using Initial Secrets, which are derived from a per-version constant and the client's destination connection ID; they are therefore observable by any on-path device that knows the per-version constant. We therefore consider these as visible in our illustration. The content of QUIC Handshake packets are encrypted using keys established during the initial handshake exchange, and are therefore not visible.¶
Initial, Handshake, and the Short Header packets transmitted after the handshake belong to cryptographic and transport contexts. The Initial Completion Figure 4 and the Handshake Completion Figure 5 datagrams finish these first two contexts, by sending the final acknowledgment and finishing the transmission of CRYPTO frames.¶
+----------------------------------------------------------+ | UDP header (source and destination UDP ports) | +----------------------------------------------------------+ | QUIC long header (type = Initial, Version, DCID, SCID) (Length) +----------------------------------------------------------+ | | QUIC CRYPTO frame header | | +----------------------------------------------------------+ | | TLS Client Hello (incl. TLS SNI) | | +----------------------------------------------------------+ | | QUIC PADDING frames | | +----------------------------------------------------------+<-+
The Client Hello datagram exposes version number, source and destination connection IDs in the clear. Information in the TLS Client Hello frame, including any TLS Server Name Indication (SNI) present, is obfuscated using the Initial secret. Note that the location of PADDING is implementation-dependent, and PADDING frames may not appear in a coalesced Initial packet.¶
+------------------------------------------------------------+ | UDP header (source and destination UDP ports) | +------------------------------------------------------------+ | QUIC long header (type = Initial, Version, DCID, SCID) (Length) +------------------------------------------------------------+ | | QUIC CRYPTO frame header | | +------------------------------------------------------------+ | | TLS Server Hello | | +------------------------------------------------------------+ | | QUIC ACK frame (acknowledging client hello) | | +------------------------------------------------------------+<-+ | QUIC long header (type = Handshake, Version, DCID, SCID) (Length) +------------------------------------------------------------+ | | encrypted payload (presumably CRYPTO frames) | | +------------------------------------------------------------+<-+ | QUIC short header | +------------------------------------------------------------+ | 1-RTT encrypted payload | +------------------------------------------------------------+
The Server Hello datagram also exposes version number, source and destination connection IDs and information in the TLS Server Hello message which is obfuscated using the Initial secret.¶
+------------------------------------------------------------+ | UDP header (source and destination UDP ports) | +------------------------------------------------------------+ | QUIC long header (type = Initial, Version, DCID, SCID) (Length) +------------------------------------------------------------+ | | QUIC ACK frame (acknowledging Server Hello Initial) | | +------------------------------------------------------------+<-+ | QUIC long header (type = Handshake, Version, DCID, SCID) (Length) +------------------------------------------------------------+ | | encrypted payload (presumably CRYPTO/ACK frames) | | +------------------------------------------------------------+<-+ | QUIC short header | +------------------------------------------------------------+ | 1-RTT encrypted payload | +------------------------------------------------------------+
The Initial Completion datagram does not expose any additional information; however, recognizing it can be used to determine that a handshake has completed (see Section 3.2), and for three-way handshake RTT estimation as in Section 3.7.¶
+------------------------------------------------------------+ | UDP header (source and destination UDP ports) | +------------------------------------------------------------+ | QUIC long header (type = Handshake, Version, DCID, SCID) (Length) +------------------------------------------------------------+ | | encrypted payload (presumably ACK frame) | | +------------------------------------------------------------+<-+ | QUIC short header | +------------------------------------------------------------+ | 1-RTT encrypted payload | +------------------------------------------------------------+
Similar to Initial Completion, Handshake Completion also exposes no additional information; observing it serves only to determine that the handshake has completed.¶
When the client uses 0-RTT connection resumption, 0-RTT data may also be seen in the QUIC Client Hello datagram, as shown in Figure 6.¶
