| Internet-Draft | QUIC-LB | February 2021 |
| Duke & Banks | Expires 8 August 2021 | [Page] |
The QUIC protocol design is resistant to transparent packet inspection, injection, and modification by intermediaries. However, the server can explicitly cooperate with network services by agreeing to certain conventions and/or sharing state with those services. This specification provides a standardized means of solving three problems: (1) maintaining routability to servers via a low-state load balancer even when the connection IDs in use change; (2) explicit encoding of the connection ID length in all packets to assist hardware accelerators; and (3) injection of QUIC Retry packets by an anti-Denial-of-Service agent on behalf of the server.¶
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 8 August 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 packets [QUIC-TRANSPORT] usually contain a connection ID to allow endpoints to associate packets with different address/ port 4-tuples to the same connection context. This feature makes connections robust in the event of NAT rebinding. QUIC endpoints usually designate the connection ID which peers use to address packets. Server-generated connection IDs create a potential need for out-of-band communication to support QUIC.¶
QUIC allows servers (or load balancers) to designate an initial connection ID to encode useful routing information for load balancers. It also encourages servers, in packets protected by cryptography, to provide additional connection IDs to the client. This allows clients that know they are going to change IP address or port to use a separate connection ID on the new path, thus reducing linkability as clients move through the world.¶
There is a tension between the requirements to provide routing information and mitigate linkability. Ultimately, because new connection IDs are in protected packets, they must be generated at the server if the load balancer does not have access to the connection keys. However, it is the load balancer that has the context necessary to generate a connection ID that encodes useful routing information. In the absence of any shared state between load balancer and server, the load balancer must maintain a relatively expensive table of server-generated connection IDs, and will not route packets correctly if they use a connection ID that was originally communicated in a protected NEW_CONNECTION_ID frame.¶
This specification provides common algorithms for encoding the server mapping in a connection ID given some shared parameters. The mapping is generally only discoverable by observers that have the parameters, preserving unlinkability as much as possible.¶
Aside from load balancing, a QUIC server may also desire to offload other protocol functions to trusted intermediaries. These intermediaries might include hardware assist on the server host itself, without access to fully decrypted QUIC packets. For example, this document specifies a means of offloading stateless retry to counter Denial of Service attacks. It also proposes a system for self-encoding connection ID length in all packets, so that crypto offload can consistently look up key information.¶
While this document describes a small set of configuration parameters to make the server mapping intelligible, the means of distributing these parameters between load balancers, servers, and other trusted intermediaries is out of its scope. There are numerous well-known infrastructures for distribution of configuration.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].¶
In this document, these words will appear with that interpretation only when in ALL CAPS. Lower case uses of these words are not to be interpreted as carrying significance described in RFC 2119.¶
In this document, "client" and "server" refer to the endpoints of a QUIC connection unless otherwise indicated. A "load balancer" is an intermediary for that connection that does not possess QUIC connection keys, but it may rewrite IP addresses or conduct other IP or UDP processing. A "configuration agent" is the entity that determines the QUIC-LB configuration parameters for the network and leverages some system to distribute that configuration.¶
Note that stateful load balancers that act as proxies, by terminating a QUIC connection with the client and then retrieving data from the server using QUIC or another protocol, are treated as a server with respect to this specification.¶
For brevity, "Connection ID" will often be abbreviated as "CID".¶
All wire formats will be depicted using the notation defined in Section 1.3 of [QUIC-TRANSPORT]. There is one addition: the function len() refers to the length of a field which can serve as a limit on a different field, so that the lengths of two fields can be concisely defined as limited to a sum, for example:¶
x(A..B) y(C..B-len(x))¶
indicates that x can be of any length between A and B, and y can be of any length between C and B provided that (len(x) + len(y)) does not exceed B.¶
The example below illustrates the basic framework:¶
Example Structure {
One-bit Field (1),
7-bit Field with Fixed Value (7) = 61,
Field with Variable-Length Integer (i),
Arbitrary-Length Field (..),
Variable-Length Field (8..24),
Variable-Length Field with Dynamic Limit (8..24-len(Variable-Length Field)),
Field With Minimum Length (16..),
Field With Maximum Length (..128),
[Optional Field (64)],
Repeated Field (8) ...,
}
QUIC is intended to provide unlinkability across connection migration, but servers are not required to provide additional connection IDs that effectively prevent linkability. If the coordination scheme is too difficult to implement, servers behind load balancers using connection IDs for routing will use trivially linkable connection IDs. Clients will therefore be forced to choose between terminating the connection during migration or remaining linkable, subverting a design objective of QUIC.¶
The solution should be both simple to implement and require little additional infrastructure for cryptographic keys, etc.¶
In the limit where there are very few connections to a pool of servers, no scheme can prevent the linking of two connection IDs with high probability. In the opposite limit, where all servers have many connections that start and end frequently, it will be difficult to associate two connection IDs even if they are known to map to the same server.¶
QUIC-LB is relevant in the region between these extremes: when the information that two connection IDs map to the same server is helpful to linking two connection IDs. Obviously, any scheme that transparently communicates this mapping to outside observers compromises QUIC's defenses against linkability.¶
Though not an explicit goal of the QUIC-LB design, concealing the server mapping also complicates attempts to focus attacks on a specific server in the pool.¶
The first octet of a Connection ID is reserved for two special purposes, one mandatory (config rotation) and one optional (length self-description).¶
Subsequent sections of this document refer to the contents of this octet as the "first octet."¶
The first two bits of any connection ID MUST encode an identifier for the configuration that the connection ID uses. This enables incremental deployment of new QUIC-LB settings (e.g., keys).¶
