This template uses kramdown-rfc: https://github.com/cabo/kramdown-rfc
title: "A Sequence Number Extension for HTTP Datagrams" abbrev: "TODO - Abbreviation" category: std
docname: draft-ihlar-masque-datagram-numbers-latest submissiontype: IETF # also: "independent", "IAB", or "IRTF" number: date: consensus: true v: 3 area: TSV workgroup: Masque Working Group keyword:
- masque
- datagram
- multipath
venue: group: WG type: Working Group mail: masque@ietf.org arch: https://mailarchive.ietf.org/arch/browse/masque/ github: ihlar/draft-ihlar-masque-datagram-numbers
latest: https://example.com/LATEST
fullname: Marcus Ihlar
organization: Ericsson AB
email: marcus.ihlar@ericsson.com
- ins: M. Westerlund name: Magnus Westerlund org: Ericsson email: magnus.westerlund@ericsson.com
normative:
informative: 3GPPTS23.501: target: https://www.3gpp.org/ftp/Specs/archive/23_series/23.501/23501-i22.zip title: System architecture for the 5G System (5GS) - Release 18 # seriesinfo: # ANSSI Technical Report DAT-NT-003 author: - ins: 3rd Generation Partnership Project date: July 2023
--- abstract
This document defines a sequence number extension to HTTP datagrams used to carry proxied UDP payload or IP datagrams. This extension is useful when HTTP datagrams are transported on top of a multipath protocol that does not ensure in-order delivery as it allows for example a masque endpoint to implement reordering logic specific to its needs.
--- middle
This document defines a sequence number extension to HTTP datagrams {{!RFC9297}}. Sequence numbers at the HTTP datagram layer allows a receiving endpoint to implement arbitrary reordering logic, which can be useful when proxied datagrams are sent over multiple paths simultaneously, such as when using the multipath QUIC extension {{?MPQUIC=I-D.ietf-quic-multipath-04}}. The extension applies to HTTP datagrams and specifices its use with the extended CONNECT method, and the protocols connect-ip {{!CONNECT-IP=I-D.ietf-masque-connect-ip}} and connect-udp {{!RFC9298}}.
This extension is motivated by the Access Traffic Steering, Switching, and Splitting (ATSSS) feature outlined for the 5G System by 3GPP in section 5.32 of {{3GPPTS23.501}}.
ATSSS, an optional feature of the 5G system, permits the concurrent usage of 3GPP and non-3GPP accesses within a single PDU session. This is managed by a number of steering functionalities and modes, determining the types of supported concurrent path usage. As of Release 18 of the 5G System Architecture specification, three steering functionalities have been defined for ATSSS: ATSSS-LL, MPTCP, and Multipath QUIC.
ATSSS-LL, a "Lower Layer Functionality," operates beneath the IP layer. It's capable of steering to one path, switching from one path to another, and splitting all traffic types, encompassing both IP and Ethernet PDU sessions. However, it doesn't support splitting a single traffic flow among multiple paths. In contrast, MPTCP and Multipath QUIC, termed as "Higher Layer Functionalities," operate above the IP layer, steering, switching, and splitting TCP and UDP traffic respectively.
The Multipath QUIC steering functionality uses multipath capable HTTP3 proxies supporting the extended CONNECT method with the connect-udp protocol. It establishes two datagram modes for UDP payload encapsulation. The default mode sends HTTP datagrams unreliably over QUIC datagrams, while the optional mode encapsulates UDP payload in HTTP datagrams augmented with sequence numbers.
Steering modes influence how traffic flows utilize concurrent paths. Load Balancing traffic steering and Redundant traffic steering are two modes where sequence numbers prove beneficial.
The Load Balancing steering mode involves parallel transmission over the 3GPP and non-3GPP accesses, a process often referred to as bandwidth aggregation. Distributing data transmission over multiple paths, while increasing available bandwidth, can result in out-of-order packet delivery. The impact of packet disorder is largely dependent on the properties of the protocols and applications conveyed over the proxied payload. Negative effects of large degrees of packet reordering may include increased frequency of packet acknowledgements, inaccurate loss detection and spurious retransmissions. By buffering out-of-order data, an ATSSS endpoint can reconcile path latency differences and reduce the volume of data delivered out-of-order to the final endpoints. Furthermore, an ATSSS endpoint can set an upper bound on the time packets are delayed in its reorder buffer, thus incurring less packet delay variation in the face of loss than if the proxied payload is encoded over reliable streams. Which datagram mode is used for load balancing traffic steering depends on application requirements expressed as ATSSS rules.