+----------------------------------------------------------+ | UDP header (source and destination UDP ports) | +----------------------------------------------------------+ | QUIC long header (type = Initial, Version, DCID, SCID) (Length) +----------------------------------------------------------+ | | QUIC CRYPTO frame header | | +----------------------------------------------------------+ | | TLS Client Hello (incl. TLS SNI) | | +----------------------------------------------------------+<-+ | QUIC long header (type = 0RTT, Version, DCID, SCID) (Length) +----------------------------------------------------------+ | | 0-rtt encrypted payload | | +----------------------------------------------------------+<-+
In a 0-RTT QUIC Client Hello datagram, the PADDING frame is only present if necessary to increase the size of the datagram with 0RTT data to at least 1200 bytes. Additional datagrams containing only 0-RTT protected long header packets may be sent from the client to the server after the Client Hello datagram, containing the rest of the 0-RTT data. The amount of 0-RTT protected data is limited by the initial congestion window, typically around 10 packets [RFC6928].¶
As soon as the cryptographic context is established, all information in the QUIC header, including exposed information, is integrity protected. Further, information that was sent and exposed in handshake packets sent before the cryptographic context was established are validated later during the cryptographic handshake. Therefore, devices on path cannot alter any information or bits in QUIC packet headers, except specific parts of Initial packets, since alteration of header information will lead to a failed integrity check at the receiver, and can even lead to connection termination.¶
The connection ID in the QUIC packet headers allows routing of QUIC packets at load balancers on other than five-tuple information, ensuring that related flows are appropriately balanced together; and to allow rebinding of a connection after one of the endpoint's addresses changes - usually the client's. Client and server negotiate connection IDs during the handshake; typically, however, only the server will request a connection ID for the lifetime of the connection. Connection IDs for either endpoint may change during the lifetime of a connection, with the new connection ID being negotiated via encrypted frames. See Section 5.1 of [QUIC-TRANSPORT]. Therefore, observing a new connection ID does not necessary indicate a new connection.¶
Server-generated connection IDs should seek to obscure any encoding, of routing identities or any other information. Exposing the server mapping would allow linkage of multiple IP addresses to the same host if the server also supports migration. Furthermore, this opens an attack vector on specific servers or pools.¶
The best way to obscure an encoding is to appear random to observers, which is most rigorously achieved with encryption. Even when encrypted, a scheme could embed the unencrypted length of the connection ID in the connection ID itself, instead of remembering it.¶
[QUIC_LB] further specified possible algorithms to generate connection IDs at load balancers.¶
The packet number field is always present in the QUIC packet header; however, it is always encrypted. The encryption key for packet number protection on handshake packets sent before cryptographic context establishment is specific to the QUIC version, while packet number protection on subsequent packets uses secrets derived from the end-to-end cryptographic context. Packet numbers are therefore not part of the wire image that is visible to on-path observers.¶
Version Negotiation packets are not intrinsically protected, but QUIC versions can use later encrypted messages to verify that they were authentic. Therefore any manipulation of this list will be detected and may cause the endpoints to terminate the connection attempt.¶
Also note that the list of versions in the Version Negotiation packet may contain reserved versions. This mechanism is used to avoid ossification in the implementation on the selection mechanism. Further, a client may send a Initial Client packet with a reserved version number to trigger version negotiation. In the Version Negotiation packet the connection ID and packet number of the Client Initial packet are reflected to provide a proof of return-routability. Therefore changing this information will also cause the connection to fail.¶
QUIC is expected to evolve rapidly, so new versions, both experimental and IETF standard versions, will be deployed in the Internet more often than with traditional Internet- and transport-layer protocols. Using a particular version number to recognize valid QUIC traffic is likely to persistently miss a fraction of QUIC flows and completely fail in the near future, and is therefore not recommended. In addition, due to the speed of evolution of the protocol, devices that attempt to distinguish QUIC traffic from non-QUIC traffic for purposes of network admission control should admit all QUIC traffic regardless of version.¶
This section addresses the different kinds of observations and inferences that can be made about QUIC flows by a passive observer in the network based on the wire image in Section 2. Here we assume a bidirectional observer (one that can see packets in both directions in the sequence in which they are carried on the wire) unless noted.¶
The QUIC wire image is not specifically designed to be distinguishable from other UDP traffic.¶