When new configuration is distributed to servers, there will be a transition period when connection IDs reflecting old and new configuration coexist in the network. The rotation bits allow load balancers to apply the correct routing algorithm and parameters to incoming packets.¶
Configuration Agents SHOULD deliver new configurations to load balancers before doing so to servers, so that load balancers are ready to process CIDs using the new parameters when they arrive.¶
A Configuration Agent SHOULD NOT use a codepoint to represent a new configuration until it takes precautions to make sure that all connections using CIDs with an old configuration at that codepoint have closed or transitioned.¶
Servers MUST NOT generate new connection IDs using an old configuration after receiving a new one from the configuration agent. Servers MUST send NEW_CONNECTION_ID frames that provide CIDs using the new configuration, and retire CIDs using the old configuration using the "Retire Prior To" field of that frame.¶
It also possible to use these bits for more long-lived distinction of different configurations, but this has privacy implications (see Section 11.3).¶
If a server has not received a valid QUIC-LB configuration, and believes that low-state, Connection-ID aware load balancers are in the path, it SHOULD generate connection IDs with the config rotation bits set to '11' and SHOULD use the "disable_active_migration" transport parameter in all new QUIC connections. It SHOULD NOT send NEW_CONNECTION_ID frames with new values.¶
A load balancer that sees a connection ID with config rotation bits set to '11' MUST revert to 5-tuple routing.¶
Local hardware cryptographic offload devices may accelerate QUIC servers by receiving keys from the QUIC implementation indexed to the connection ID. However, on physical devices operating multiple QUIC servers, it is impractical to efficiently lookup these keys if the connection ID does not self-encode its own length.¶
Note that this is a function of particular server devices and is irrelevant to load balancers. As such, load balancers MAY omit this from their configuration. However, the remaining 6 bits in the first octet of the Connection ID are reserved to express the length of the following connection ID, not including the first octet.¶
A server not using this functionality SHOULD make the six bits appear to be random.¶
First Octet {
Config Rotation (2),
CID Len or Random Bits (6),
}
The first octet has the following fields:¶
Config Rotation: Indicates the configuration used to interpret the CID.¶
CID Len or Random Bits: Length Self-Description (if applicable), or random bits otherwise. Encodes the length of the Connection ID following the First Octet.¶
In QUIC-LB, load balancers do not generate individual connection IDs for servers. Instead, they communicate the parameters of an algorithm to generate routable connection IDs.¶
The algorithms differ in the complexity of configuration at both load balancer and server. Increasing complexity improves obfuscation of the server mapping.¶
This section describes three participants: the configuration agent, the load balancer, and the server. For any given QUIC-LB configuration that enables connection-ID-aware load balancing, there must be a choice of (1) routing algorithm, (2) server ID allocation strategy, and (3) algorithm parameters.¶
Fundamentally, servers generate connection IDs that encode their server ID. Load balancers decode the server ID from the CID in incoming packets to route to the correct server.¶
There are situations where a server pool might be operating two or more routing algorithms or parameter sets simultaneously. The load balancer uses the first two bits of the connection ID to multiplex incoming DCIDs over these schemes (see Section 3.1).¶
QUIC-LB servers will generate Connection IDs that are decodable to extract a server ID in accordance with a specified algorithm and parameters. However, QUIC often uses client-generated Connection IDs prior to receiving a packet from the server.¶
These client-generated CIDs might not conform to the expectations of the routing algorithm and therefore not be routable by the load balancer. These are called "non-compliant DCIDs":¶
All other DCIDs are compliant.¶
Load balancers MUST forward packets with compliant DCIDs to a server in accordance with the chosen routing algorithm.¶
Load balancers SHOULD drop packets with non-compliant DCIDs in a short header.¶
The routing of long headers with non-compliant DCIDs depends on the server ID allocation strategy, described in Section 4.3. However, the load balancer MUST NOT drop these packets, with one exception.¶
Load balancers MAY drop packets with long headers and non-compliant DCIDs if and only if it knows that the encoded QUIC version does not allow a non- compliant DCID in a packet with that signature. For example, a load balancer can safely drop a QUIC version 1 Handshake packet with a non-compliant DCID, as a version 1 Handshake packet sent to a QUIC-LB compliant server will always have a server-generated compliant CID. The prohibition against dropping packets with long headers remains for unknown QUIC versions.¶
Furthermore, while the load balancer function MUST NOT drop packets, the device might implement other security policies, outside the scope of this specification, that might force a drop.¶
Servers that receive packets with noncompliant CIDs MUST use the available mechanisms to induce the client to use a compliant CID in future packets. In QUIC version 1, this requires using a compliant CID in the Source CID field of server-generated long headers.¶
There are conditions described below where a load balancer routes a packet using an "arbitrary algorithm." It can choose any algorithm, without coordination with the servers, but the algorithm SHOULD be deterministic over short time scales so that related packets go to the same server. The design of this algorithm SHOULD consider the version-invariant properties of QUIC described in [QUIC-INVARIANTS] to maximize its robustness to future versions of QUIC.¶
An arbitrary algorithmr MUST NOT make the routing behavior dependent on any bits in the first octet of the QUIC packet header, except the first bit, which indicates a long header. All other bits are QUIC version-dependent and intermediaries should not base their design on version-specific templates.¶
For example, one arbitrary algorithm might convert a non-compliant DCID to an integer and divided by the number of servers, with the modulus used to forward the packet. The number of servers is usually consistent on the time scale of a QUIC connection handshake. Another might simply hash the address/port 4-tuple. See also Section 10.¶
For any given configuration, the configuration agent must specify if server IDs will be statically or dynamically allocated. Load Balancer configurations with statically allocated server IDs explicitly include a mapping of server IDs to forwarding addresses. The corresponding server configurations contain one or more unique server IDs.¶
A dynamically allocated configuration does not include any bespoke assignment, reducing configuration complexity. However, it places limits on the maximum server ID length and requires more state at the load balancer. In certain edge cases, it can force parts of the system to fail over to 5-tuple routing for a short time.¶
