Using HTTP/3 Stream Limits in HTTP/2

Status of This Memo

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 10 May 2024.¶

Copyright Notice

Copyright (c) 2023 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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶

Table of Contents

1. Introduction

HTTP/2 allows endpoints to declare the concurrency limit for streams created by their peer (SETTINGS_MAX_CONCURRENT_STREAMS; see Section 5.1.2 of [HTTP/2]). Initially, the stream concurrency limit is only bounded by the maximum number of streams that can be created, which is 2³⁰ for each endpoint, but the value can be changed at any time. Most endpoints set a smaller value in an attempt to protect the resources that are committed when processing a stream.¶

This limit is not effective in the case that streams are quickly cancelled. The creator of a stream can cancel it using the RST_STREAM frame. The creator of a stream could also cause its peer to send RST_STREAM by purposefully sending frames that violate HTTP/2 rules, unless errors also cause the peer to close the connection. Either of these methods has deterministic and immediate effect, which means that the stream no longer counts against the concurrent stream limit. This means that a malicious endpoint can create or cancel an unbounded number of streams as long as its peer does not set a limit of zero. For clarity, we'll refer to either of these methods as "cancelling".¶

If the creation of the stream results in the expenditure of resources, rapidly creating and cancelling streams can exhaust resources. For a server, clients that create and cancel many requests can effectively deny service; for a client, reserving and cancelling promised streams might have similar effect if a client does not disable server push using the SETTINGS_ENABLE_PUSH setting. This creates a denial of service exposure for which a remedy is not supported in the protocol.¶

However, as noted in Section 10.5 of [HTTP/2], many of the features in HTTP/2 that have some potential to create denial of service attacks also have constructive uses. Distinguishing constructive and destructive uses is often challenging. Some applications of HTTP might find that cancelling many requests is necessary; if cancellation is treated as abusive, it is possible that the treatment necessary to prevent attacks might unintentionally punish some clients.¶

The QUIC protocol [QUIC] on which HTTP/3 [HTTP/3] is built contains an alternative mechanism for limiting concurrency. The scheme in QUIC, as described in Section 4.6 of [QUIC], does not set a concurrent stream limit, but instead relies on a peer increasing the maximum allowed stream identifier. An endpoint cannot create new streams immediately after cancelling an open stream; their peer needs to send a message to make more streams available. In QUIC therefore, a malicious endpoint can only create and cancel a finite number of streams, after which their peer needs to provide consent to continue.¶

This document ports the QUIC stream concurrency limit mechanisms to HTTP/2. As use of this system is voluntary and therefore not necessarily guaranteed, this does not prevent abusive use of stream cancellation. However, deployment of this mechanism in popular implementations might allow endpoints to deploy more aggressive strategies for managing abuse in its absence.¶

2. Conventions and Definitions

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 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

3. The HTTP/2 MAX_STREAMS Frame

The MAX_STREAMS frame (type=0xTBD) is used to limit the streams that a recipient is permitted to create.¶

The format of the MAX_STREAMS frame is illustrated in Figure 1, using the notation defined in Section 1.3 of [QUIC].¶

MAX_STREAMS Frame {
  Length (24) = 0x04,
  Type (8) = 0x0TBD,

  Unused Flags (8),

  Reserved (1),
  Stream Identifier (31) = 0,

  Reserved (1),
  Maximum Stream Identifier (31),
}

In addition to the common HTTP/2 frame header (see Section 4 of [HTTP/2]), the MAX_STREAMS contains a one-bit reserved field and a 31-bit Maxmimum Stream Identifier field.¶

The Reserved field MUST be set to zero when sending and ignored on receipt.¶

The Maximum Stream Identifier field contains the maximum value of stream identifier that a recipient of this frame can use when it creates or reserves a stream; see Section 5.1 of [HTTP/2] for how streams are created.¶

A MAX_STREAMS frame MUST be sent on stream 0. Receipt of a MAX_STREAMS frame on any other stream MUST be treated as a connection error of type PROTOCOL_ERROR; see Section 5.4.1 of [HTTP/2].¶

An endpoint MUST treat receipt of a MAX_STREAMS frame with a length other than 4 as a connection error of type FRAME_SIZE_ERROR; see Section 5.4.1 of [HTTP/2].¶

A client MUST treat receipt of an even-numbered Maximum Stream Identifier as a connection error of type PROTOCOL_ERROR; see Section 5.4.1 of [HTTP/2], with an exception of a value of 0, which can be used to indicate support for the feature without enabling the creation of streams (see Section 3.2). Similarly, a server MUST treat receipt of an odd-numbered Maximum Stream Identifier as a connection error.¶

4. Security Considerations

HTTP/2 considered the cancellation of a stream to reclaim the resources that might have been committed during its creation. The stream limits in HTTP/2 therefore did not provide a means to limit stream creation. Though cancellation is a potential source of abusive traffic, it was not explicitly mentioned (see Section 10.5 of [HTTP/2]). However, experience has shown that the creation of a stream has costs that cannot always be recovered when that stream is cancelled.¶

The use of the MAX_STREAMS frame provides endpoints with a means of limiting stream creation, using a method that has proven to be effective in QUIC. However, repeated use of the MAX_STREAMS frame is also a potential source of abuse, as described in Section 10.5 of [HTTP/2]. Though processing a MAX_STREAMS frame is likely to be trivial, receiving significantly more MAX_STREAMS frames than the number of streams that have been closed might indicate an attempt to waste effort.¶

5. IANA Considerations

This document registers a new entry in the "HTTP/2 Frame Type" registry, as documented in Section 11 of [HTTP/2]. This entry has the following values.¶

Frame Type:

MAX_STREAMS¶

Code:

0xTBD¶

Specification:

Section 3¶

Acknowledgments

This document is written in response to the problems discovered as part of [CVE-2023-44487], so maybe those behind the botnet responsible for that attack deserve some credit.¶

Authors' Addresses