The only application binding defined by the IETF QUIC WG is HTTP/3 [QUIC-HTTP] at the time of this writing; however, many other applications are currently being defined and deployed over QUIC, so an assumption that all QUIC traffic is HTTP/3 is not valid. HTTP over QUIC uses UDP port 443 by default, although URLs referring to resources available over HTTP/3 may specify alternate port numbers. Simple assumptions about whether a given flow is using QUIC based upon a UDP port number may therefore not hold; see also [RFC7605] section 5.¶
While the second most significant bit (0x40) of the first octet is set to 1 in most QUIC packets of the current version (see Section 2.1), this method of recognizing QUIC traffic is NOT RECOMMENDED. First, it only provides one bit of information and is quite prone to collide with UDP-based protocols other than those that this static bit is meant to allow multiplexing with. Second, this feature of the wire image is not invariant [QUIC-INVARIANTS] and may change in future versions of the protocol, or even be negotiated during the handshake via the use of transport parameters.¶
Even though transport parameters transmitted in the client initial are obserable by the network, they cannot be modified by the network without risking connection failure. Further, the negotiated reply from the server cannot be observed, so observers on the network cannot know which parameters are actually in use.¶
An in-network observer assuming that a set of packets belongs to a QUIC flow can infer the version number in use by observing the handshake: an Initial packet with a given version from a client to which a server responds with an Initial packet with the same version implies acceptance of that version.¶
Negotiated version cannot be identified for flows for which a handshake is not observed, such as in the case of connection migration; however, these flows can be associated with flows for which a version has been identified; see Section 3.4.¶
This document focuses on QUIC Version 1, and this section applies only to packets belonging to Version 1 QUIC flows; for purposes of on-path observation, it assumes that these packets have been identified as such through the observation of a version number exchange as described above.¶
A related question is whether a first packet of a given flow on a known QUIC-associated port is a valid QUIC packet, to support in-network filtering of garbage UDP packets (reflection attacks, random backscatter). While heuristics based on the first byte of the packet (packet type) could be used to separate valid from invalid first packet types, the deployment of such heuristics is not recommended, as packet types may have different meanings in future versions of the protocol.¶
Connection establishment uses Initial and Handshake packets containing a TLS handshake, and Retry packets that do not contain parts of the handshake. Connection establishment can therefore be detected using heuristics similar to those used to detect TLS over TCP. A client initiating a 0-RTT connection may also send data packets in 0-RTT Protected packets directly after the Initial packet containing the TLS Client Hello. Since these packets may be reordered in the network, 0-RTT Protected data packets could be seen before the Initial packet.¶
Note that clients send Initial packets before servers do, servers send Handshake packets before clients do, and only clients send Initial packets with tokens. Therefore, the role as a client or server can generally be confirmed by an on- path observer. An attempted connection after Retry can be detected by correlating the token on the Retry with the token on the subsequent Initial packet and the destination connection ID of the new Initial packet.¶
The cleartext TLS handshake may contain Server Name Indication (SNI) [RFC6066], by which the client reveals the name of the server it intends to connect to, in order to allow the server to present a certificate based on that name. It may also contain information from Application-Layer Protocol Negotiation (ALPN) [RFC7301], by which the client exposes the names of application-layer protocols it supports; an observer can deduce that one of those protocols will be used if the connection continues.¶
Work is currently underway in the TLS working group to encrypt the SNI in TLS 1.3 [TLS-ESNI]. This would make SNI-based application identification impossible through passive measurement for QUIC and other protocols that use TLS.¶
If the SNI is not encrypted it can be derived from the QUIC Initial packet by calculating the Initial Secret to decrypt the packet payload and parse the QUIC CRYPTO Frame containing the TLS ClientHello.¶
As both the initial salt for the Initial Secret as well as CRYPTO frame itself are version-specific, the first step is always to parse the version number (second to sixth byte of the long header). Note that only long header packets carry the version number, so it is necessary to also check the if first bit of the QUIC packet is set to 1, indicating a long header.¶
Note that proprietary QUIC versions, that have been deployed before standardization, might not set the first bit in a QUIC long header packets to 1. To parse these versions, example code is provided in the appendix (see Section 9.1), however, it is expected that these versions will gradually disappear over time.¶