In either case, the configuration agent chooses a server ID length for each configuration that MUST be at least one octet. For Static Allocation, the maximum length depends on the algorithm. For dynamic allocation, the maximum length is 7 octets.¶
A QUIC-LB configuration MAY significantly over-provision the server ID space (i.e., provide far more codepoints than there are servers) to increase the probability that a randomly generated Destination Connection ID is non- compliant.¶
Conceptually, each configuration has its own set of server ID allocations, though two static configurations with identical server ID lengths MAY use a common allocation between them.¶
A server encodes one of its assigned server IDs in any CID it generates using the relevant configuration.¶
In the manual allocation method, the configuration agent assigns at least one server ID to each server.¶
When forwarding a packet with a long header and non-compliant DCID, load balancers MUST forward packets with long headers and non-compliant DCIDs using an arbitrary algorithm as specified in Section 4.2.¶
In the dynamic allocation method, the load balancer assigns server IDs dynamically so that configuration does not require bespoke server ID assignment. This also reduces linkability. However, it requires state at the load balancer that roughly scales with the number of connections, until the server ID codespace is exhausted.¶
The configuration agent does not assign server IDs, but does configure a server ID length and an "LB timeout". The server ID MUST be at least one and no more than seven octets.¶
The load balancer maintains a table of all assigned server IDs and corresponding routing information, which is initialized empty. These tables are independent for each operating configuration.¶
The load balancer MUST keep track of the most recent observation of each server ID, in any sort of packet it forwards, in the table and delete the entries when the time since that observation exceeds the LB Timeout.¶
Note that when the load balancer's table for a configuration is empty, all incoming DCIDs corresponding to that configuration are non-compliant by definition.¶
The handling of a non-compliant long-header packet depends on the reason for non-compliance. The load balancer MUST applyt this logic:¶
Each server maintains a list of server IDs assigned to it, initialized empty. For each SID, it records the last time it received any packet with an CID that encoded that SID.¶
Upon receipt of a packet with a client-generated DCID, the server MUST follow these steps in order:¶
After any possible SID is extracted, the server processes the packet normally.¶
When a server needs a new connection ID, it uses one of the server IDs in its list to populate the server ID field of that CID. It SHOULD vary this selection to reduce linkability within a connection.¶
After loading a new configuration or long periods of idleness, a server may not have any available SIDs. This is because an incoming packet may not the config rotation bits necessary to extract a server ID in accordance with the algorithm above. When required to generate a CID under these conditions, the server MUST generate CIDs using the 5-tuple routing codepoint (see Section 3.2. Note that these connections will not be robust to client address changes while they use this connection ID. For this reason, a server SHOULD retire these connection IDs and replace them with routable ones once it receives a client-generated CID that allows it to acquire a server ID. As, statistically, one in every four such CIDs can provide a server ID, this is typically a short interval.¶
If a server has not received a connection ID encoding a particular server ID within the LB timeout, it MUST retire any outstanding CIDs that use that server ID and cease generating any new ones.¶
A server SHOULD have a mechanism to stop using some server IDs if the list gets large relative to its share of the codepoint space, so that these allocations time out and are freed for reuse by servers that have recently joined the pool.¶
Encryption in the algorithms below uses the AES-128-ECB cipher. Future standards could add new algorithms that use other ciphers to provide cryptographic agility in accordance with [RFC7696]. QUIC-LB implementations SHOULD be extensible to support new algorithms.¶
The Plaintext CID Algorithm makes no attempt to obscure the mapping of connections to servers, significantly increasing linkability. The format is depicted in the figure below.¶
Plaintext CID {
First Octet (8),
Server ID (8..128),
For Server Use (8..152-len(Server ID)),
}
For static SID allocation, the server ID length is limited to 16 octets. There are no parameters specific to this algorithm.¶
On each incoming packet, the load balancer extracts consecutive octets, beginning with the second octet. These bytes represent the server ID.¶
The server chooses how many octets to reserve for its own use, which MUST be at least one octet.¶
When a server needs a new connection ID, it encodes one of its assigned server IDs in consecutive octets beginning with the second. All other bits in the connection ID, except for the first octet, MAY be set to any other value. These other bits SHOULD appear random to observers.¶
The Stream Cipher CID algorithm provides cryptographic protection at the cost of additional per-packet processing at the load balancer to decrypt every incoming connection ID. The CID format is depicted below.¶
Stream Cipher CID {
First Octet (8),
Nonce (64..120),
Encrypted Server ID (8..128-len(Nonce)),
For Server Use (0..152-len(Nonce)-len(Encrypted Server ID)),
}
The configuration agent assigns a server ID to every server in its pool, and determines a server ID length (in octets) sufficiently large to encode all server IDs, including potential future servers.¶
The configuration agent also selects a nonce length and an 16-octet AES-ECB key to use for connection ID decryption. The nonce length MUST be at least 8 octets and no more than 16 octets. The nonce length and server ID length MUST sum to 19 or fewer octets, but SHOULD sum to 15 or fewer to allow space for server use.¶
Upon receipt of a QUIC packet, the load balancer extracts as many of the earliest octets from the destination connection ID as necessary to match the nonce length. The server ID immediately follows.¶
The load balancer decrypts the nonce and the server ID using the following three pass algorithm:¶
server_id_intermediate = encrypted_server_id ^ AES-ECB(key, padded-encrypted-nonce)¶
nonce = encrypted_nonce ^ AES-ECB(key, padded-server_id_intermediate)¶
server_id = server_id_intermediate ^ AES-ECB(key, padded-nonce)¶
For example, if the nonce length is 10 octets and the server ID length is 2 octets, the connection ID can be as small as 13 octets. The load balancer uses the the second through eleventh octets of the connection ID for the nonce, zero-pads it to 16 octets, uses xors the result with the twelfth and thirteenth octet. The result is padded with 14 octets of zeros and encrypted to obtain a mask that is xored with the nonce octets. Finally, the nonce octets are padded with six octets of zeros, encrypted, and the first two octets xored with the server ID octets to obtain the actual server ID.¶