With the Redundant steering mode proxied payload is duplicated over the 3GPP and non-3GPP accesses. Despite the added cost, this steering mode is useful for applications and users with stringent availability requirements. Data duplication at one end necessitates de-duplication at the other. This can be efficiently accomplished through sequence numbering, which provides a straightforward method for de-duplication.
{::boilerplate bcp14-tagged}
This document uses notational conventions described in {{Section 1.3 of !RFC9000}}.
The Sequence Number datagram extension prepends sequence numbers to HTTP datagrams. Datagram sequence numbers are unsigned integers initiated to 0 and are incremented by 1 for every transmitted HTTP datagram, except for when the integer overflows and is reset to 0. The extension can be used with the HTTP CONNECT method when the :protocol pseudo header is equal to "connect-udp" or "connect-ip". Use of the sequence number extension is determined per request, and the scope of a datagram sequence is limited to a single request stream. Datagrams with different quarter stream IDs have distinct sequence number spaces.
Endpoints indicate support for Sequence Number Datagram type by including the boolean-valued Item Structured Field "DG-Sequence: ?1" in the HTTP Request and Response headers (See {{Section 3.3.6 of !RFC8941}} for information about the boolean format.).
A datagram sequence is registered by sending a REGISTER_SEQUENCE_CONTEXT capsule. An endpoint MAY send multiple REGISTER_SEQUENCE_CONTEXT capsules in order to support multiple payload formats.
REGISTER_SEQUENCE_CONTEXT Capsule {
Type (i) = REGISTER_SEQUENCE_CONTEXT,
Length (i),
Context ID (i),
Payload Context ID (i),
[Representation (8)]
}
The capsule has the following fields:
Context ID: Identifies a sequence number context. The value MUST be unique within the scope of a request stream.
Payload Context ID: Identifies the type of payload that follows a sequence number. The value MUST be equal to a previously registered Context ID.
Representation: The size in bits of the unsigned interger used to encode the sequence number, the value MUST be one of the following: 8, 16, 32 or 64. This field MUST be present in the first REGISTER_SEQUENCE_CONTEXT capsule sent on a request stream and it MAY be omitted in subsequent capsules.
A Sequence Number Datagram has the following format:
Sequence Number Datagram {
Context ID (i),
Sequence Number (8..64),
Payload (..)
}
Context ID: This value indicates that the datagram contains a sequence number and the format of the data that follows the sequence number.
Sequence Number: Unsigned integer of size specified in registration, indicates the transmission order of the datagagram.
Payload: Datagram payload.
Although the usage of the sequence number is not defined by this specification, there is an underlying assumption that the sequence numbers are assigned in transmission order of HTTP datagram sent in the context of this HTTP request. Any attacker that can break that assumption will thus impact any node that uses the sequence number. By altering the sequence number in HTTP datagrams, an attacker can impact how much data a receiver is buffering for the following purposes:
-
Resource exhaustion attack by maximizing the amount of data buffered in each HTTP request context
-
Introducing reordering, jitter and additional delay in the path properties for these datagram
-
Cause the sequence number using node to drop some HTTP Datagrams by causing them to be so far reordered that some policy in the receiving node drops the datagram.
A malicious endpoint is more likely to mount a resource exhaustion attack, while HTTP intermediares could be used by an third party attacker to impact the HTTP datagram flow between a source and a destination.
A user that buffers datagrams based on sequence numbers should ensure that they have protection against resource exhaustion attacks by limiting the size of their buffers.
This document adds following entries to the "HTTP Capsule Types" registry:
| Capsule Type | Value | Specification |
|---|---|---|
| REGISTER_SEQUENCE_CONTEXT | TBD | (This document) |
| {: #iana-capsule-type title="New Capsule Type to register" cols="l l l"} |
This document adds following entry to the "Hypertext Transfer Protocol (HTTP) Field Name Registry":
| Field Name | Template | Status | Reference | Comments |
|---|---|---|---|---|
| DG-Sequence | permanent | (This document) | ||
| {: #iana-http-field title="HTTP Field Name to register" cols="l l l l l"} | ||||
| --- back |
{:numbered="false"}
TODO acknowledge.