When the version has been identified as QUIC version 1, the packet type needs to be verified as an Initial packet by checking that the third and fourth bit of the header are both set to 0. Then the client destination connection ID needs to be extracted to calculate the Initial Secret together with the version specific initial salt, as described in [QUIC-TLS]. The length of the connection ID is indicated in the 6th byte of the header followed by the connection ID itself.¶
To determine the end of the header and find the start of the payload, the packet number length, the source connection ID length, and the token length need to be extracted. The packet number length is defined by the seventh and eight bits of the header as described in section 17.2. of [QUIC-TRANSPORT], but is obfuscated as described in [QUIC-TLS]. The source connection ID length is specified in the byte after the destination connection ID. And the token length, which follows the source connection ID, is a variable length integer as specified in Section 16 of [QUIC-TRANSPORT].¶
After decryption, the Initial Client packet can be parsed to detect the CRYPTO frame that contains the TLS Client Hello, which then can be parsed similarly to TLS over TCP connections. The Initial client packet may contain other frames, so the first bytes of each frame need to be checked to identify the frame type, and if needed skip over it. Note that the length of the frames is dependent on the frame type. In QUIC version 1, the packet is expected to only carry the CRYPTO frame and optionally padding frames. However, PADDING frames, which are each one byte of zeros, may also occur before or after the CRYPTO frame.¶
Note that client Initial packets after the first do not always use the destination connection ID that was used to generate the Initial keys. Therefore, attempts to decrypt these packets using the procedure above might fail.¶
The QUIC connection ID (see Section 2.6) is designed to allow an on-path device such as a load-balancer to associate two flows as identified by five-tuple when the address and port of one of the endpoints changes; e.g. due to NAT rebinding or server IP address migration. An observer keeping flow state can associate a connection ID with a given flow, and can associate a known flow with a new flow when when observing a packet sharing a connection ID and one endpoint address (IP address and port) with the known flow.¶
However, since the connection ID may change multiple times during the lifetime of a flow, and the negotiation of connection ID changes is encrypted, packets with the same 5-tuple but different connection IDs may or may not belong to the same connection.¶
The connection ID value should be treated as opaque; see Section 4.3 for caveats regarding connection ID selection at servers.¶
QUIC does not expose the end of a connection; the only indication to on-path devices that a flow has ended is that packets are no longer observed. Stateful devices on path such as NATs and firewalls must therefore use idle timeouts to determine when to drop state for QUIC flows, see further section Section 4.1.¶
QUIC explicitly exposes which side of a connection is a client and which side is a server during the handshake. In addition, the symmetry of a flow (whether primarily client-to-server, primarily server-to-client, or roughly bidirectional, as input to basic traffic classification techniques) can be inferred through the measurement of data rate in each direction. While QUIC traffic is protected and ACKs may be padded, padding is not required.¶
Round-trip time of QUIC flows can be inferred by observation once per flow, during the handshake, as in passive TCP measurement; this requires parsing of the QUIC packet header and recognition of the handshake, as illustrated in Section 2.4. It can also be inferred during the flow's lifetime, if the endpoints use the spin bit facility described below and in [QUIC-TRANSPORT], section 17.3.1.¶
In the common case, the delay between the Initial packet containing the TLS Client Hello and the Handshake packet containing the TLS Server Hello represents the RTT component on the path between the observer and the server. The delay between the TLS Server Hello and the Handshake packet containing the TLS Finished message sent by the client represents the RTT component on the path between the observer and the client. While the client may send 0-RTT Protected packets after the Initial packet during 0-RTT connection re-establishment, these can be ignored for RTT measurement purposes.¶
Handshake RTT can be measured by adding the client-to-observer and observer-to-server RTT components together. This measurement necessarily includes any transport and application layer delay (the latter mainly caused by the asymmetric crypto operations associated with the TLS handshake) at both sides.¶