This three-pass algorithm is a simplified version of the FFX algorithm, with the property that each encrypted nonce value depends on all server ID bits, and each encrypted server ID bit depends on all nonce bits and all server ID bits. This mitigates attacks against stream ciphers in which attackers simply flip encrypted server-ID bits.¶
The output of the decryption is the server ID that the load balancer uses for routing.¶
When generating a routable connection ID, the server writes arbitrary bits into its nonce octets, and its provided server ID into the server ID octets. Servers MAY opt to have a longer connection ID beyond the nonce and server ID. The additional bits MAY encode additional information, but SHOULD appear essentially random to observers.¶
If the decrypted nonce bits increase monotonically, that guarantees that nonces are not reused between connection IDs from the same server.¶
The server encrypts the server ID using exactly the algorithm as described in Section 5.2.2, performing the three passes in reverse order.¶
The Block Cipher CID Algorithm, by using a full 16 octets of plaintext and a 128-bit cipher, provides higher cryptographic protection and detection of non-compliant connection IDs. However, it also requires connection IDs of at least 17 octets, increasing overhead of client-to-server packets.¶
Block Cipher CID {
First Octet (8),
Encrypted Server ID (8..128),
Encrypted Bits for Server Use (128-len(Encrypted Server ID)),
Unencrypted Bits for Server Use (0..24),
}
If server IDs are statically allocated, the server ID length MUST be no more than 12 octets, to provide servers adequate entropy to generate unique CIDs.¶
The configuration agent also selects an 16-octet AES-ECB key to use for connection ID decryption.¶
Upon receipt of a QUIC packet, the load balancer reads the first octet to obtain the config rotation bits. It then decrypts the subsequent 16 octets using AES-ECB decryption and the chosen key.¶
The decrypted plaintext contains the server id and opaque server data in that order. The load balancer uses the server ID octets for routing.¶
When generating a routable connection ID, the server MUST choose a connection ID length between 17 and 20 octets. The server writes its server ID into the server ID octets and arbitrary bits into the remaining bits. These arbitrary bits MAY encode additional information, and MUST differ between connection IDs. Bits in the eighteenth, nineteenth, and twentieth octets SHOULD appear essentially random to observers. The first octet is reserved as described in Section 3.¶
The server then encrypts the second through seventeenth octets using the 128-bit AES-ECB cipher.¶
For protocols where 4-tuple load balancing is sufficient, it is straightforward to deliver ICMP packets from the network to the correct server, by reading the echoed IP and transport-layer headers to obtain the 4-tuple. When routing is based on connection ID, further measures are required, as most QUIC packets that trigger ICMP responses will only contain a client-generated connection ID that contains no routing information.¶
To solve this problem, load balancers MAY maintain a mapping of Client IP and port to server ID based on recently observed packets.¶
Alternatively, servers MAY implement the technique described in Section 14.4.1 of [QUIC-TRANSPORT] to increase the likelihood a Source Connection ID is included in ICMP responses to Path Maximum Transmission Unit (PMTU) probes. Load balancers MAY parse the echoed packet to extract the Source Connection ID, if it contains a QUIC long header, and extract the Server ID as if it were in a Destination CID.¶
When a server is under load, QUICv1 allows it to defer storage of connection state until the client proves it can receive packets at its advertised IP address. Through the use of a Retry packet, a token in subsequent client Initial packets, and transport parameters, servers verify address ownership and clients verify that there is no on-path attacker generating Retry packets.¶
A "Retry Service" detects potential Denial of Service attacks and handles sending of Retry packets on behalf of the server. As it is, by definition, literally an on-path entity, the service must communicate some of the original connection IDs back to the server so that it can pass client verification. It also must either verify the address itself (with the server trusting this verification) or make sure there is common context for the server to verify the address using a service-generated token.¶
There are two different mechanisms to allow offload of DoS mitigation to a trusted network service. One requires no shared state; the server need only be configured to trust a retry service, though this imposes other operational constraints. The other requires a shared key, but has no such constraints.¶
Retry services MUST forward all QUIC packets that are not of type Initial or 0-RTT. Other packet types might involve changed IP addresses or connection IDs, so it is not practical for Retry Services to identify such packets as valid or invalid.¶
Regardless of mechanism, a retry service has an active mode, where it is generating Retry packets, and an inactive mode, where it is not, based on its assessment of server load and the likelihood an attack is underway. The choice of mode MAY be made on a per-packet or per-connection basis, through a stochastic process or based on client address.¶
A configuration agent MUST distribute a list of QUIC versions the Retry Service supports. It MAY also distribute either an "Allow-List" or a "Deny-List" of other QUIC versions. It MUST NOT distribute both an Allow-List and a Deny-List.¶
The Allow-List or Deny-List MUST NOT include any versions included for Retry Service Support.¶
The Configuration Agent MUST provide a means for the entity that controls the Retry Service to report its supported version(s) to the configuration Agent. If the entity has not reported this information, it MUST NOT activate the Retry Service and the configuration agent MUST NOT distribute configuration that activates it.¶
The configuration agent MAY delete versions from the final supported version list if policy does not require the Retry Service to operate on those versions.¶
The configuration Agent MUST provide a means for the entities that control servers behind the Retry Service to report either an Allow-List or a Deny-List.¶
If all entities supply Allow-Lists, the consolidated list MUST be the union of these sets. If all entities supply Deny-Lists, the consolidated list MUST be the intersection of these sets.¶
If entities provide a mixture of Allow-Lists and Deny-Lists, the consolidated list MUST be a Deny-List that is the intersection of all provided Deny-Lists and the inverses of all Allow-Lists.¶
If no entities that control servers have reported Allow-Lists or Deny-Lists, the default is a Deny-List with the null set (i.e., all unsupported versions will be admitted). This preserves the future extensibilty of QUIC.¶
A retry service MUST forward all packets for a QUIC version it does not support that are not on a Deny-List or absent from an Allow-List. Note that if servers support versions the retry service does not, this may increase load on the servers.¶
Note that future versions of QUIC might not have Retry packets, require different information in Retry, or use different packet type indicators.¶
QUIC-LB requires common configuration to synchronize understanding of encodings and guarantee explicit consent of the server.¶
The load balancer and server MUST agree on a routing algorithm, server ID allocation method, and the relevant parameters for that algorithm.¶
All algorithms require a server ID length. If server IDs are statically allocated, the load balancer MUST receive the full table of mappings, and each server must receive its assigned SID(s), from the configuration agent.¶
For Stream Cipher CID Routing, the servers and load balancer also MUST have a common understanding of the key and nonce length.¶
For Block Cipher CID Routing, the servers and load balancer also MUST have a common understanding of the key.¶
Note that server IDs are opaque bytes, not integers, so there is no notion of network order or host order.¶
A server configuration MUST specify if the first octet encodes the CID length. Note that a load balancer does not need the CID length, as the required bytes are present in the QUIC packet.¶
A full QUIC-LB server configuration MUST also specify the supported QUIC versions of any Retry Service. If a shared-state service, the server also must have the token key.¶
A non-shared-state Retry Service need only be configured with the QUIC versions it supports, and an Allow- or Deny-List. A shared-state Retry Service also needs the token key, and to be aware if a NAT sits between it and the servers.¶
Appendix A provides a YANG Model of the a full QUIC-LB configuration.¶
This section discusses considerations for some deployment scenarios not implied by the specification above.¶
Some network architectures may have multiple tiers of low-state load balancers, where a first tier of devices makes a routing decision to the next tier, and so on, until packets reach the server. Although QUIC-LB is not explicitly designed for this use case, it is possible to support it.¶
If each load balancer is assigned a range of server IDs that is a subset of the range of IDs assigned to devices that are closer to the client, then the first devices to process an incoming packet can extract the server ID and then map it to the correct forwarding address. Note that this solution is extensible to arbitrarily large numbers of load-balancing tiers, as the maximum server ID space is quite large.¶
Some deployments may transparently move a connection from one server to another. The means of transferring connection state between servers is out of scope of this document.¶
To support a handover, a server involved in the transition could issue CIDs that map to the new server via a NEW_CONNECTION_ID frame, and retire CIDs associated with the new server using the "Retire Prior To" field in that frame.¶
Alternately, if the old server is going offline, the load balancer could simply map its server ID to the new server's address.¶
Non-shared-state Retry Services are inherently dependent on the format (and existence) of Retry Packets in each version of QUIC, and so Retry Service configuration explicitly includes the supported QUIC versions.¶
The server ID encodings, and requirements for their handling, are designed to be QUIC version independent (see [QUIC-INVARIANTS]). A QUIC-LB load balancer will generally not require changes as servers deploy new versions of QUIC. However, there are several unlikely future design decisions that could impact the operation of QUIC-LB.¶
The maximum Connection ID length could be below the minimum necessary for one or more encoding algorithms.¶
Section 4.1 provides guidance about how load balancers should handle non-compliant DCIDs. This guidance, and the implementation of an algorithm to handle these DCIDs, rests on some assumptions:¶
While this document does not update the commitments in [QUIC-INVARIANTS], the additional assumptions are minimal and narrowly scoped, and provide a likely set of constants that load balancers can use with minimal risk of version- dependence.¶
If these assumptions are invalid, this specification is likely to lead to loss of packets that contain non-compliant DCIDs, and in extreme cases connection failure.¶
Some load balancers might inspect elements of the Server Name Indication (SNI) extension in the TLS Client Hello to make a routing decision. Note that the format and cryptographic protection of this information may change in future versions or extensions of TLS or QUIC, and therefore this functionality is inherently not version-invariant.¶
QUIC-LB is intended to prevent linkability. Attacks would therefore attempt to subvert this purpose.¶
Note that the Plaintext CID algorithm makes no attempt to obscure the server mapping, and therefore does not address these concerns. It exists to allow consistent CID encoding for compatibility across a network infrastructure, which makes QUIC robust to NAT rebinding. Servers that are running the Plaintext CID algorithm SHOULD only use it to generate new CIDs for the Server Initial Packet and SHOULD NOT send CIDs in QUIC NEW_CONNECTION_ID frames, except that it sends one new Connection ID in the event of config rotation Section 3.1. Doing so might falsely suggest to the client that said CIDs were generated in a secure fashion.¶
A linkability attack would find some means of determining that two connection IDs route to the same server. As described above, there is no scheme that strictly prevents linkability for all traffic patterns, and therefore efforts to frustrate any analysis of server ID encoding have diminishing returns.¶
Any attacker might open a connection to the server infrastructure and aggressively simulate migration to obtain a large sample of IDs that map to the same server. It could then apply analytical techniques to try to obtain the server encoding.¶
The Stream and Block Cipher CID algorithms provide robust protection against any sort of linkage. The Plaintext CID algorithm makes no attempt to protect this encoding.¶
Were this analysis to obtain the server encoding, then on-path observers might apply this analysis to correlating different client IP addresses.¶
Attackers in this privileged position are intrinsically able to map two connection IDs to the same server. The QUIC-LB algorithms do prevent the linkage of two connection IDs to the same individual connection if servers make reasonable selections when generating new IDs for that connection.¶
During the period in which there are multiple deployed configuration IDs (see Section 3.1), there is a slight increase in linkability. The server space is effectively divided into segments with CIDs that have different config rotation bits. Entities that manage servers SHOULD strive to minimize these periods by quickly deploying new configurations across the server pool.¶
A simple deployment of QUIC-LB in a cloud provider might use the same global QUIC-LB configuration across all its load balancers that route to customer servers. An attacker could then simply become a customer, obtain the configuration, and then extract server IDs of other customers' connections at will.¶
To avoid this, the configuration agent SHOULD issue QUIC-LB configurations to mutually distrustful servers that have different keys for encryption algorithms. The load balancers can distinguish these configurations by external IP address, or by assigning different values to the config rotation bits (Section 3.1). Note that either solution has a privacy impact; see Section 11.3.¶