The spin bit provides an additional method to measure per-flow RTT from observation points on the network path throughout the duration of a connection. Endpoint participation in spin bit signaling is optional in QUIC. That is, while its location is fixed in this version of QUIC, an endpoint can unilaterally choose to not support "spinning" the bit. Use of the spin bit for RTT measurement by devices on path is only possible when both endpoints enable it. Some endpoints may disable use of the spin bit by default, others only in specific deployment scenarios, e.g. for servers and clients where the RTT would reveal the presence of a VPN or proxy. To avoid making these connections identifiable based on the usage of the spin bit, all endpoints randomly disable "spinning" for at least one eighth of connections, even if otherwise enabled by default. An endpoint not participating in spin bit signaling for a given connection can use a fixed spin value for the duration of the connection, or can set the bit randomly on each packet sent.¶
When in use and a QUIC flow sends data continuously, the latency spin bit in each direction changes value once per round-trip time (RTT). An on-path observer can observe the time difference between edges (changes from 1 to 0 or 0 to 1) in the spin bit signal in a single direction to measure one sample of end-to-end RTT.¶
Note that this measurement, as with passive RTT measurement for TCP, includes any transport protocol delay (e.g., delayed sending of acknowledgements) and/or application layer delay (e.g., waiting for a response to be generated). It therefore provides devices on path a good instantaneous estimate of the RTT as experienced by the application. A simple linear smoothing or moving minimum filter can be applied to the stream of RTT information to get a more stable estimate.¶
However, application-limited and flow-control-limited senders can have application and transport layer delay, respectively, that are much greater than network RTT. When the sender is application-limited and e.g. only sends small amount of periodic application traffic, where that period is longer than the RTT, measuring the spin bit provides information about the application period, not the network RTT.¶
Since the spin bit logic at each endpoint considers only samples from packets that advance the largest packet number, signal generation itself is resistant to reordering. However, reordering can cause problems at an observer by causing spurious edge detection and therefore inaccurate (i.e., lower) RTT estimates, if reordering occurs across a spin-bit flip in the stream.¶
Simple heuristics based on the observed data rate per flow or changes in the RTT series can be used to reject bad RTT samples due to lost or reordered edges in the spin signal, as well as application or flow control limitation; for example, QoF [TMA-QOF] rejects component RTTs significantly higher than RTTs over the history of the flow. These heuristics may use the handshake RTT as an initial RTT estimate for a given flow. Usually such heuristics would also detect if the spin is either constant or randomly set for a connection.¶
An on-path observer that can see traffic in both directions (from client to server and from server to client) can also use the spin bit to measure "upstream" and "downstream" component RTT; i.e, the component of the end-to-end RTT attributable to the paths between the observer and the server and the observer and the client, respectively. It does this by measuring the delay between a spin edge observed in the upstream direction and that observed in the downstream direction, and vice versa.¶
In this section, we review specific network management and measurement techniques and how QUIC's design impacts them.¶
Stateful treatment of QUIC traffic (e.g., at a firewall or NAT middlebox) is possible through QUIC traffic and version identification (Section 3.1) and observation of the handshake for connection confirmation (Section 3.2). The lack of any visible end-of-flow signal (Section 3.5) means that this state must be purged either through timers or through least-recently-used eviction, depending on application requirements.¶
[RFC4787] recommends a 2 minute timeout interval for UDP. However, timers can be lower, in the range of 15 to 30 seconds. In contrast, [RFC5382] recommends a timeout of more than 2 hours for TCP, given that TCP is a connection-oriented protocol with well-defined closure semantics. For network devices that are QUIC-aware, it is recommended to also use longer timeouts for QUIC traffic, as QUIC is connection-oriented. As such, a handshake packet from the server indicates the willingness of the server to communicate with the client.¶
The QUIC header optionally contains a connection ID which can be used as additional entropy beyond the 5-tuple, if needed. The QUIC handshake needs to be observed in order to understand whether the connection ID is present and what length it has. However, connection IDs may be renegotiated during a connection, and this renegotiation is not visible to the path. Keying state off the connection ID may therefore cause undetectable and unrecoverable loss of state in the middle of a connection. Use of connection ID specifically discouraged for NAT applications.¶
Limited RTT measurement is possible by passive observation of QUIC traffic; see Section 3.7. No passive measurement of loss is possible with the present wire image. Extremely limited observation of upstream congestion may be possible via the observation of CE markings on ECN-enabled QUIC traffic.¶