These techniques are not necessary for the plaintext algorithm, as it does not attempt to conceal the server ID.¶
Section 21.9 of [QUIC-TRANSPORT] discusses the Stateless Reset Oracle attack. For a server deployment to be vulnerable, an attacking client must be able to cause two packets with the same Destination CID to arrive at two different servers that share the same cryptographic context for Stateless Reset tokens. As QUIC-LB requires deterministic routing of DCIDs over the life of a connection, it is a sufficient means of avoiding an Oracle without additional measures.¶
The Stream Cipher and Block Cipher algorithms need to generate different cipher text for each generated Connection ID instance to protect the Server ID. To do so, at least four octets of the Block Cipher CID and at least eight octets of the Stream Cipher CID are reserved for a nonce that, if used only once, will result in unique cipher text for each Connection ID.¶
If servers simply increment the nonce by one with each generated connection ID, then it is safe to use the existing keys until any server's nonce counter exhausts the allocated space and rolls over to zero. Whether or not it implements this method, the server MUST NOT reuse a nonce until it switches to a configuration with new keys.¶
Configuration agents SHOULD implement an out-of-band method to discover when servers are in danger of exhausting their nonce space, and SHOULD respond by issuing a new configuration. A server that has exhausted its nonces MUST either switch to a different configuration, or if none exists, use the 4-tuple routing config rotation codepoint.¶
There are no IANA requirements.¶
This YANG model conforms to [RFC6020] and expresses a complete QUIC-LB configuration.¶
module ietf-quic-lb {
yang-version "1.1";
namespace "urn:ietf:params:xml:ns:yang:ietf-quic-lb";
prefix "quic-lb";
import ietf-yang-types {
prefix yang;
reference
"RFC 6991: Common YANG Data Types.";
}
import ietf-inet-types {
prefix inet;
reference
"RFC 6991: Common YANG Data Types.";
}
organization
"IETF QUIC Working Group";
contact
"WG Web: <http://datatracker.ietf.org/wg/quic>
WG List: <quic@ietf.org>
Authors: Martin Duke (martin.h.duke at gmail dot com)
Nick Banks (nibanks at microsoft dot com)";
description
"This module enables the explicit cooperation of QUIC servers with
trusted intermediaries without breaking important protocol features.
Copyright (c) 2021 IETF Trust and the persons identified as
authors of the code. All rights reserved.
Redistribution and use in source and binary forms, with or
without modification, is permitted pursuant to, and subject to
the license terms contained in, the Simplified BSD License set
forth in Section 4.c of the IETF Trust's Legal Provisions
Relating to IETF Documents
(https://trustee.ietf.org/license-info).
This version of this YANG module is part of RFC XXXX
(https://www.rfc-editor.org/info/rfcXXXX); see the RFC itself
for full legal notices.
The key words 'MUST', 'MUST NOT', 'REQUIRED', 'SHALL', 'SHALL
NOT', 'SHOULD', 'SHOULD NOT', 'RECOMMENDED', 'NOT RECOMMENDED',
'MAY', and 'OPTIONAL' in this document are to be interpreted as
described in BCP 14 (RFC 2119) (RFC 8174) when, and only when,
they appear in all capitals, as shown here.";
revision "2021-01-29" {
description
"Initial Version";
reference
"RFC XXXX, QUIC-LB: Generating Routable QUIC Connection IDs";
}
container quic-lb {
presence "The container for QUIC-LB configuration.";
description
"QUIC-LB container.";
typedef quic-lb-key {
type yang:hex-string {
length 47;
}
description
"This is a 16-byte key, represented with 47 bytes";
}
list cid-configs {
key "config-rotation-bits";
description
"List up to three load balancer configurations";
leaf config-rotation-bits {
type uint8 {
range "0..2";
}
mandatory true;
description
"Identifier for this CID configuration.";
}
leaf first-octet-encodes-cid-length {
type boolean;
default false;
description
"If true, the six least significant bits of the first CID
octet encode the CID length minus one.";
}
leaf cid-key {
type quic-lb-key;
description
"Key for encrypting the connection ID. If absent, the
configuration uses the Plaintext algorithm.";
}
leaf nonce-length {
type uint8 {
range "8..16";
}
must '(../cid-key)' {
error-message "nonce-length only valid if cid-key is set";
}
description
"Length, in octets, of the nonce. If absent when cid-key is
present, the configuration uses the Block Cipher Algorithm.
If present along with cid-key, the configurationuses the
Stream Cipher Algorithm.";
}
leaf lb-timeout {
type uint32;
description
"Existence means the configuration uses dynamic Server ID allocation.
Time (in seconds) to keep a server ID allocation if no packets with
that server ID arrive.";
}
leaf server-id-length {
type uint8 {
range "1..18";
}
must '(../lb-timeout and . <= 7) or
(not(../lb-timeout) and
(not(../cid-key) and . <= 16) or
((../nonce-length) and . <= (19 - ../nonce-length)) or
((../cid-key) and not(../nonce-length) and . <= 12))' {
error-message
"Server ID length too long for routing algorithm and server ID
allocation method";
}
mandatory true;
description
"Length (in octets) of a server ID. Further range-limited
by sid-allocation, cid-key, and nonce-length.";
}
list server-id-mappings {
when "not(../lb-timeout)";
key "server-id";
description "Statically allocated Server IDs";
leaf server-id {
type yang:hex-string;
must "string-length(.) = 3 * ../../server-id-length - 1";
mandatory true;
description
"An allocated server ID";
}
leaf server-address {
type inet:ip-address;
mandatory true;
description
"Destination address corresponding to the server ID";
}
}
}
container retry-service-config {
description
"Configuration of Retry Service. If supported-versions is empty, there
is no retry service. If token-keys is empty, it uses the non-shared-
state service. If present, it uses shared-state tokens.";
leaf-list supported-versions {
type uint32;
description
"QUIC versions that the retry service supports. If empty, there
is no retry service.";
}
leaf unsupported-version-default {
type enumeration {
enum allow {
description "Unsupported versions admitted by default";
}
enum deny {
description "Unsupported versions denied by default";
}
}
default allow;
description
"Are unsupported versions not in version-exceptions allowed
or denied?";
}
leaf-list version-exceptions {
type uint32;
description
"Exceptions to the default-deny or default-allow rule.";
}
list token-keys {
key "key-sequence-number";
description
"list of active keys, for key rotation purposes. Existence implies
shared-state format";
leaf key-sequence-number {
type uint8;
mandatory true;
description
"Identifies the key used to encrypt the token";
}
leaf token-key {
type quic-lb-key;
mandatory true;
description
"16-byte key to encrypt the token";
}
leaf token-iv {
type yang:hex-string {
length 23;
}
mandatory true;
description
"8-byte IV to encrypt the token, encoded in 23 bytes";
}
}
}
}
}
¶
This summary of the YANG model uses the notation in [RFC8340].¶
module: ietf-quic-lb
+--rw quic-lb
+--rw cid-configs*
| [config-rotation-bits]
| +--rw config-rotation-bits uint8
| +--rw first-octet-encodes-cid-length? boolean
| +--rw cid-key? yang:hex-string
| +--rw nonce-length? uint8
| +--rw lb-timeout? uint32
| +--rw server-id-length uint8
| +--rw server-id-mappings*?
| | [server-id]
| | +--rw server-id yang:hex-string
| | +--rw server-address inet:ip-address
+--ro retry-service-config
| +--rw supported-versions*
| | +--rw version uint32
| +--rw unsupported-version-default enumeration {allow deny}
| +--rw version-exceptions*
| | +--rw version uint32
| +--rw token-keys*?
| | [key-sequence-number]
| | +--rw key-sequence-number uint8
| | +--rw token-key yang:hex-string
| | +--rw token-iv yang:hex-string
¶
Each section of this draft includes multiple sets of load balancer configuration, each of which has five examples of server ID and server use bytes and how they are encoded in a CID.¶
In some cases, there are no server use bytes. Note that, for simplicity, the first octet bits used for neither config rotation nor length self-encoding are random, rather than listed in the server use field. Therefore, a server implementation using these parameters may generate CIDs with a slightly different first octet.¶
This section uses the following abbreviations:¶
cid Connection ID cr_bits Config Rotation Bits LB Load Balancer sid Server ID sid_len Server ID length su Server Use Bytes¶
All values except length_self_encoding and sid_len are expressed in hexidecimal format.¶
LB configuration: cr_bits 0x0 length_self_encoding: y sid_len 1 cid 01be sid be su cid 0221b7 sid 21 su b7 cid 03cadfd8 sid ca su dfd8 cid 041e0c9328 sid 1e su 0c9328 cid 050c8f6d9129 sid 0c su 8f6d9129 LB configuration: cr_bits 0x0 length_self_encoding: n sid_len 2 cid 02aab0 sid aab0 su cid 3ac4b106 sid c4b1 su 06 cid 08bd3cf4a0 sid bd3c su f4a0 cid 3771d59502d6 sid 71d5 su 9502d6 cid 1d57dee8b888f3 sid 57de su e8b888f3 LB configuration: cr_bits 0x0 length_self_encoding: y sid_len 3 cid 0336c976 sid 36c976 su cid 04aa291806 sid aa2918 su 06 cid 0586897bd8b6 sid 86897b su d8b6 cid 063625bcae4de0 sid 3625bc su ae4de0 cid 07966fb1f3cb535f sid 966fb1 su f3cb535f LB configuration: cr_bits 0x0 length_self_encoding: n sid_len 4 cid 185172fab8 sid 5172fab8 su cid 2eb7ff2c9297 sid b7ff2c92 su 97 cid 14f3eb3dd3edbe sid f3eb3dd3 su edbe cid 3feb31cece744b74 sid eb31cece su 744b74 cid 06b9f34c353ce23bb5 sid b9f34c35 su 3ce23bb5 LB configuration: cr_bits 0x0 length_self_encoding: y sid_len 5 cid 05bdcd8d0b1d sid bdcd8d0b1d su cid 06aee673725a63 sid aee673725a su 63 cid 07bbf338ddbf37f4 sid bbf338ddbf su 37f4 cid 08fbbca64c26756840 sid fbbca64c26 su 756840 cid 09e7737c495b93894e34 sid e7737c495b su 93894e34¶
In each case below, the server is using a plain text nonce value of zero.¶
LB configuration: cr_bits 0x0 length_self_encoding: y nonce_len 12 sid_len 1
key 4d9d0fd25a25e7f321ef464e13f9fa3d
cid 0d69fe8ab8293680395ae256e89c sid c5 su
cid 0e420d74ed99b985e10f5073f43027 sid d5 su 27
cid 0f380f440c6eefd3142ee776f6c16027 sid 10 su 6027
cid 1020607efbe82049ddbf3a7c3d9d32604d sid 3c su 32604d
cid 11e132d12606a1bb0fa17e1caef00ec54c10 sid e3 su 0ec54c10
LB configuration: cr_bits 0x0 length_self_encoding: n nonce_len 12 sid_len 2
key 49e1cec7fd264b1f4af37413baf8ada9
cid 3d3a5e1126414271cc8dc2ec7c8c15 sid f7fe su
cid 007042539e7c5f139ac2adfbf54ba748 sid eaf4 su 48
cid 2bc125dd2aed2aafacf59855d99e029217 sid e880 su 9217
cid 3be6728dc082802d9862c6c8e4dda3d984d8 sid 62c6 su d984d8
cid 1afe9c6259ad350fc7bad28e0aeb2e8d4d4742 sid 8502 su 8d4d4742
LB configuration: cr_bits 0x0 length_self_encoding: y nonce_len 14 sid_len 3
key 2c70df0b399bd33a7335523dcdb884ad
cid 11d62e8670565cd30b552edff6782ff5a740 sid d794bb su
cid 12c70e481f49363cabd9370d1fd5012c12bca5 sid 2cbd5d su a5
cid 133b95dfd8ad93566782f8424df82458069fc9e9 sid d126cd su c9e9
cid 13ac6ffcd635532ab60370306c7ee572d6b6e795 sid 539e42 su e795
cid 1383ed07a9700777ff450bb39bb9c1981266805c sid 9094dd su 805c