In the case of content distribution networking architectures including load balancers, the connection ID provides a way for the server to signal information about the desired treatment of a flow to the load balancers. Guidance on assigning connection IDs is given in [QUIC-APPLICABILITY].¶
Current practices in detection and mitigation of Distributed Denial of Service (DDoS) attacks generally involve classification of incoming traffic (as packets, flows, or some other aggregate) into "good" (productive) and "bad" (DDoS) traffic, and then differential treatment of this traffic to forward only good traffic. This operation is often done in a separate specialized mitigation environment through which all traffic is filtered; a generalized architecture for separation of concerns in mitigation is given in [DOTS-ARCH].¶
Key to successful DDoS mitigation is efficient classification of this traffic in the mitigation environment. Limited first-packet garbage detection as in Section 3.1.2 and stateful tracking of QUIC traffic as in Section 4.1 above may be useful during classification.¶
Note that the use of a connection ID to support connection migration renders 5-tuple based filtering insufficient and requires more state to be maintained by DDoS defense systems. For the common case of NAT rebinding, DDoS defense systems can detect a change in the client's endpoint address by linking flows based on the server's connection IDs. QUIC's linkability resistance ensures that a deliberate connection migration is accompanied by a change in the connection ID.¶
It is questionable whether connection migrations must be supported during a DDoS attack. If the connection migration is not visible to the network that performs the DDoS detection, an active, migrated QUIC connection may be blocked by such a system under attack. As soon as the connection blocking is detected by the client, the client may rely on the fast resumption mechanism provided by QUIC. When clients migrate to a new path, they should be prepared for the migration to fail and attempt to reconnect quickly.¶
TCP syncookies [RFC4937] are a well-established method of mitigating some kinds of TCP DDoS attacks. QUIC Retry packets are the functional analogue to syncookies, forcing clients to prove possession of their IP address before committing server state. However, there are safeguards in QUIC against unsolicited injection of these packets by intermediaries who do not have consent of the end server. See [QUIC_LB] for standard ways for intermediaries to send Retry packets on behalf of consenting servers.¶
Today, UDP is the most prevalent DDoS vector, since it is easy for compromised non-admin applications to send a flood of large UDP packets (while with TCP the attacker gets throttled by the congestion controller) or to craft reflection and amplification attacks. Networks should therefore be prepared for UDP flood attacks on ports used for QUIC traffic. One possible response to this threat is to police UDP traffic on the network, allocating a fixed portion of the network capacity to UDP and blocking UDP datagram over that cap.¶
The recommended way to police QUIC packets is to either drop them all or to throttle them based on the hash of the UDP datagram's source and destination addresses, blocking a portion of the hash space that corresponds to the fraction of UDP traffic one wishes to drop. When the handshake is blocked, QUIC-capable applications may failover to TCP (at least applications using well-known UDP ports). However, blindly blocking a significant fraction of QUIC packets will allow many QUIC handshakes to complete, preventing a TCP failover, but the connections will suffer from severe packet loss.¶
Some deployed in-network functions distinguish pure-acknowledgment (ACK) packets from packets carrying upper-layer data in order to attempt to enhance performance, for example by queueing ACKs differently or manipulating ACK signaling. Distinguishing ACK packets is trivial in TCP, but not supported by QUIC, since acknowledgment signaling is carried inside QUIC's encrypted payload, and ACK manipulation is impossible. Specifically, heuristics attempting to distinguish ACK-only packets from payload-carrying packets based on packet size are likely to fail, and are emphatically NOT RECOMMENDED.¶
It is expected that any QoS handling in the network, e.g. based on use of DiffServ Code Points (DSCPs) [RFC2475] as well as Equal-Cost Multi-Path (ECMP) routing, is applied on a per flow-basis (and not per-packet) and as such that all packets belonging to the same QUIC connection get uniform treatment. Using ECMP to distribute packets from a single flow across multiple network paths or any other non-uniform treatment of packets belong to the same connection could result in variations in order, delivery rate, and drop rate. As feedback about loss or delay of each packet is used as input to the congestion controller, these variations could adversely affect performance.¶
Depending of the loss recovery mechanism implemented, QUIC may be more tolerant of packet re-ordering than traditional TCP traffic (see Section 2.7). However, it cannot be known by the network which exact recovery mechanism is used and therefore reordering tolerance should be considered as unknown.¶