LB configuration: cr_bits 0x0 length_self_encoding: n nonce_len 12 sid_len 4
key 2297b8a95c776cf9c048b76d9dc27019
cid 32873890c3059ca62628089439c44c1f84 sid 7398d8ca su
cid 1ff7c7d7b9823954b178636c99a7dc93ac83 sid 9655f091 su 83
cid 31044000a5ebb3bf2fa7629a17f2c78b077c17 sid 8b035fc6 su 7c17
cid 1791bd28c66721e8fea0c6f34fd2d8e663a6ef70 sid 6672e0e2 su a6ef70
cid 3df1d90ad5ccd5f8f475f040e90aeca09ec9839d sid b98b1fff su c9839d
LB configuration: cr_bits 0x0 length_self_encoding: y nonce_len 8 sid_len 5
key 484b2ed942d9f4765e45035da3340423
cid 0da995b7537db605bfd3a38881ae sid 391a7840dc su
cid 0ed8d02d55b91d06443540d1bf6e98 sid 10f7f7b284 su 98
cid 0f3f74be6d46a84ccb1fd1ee92cdeaf2 sid 0606918fc0 su eaf2
cid 1045626dbf20e03050837633cc5650f97c sid e505eea637 su 50f97c
cid 11bb9a17f691ab446a938427febbeb593eaa sid 99343a2a96 su eb593eaa
¶
LB configuration: cr_bits 0x0 length_self_encoding: y sid_len 1
key 411592e4160268398386af84ea7505d4
cid 10564f7c0df399f6d93bdddb1a03886f25 sid 23 su 05231748a80884ed58007847eb9fd0
cid 10d5c03f9dd765d73b3d8610b244f74d02 sid 15 su 76cd6b6f0d3f0b20fc8e633e3a05f3
cid 108ca55228ab23b92845341344a2f956f2 sid 64 su 65c0ce170a9548717498b537cb8790
cid 10e73f3d034aef2f6f501e3a7693d6270a sid 07 su f9ad10c84cc1e89a2492221d74e707
cid 101a6ce13d48b14a77ecfd365595ad2582 sid 6c su 76ce4689b0745b956ef71c2608045d
LB configuration: cr_bits 0x0 length_self_encoding: n sid_len 2
key 92ce44aecd636aeeff78da691ef48f77
cid 20aa09bc65ed52b1ccd29feb7ef995d318 sid a52f su 99278b92a86694ff0ecd64bc2f73
cid 30b8dbef657bd78a2f870e93f9485d5211 sid 6c49 su 7381c8657a388b4e9594297afe96
cid 043a8137331eacd2e78383279b202b9a6d sid 4188 su 5ac4b0e0b95f4e7473b49ee2d0dd
cid 3ba71ea2bcf0ab95719ab59d3d7fde770d sid 8ccc su 08728807605db25f2ca88be08e0f
cid 37ef1956b4ec354f40dc68336a23d42b31 sid c89d su 5a3ccd1471caa0de221ad6c185c0
LB configuration: cr_bits 0x0 length_self_encoding: y sid_len 3
key 5c49cb9265efe8ae7b1d3886948b0a34
cid 10efcffc161d232d113998a49b1dbc4aa0 sid 0690b3 su 958fc9f38fe61b83881b2c5780
cid 10fc13bdbcb414ba90e391833400c19505 sid 031ac3 su 9a55e1e1904e780346fcc32c3c
cid 10d3cc1efaf5dc52c7a0f6da2746a8c714 sid 572d3a su ff2ec9712664e7174dc03ca3f8
cid 107edf37f6788e33c0ec7758a485215f2b sid 562c25 su 02c5a5dcbea629c3840da5f567
cid 10bc28da122582b7312e65aa096e9724fc sid 2fa4f0 su 8ae8c666bfc0fc364ebfd06b9a
LB configuration: cr_bits 0x0 length_self_encoding: n sid_len 4
key e787a3a491551fb2b4901a3fa15974f3
cid 26125351da12435615e3be6b16fad35560 sid 0cb227d3 su 65b40b1ab54e05bff55db046
cid 14de05fc84e41b611dfbe99ed5b1c9d563 sid 6a0f23ad su d73bee2f3a7e72b3ffea52d9
cid 1306052c3f973db87de6d7904914840ff1 sid ca21402d su 5829465f7418b56ee6ada431
cid 1d202b5811af3e1dba9ea2950d27879a92 sid b14e1307 su 4902aba8b23a5f24616df3cf
cid 26538b78efc2d418539ad1de13ab73e477 sid a75e0148 su 0040323f1854e75aeb449b9f
LB configuration: cr_bits 0x0 length_self_encoding: y sid_len 5
key d5a6d7824336fbe0f25d28487cdda57c
cid 10a2794871aadb20ddf274a95249e57fde sid 82d3b0b1a1 su 0935471478c2edb8120e60
cid 108122fe80a6e546a285c475a3b8613ec9 sid fbcc902c9d su 59c47946882a9a93981c15
cid 104d227ad9dd0fef4c8cb6eb75887b6ccc sid 2808e22642 su 2a7ef40e2c7e17ae40b3fb
cid 10b3f367d8627b36990a28d67f50b97846 sid 5e018f0197 su 2289cae06a566e5cb6cfa4
cid 1024412bfe25f4547510204bdda6143814 sid 8a8dd3d036 su 4b12933a135e5eaaebc6fd
¶
The authors would like to thank Christian Huitema and Ian Swett for their major design contributions.¶
Manasi Deval, Erik Fuller, Toma Gavrichenkov, Jana Iyengar, Subodh Iyengar, Ladislav Lhotka, Jan Lindblad, Ling Tao Nju, Kazuho Oku, Udip Pant, Martin Thomson, Dmitri Tikhonov, Victor Vasiliev, and William Zeng Ke all provided useful input to this document.¶