QUIC Connection IDs are opaque byte fields that are expressed consistently across all QUIC versions [QUIC-INVARIANTS], see Section 2.6. This feature may appear to present opportunities to optimize NAT port usage and simplify the work of the QUIC server. In fact, NAT behavior that relies on CID may instead cause connection failure when endpoints change Connection ID, and disable important protocol security features. NATs should retain their existing 4-tuple-based operation and refrain from parsing or otherwise using QUIC connection IDs.¶
This section uses the colloquial term NAT to mean NAPT (section 2.2 of [RFC3022]), which overloads several IP addresses to one IP address or to an IP address pool, as commonly deployed in carrier-grade NATs or residential NATs.¶
The remainder of this section explains how QUIC supports NATs better than other connection-oriented protocols, why NAT use of Connection ID might appear attractive, and how NAT use of CID can create serious problems for the endpoints.¶
[RFC4787] contains some guidance on building NATs to interact constructively with a wide range of applications. This section extends the discussion to QUIC.¶
By using the CID, QUIC connections can survive NAT rebindings as long as no routing function in the path is dependent on client IP address and port to deliver packets between server and NAT. Reducing the timeout on UDP NATs might be tempting in light of this property, but not all QUIC server deployments will be robust to rebinding.¶
NATs sometimes hit an operational limit where they exhaust available public IP addresses and ports, and must evict flows from their address/port mapping. CIDs might appear to offer a way to multiplex many connections over a single address and port.¶
However, QUIC endpoints may negotiate new connection IDs inside cryptographically protected packets, and begin using them at will. Imagine two clients behind a NAT that are sharing the same public IP address and port. The NAT is differentiating them using the incoming Connection ID. If one client secretly changes its connection ID, there will be no mapping for the NAT, and the connection will suddenly break.¶
QUIC is deliberately designed to fail rather than persist when the network cannot support its operation. For HTTP/3, this extends to recommending a fallback to TCP-based versions of HTTP rather than persisting with a QUIC connection that might be unstable. And [I-D.ietf-quic-applicability] recommends TCP fallback for other protocols on the basis that this is preferable to sudden connection errors and time outs. Furthermore, wide deployment of NATs with this behavior hinders the use of QUIC's migration function, which relies on the ability to change the connection ID any time during the lifetime of a QUIC connection.¶
It is possible, in principle, to encode the client's identity in a connection ID using the techniques described in [QUIC_LB] and explicit coordination with the NAT. However, this implies that the client shares configuration with the NAT, which might be logistically difficult. This adds administrative overhead while not resolving the case where a client migrates to a point behind the NAT.¶
Note that multiplexing connection IDs over a single port anyway violates the best common practice to avoid "port overloading" as described in [RFC4787].¶
Concealing client address changes in order to simplify operational routing issues will mask important signals that drive security mechanisms, and therefore opens QUIC up to various attacks.¶
One challenge in QUIC deployments that want to benefit from QUIC's migration capability is server infrastructures with routers and switches that direct traffic based on address-port 4-tuple rather than connection ID. The use of source IP address means that a NAT rebinding or address migration will deliver packets to the wrong server. As all QUIC payloads are encrypted, routers and switches will not have access to negotiated but not-yet-in-use CIDs. This is a particular problem for low-state load balancers. [QUIC_LB] addresses this problem proposing a QUIC extension to allow some server-load balancer coordination for routable CIDs.¶
It seems that a NAT anywhere in the front of such an infrastructure setup could save the effort of converting all these devices by decoding routable connection IDs and rewriting the packet IP addresses to allow consistent routing by legacy devices.¶
Unfortunately, the change of IP address or port is an important signal to QUIC endpoints. It requires a review of path-dependent variables like congestion control parameters. It can also signify various attacks that mislead one endpoint about the best peer address for the connection (see section 9 of [QUIC-TRANSPORT]). The QUIC PATH_CHALLENGE and PATH_RESPONSE frames are intended to detect and mitigate these attacks and verify connectivity to the new address. This mechanism cannot work if the NAT is bleaching peer address changes.¶
For example, an attacker might copy a legitimate QUIC packet and change the source address to match its own. In the absence of a bleaching NAT, the receiving endpoint would interpret this as a potential NAT rebinding and use a PATH_CHALLENGE frame to prove that the peer endpoint is not truly at the new address, thus thwarting the attack. A bleaching NAT has no means of sending an encrypted PATH_CHALLENGE frame, so it might start redirecting all QUIC traffic to the attacker address and thus allow an observer to break the connection.¶
[RFC4787] describes possible packet filtering behaviors that relate to NATs. Though the guidance there holds, a particularly unwise behavior is to admit a handful of UDP packets and then make a decision as to whether or not to filter it. QUIC applications are encouraged to fail over to TCP if early packets do not arrive at their destination. Admitting a few packets allows the QUIC endpoint to determine that the path accepts QUIC. Sudden drops afterwards will result in slow and costly timeouts before abandoning the connection.¶
This document has no actions for IANA.¶
QUIC is an encrypted and authenticated transport. That means, once the cryptographic handshake is complete, QUIC endpoints discard most packets that are not authenticated, greatly limiting the ability of an attacker to interfere with existing connections.¶
However, some information is still observerable, as supporting manageability of QUIC traffic inherently involves tradeoffs with the confidentiality of QUIC's control information; this entire document is therefore security-relevant.¶
More security considerations for QUIC are discussed in [QUIC-TRANSPORT] and [QUIC-TLS], generally considering active or passive attackers in the network as well as attacks on specific QUIC mechanism.¶
Version Negotiation packets do not contain any mechanism to prevent version downgrade attacks. However, future versions of QUIC that use Version Negotiation packets are require to define a mechanism that is robust against version downgrade attacks. Therefore a network node should not attempt to impact version selection, as version downgrade may result in connection failure.¶
The following people have contributed text to sections of this document:¶
Special thanks to Martin Thomson and Martin Duke for the detailed reviews and feedback.¶
This work is partially supported by the European Commission under Horizon 2020 grant agreement no. 688421 Measurement and Architecture for a Middleboxed Internet (MAMI), and by the Swiss State Secretariat for Education, Research, and Innovation under contract no. 15.0268. This support does not imply endorsement.¶
This appendix uses the following conventions: array[i] - one byte at index i of array array[i:j] - subset of array starting with index i (inclusive) up to j-1 (inclusive) array[i:] - subset of array starting with index i (inclusive) up to the end of the array¶
This section contains algorithms that allows parsing versions from both Google QUIC and IETF QUIC. These mechanisms will become irrelevant when IETF QUIC is fully deployed and Google QUIC is deprecated.¶
Note that other than this appendix, nothing in this document applies to Google QUIC. And the purpose of this appendix is merely to distinguish IETF QUIC from any versions of Google QUIC.¶
Conceptually, a Google QUIC version is an opaque 32bit field. When we refer to a version with four printable characters, we use its ASCII representation: for example, Q050 refers to {'Q', '0', '5', '0'} which is equal to {0x51, 0x30, 0x35, 0x30}. Otherwise, we use its hexadecimal representation: for example, 0xff00001d refers to {0xff, 0x00, 0x00, 0x1d}.¶
QUIC versions that start with 'Q' or 'T' followed by three digits are Google QUIC versions. Versions up to and including 43 are documented by <https://docs.google.com/document/d/ 1WJvyZflAO2pq77yOLbp9NsGjC1CHetAXV8I0fQe-B_U/preview>. Versions Q046, Q050, T050, and T051 are not fully documented, but this appendix should contain enough information to allow parsing Client Hellos for those versions.¶
To extract the version number itself, one needs to look at the first byte of the QUIC packet, in other words the first byte of the UDP payload.¶
first_byte = packet[0]
first_byte_bit1 = ((first_byte & 0x80) != 0)
first_byte_bit2 = ((first_byte & 0x40) != 0)
first_byte_bit3 = ((first_byte & 0x20) != 0)
first_byte_bit4 = ((first_byte & 0x10) != 0)
first_byte_bit5 = ((first_byte & 0x08) != 0)
first_byte_bit6 = ((first_byte & 0x04) != 0)
first_byte_bit7 = ((first_byte & 0x02) != 0)
first_byte_bit8 = ((first_byte & 0x01) != 0)
if (first_byte_bit1) {
version = packet[1:5]
} else if (first_byte_bit5 && !first_byte_bit2) {
if (!first_byte_bit8) {
abort("Packet without version")
}
if (first_byte_bit5) {
version = packet[9:13]
} else {
version = packet[5:9]
}
} else {
abort("Packet without version")
}
¶
counter = 0
while (payload[counter] == 0) {
counter += 1
}
first_nonzero_payload_byte = payload[counter]
fnz_payload_byte_bit3 = ((first_nonzero_payload_byte & 0x20) != 0)
if (first_nonzero_payload_byte != 0x06) {
abort("Unexpected frame")
}
if (payload[counter+1] != 0x00) {
abort("Unexpected crypto stream offset")
}
counter += 2
if ((payload[counter] & 0xc0) == 0) {
crypto_data_length = payload[counter]
counter += 1
} else {
crypto_data_length = payload[counter:counter+2]
counter += 2
}
crypto_data = payload[counter:counter+crypto_data_length]
ParseTLS(crypto_data)
¶