| RFC 9687 | BGP SendHoldTimer | November 2024 |
| Snijders, et al. | Standards Track | [Page] |
This document defines the SendHoldTimer, along with the SendHoldTimer_Expires event, for the Border Gateway Protocol (BGP) Finite State Machine (FSM).
Implementation of the SendHoldTimer helps overcome situations where a BGP connection is not terminated after the local system detects that the remote system is not processing BGP messages.
This document specifies that the local system should close the BGP connection and not solely rely on the remote system for connection closure when the SendHoldTimer expires.
This document updates RFC 4271.¶
This is an Internet Standards Track document.¶
This document is a product of the Internet Engineering Task Force (IETF). It represents the consensus of the IETF community. It has received public review and has been approved for publication by the Internet Engineering Steering Group (IESG). Further information on Internet Standards is available in Section 2 of RFC 7841.¶
Information about the current status of this document, any errata, and how to provide feedback on it may be obtained at https://www.rfc-editor.org/info/rfc9687.¶
Copyright (c) 2024 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.¶
This document defines the SendHoldTimer, along with the SendHoldTimer_Expires event, for the Border Gateway Protocol (BGP) Finite State Machine (FSM) defined in Section 8 of [RFC4271].¶
Failure to terminate a blocked BGP connection can result in network reachability issues, and the subsequent failure to generate and deliver BGP UPDATE messages to another BGP speaker of the local system is detrimental to all participants of the inter-domain routing system. This phenomena is thought to have contributed to IP traffic packet loss events in the global Internet routing system [bgpzombies].¶
This specification intends to improve this situation by requiring that BGP connections be terminated if the local system has detected that the remote system cannot possibly have processed any BGP messages for the duration of the SendHoldTime.
Through standardization of the aforementioned requirement, operators will benefit from consistent behavior across different BGP implementations.¶
BGP speakers following this specification do not rely exclusively on remote systems closing blocked connections; they also locally close blocked connections.¶
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.¶
In implementations lacking the concept of a SendHoldTimer, a malfunctioning or overwhelmed remote speaker may cause data on the BGP socket in the local system to accumulate ad infinitum.
This could result in forwarding failure and traffic loss, as the overwhelmed speaker continues to utilize stale routes.¶
An example fault state: as BGP runs over TCP [RFC9293], it is possible for a BGP speaker in the Established state to encounter a BGP speaker that is advertising a TCP Receive Window (RCV.WND) of size zero. The size zero of this window prevents the local system from sending KEEPALIVE, UPDATE, or any other critical BGP messages across the network socket to the remote speaker.¶
Generally BGP implementations have no visibility into lower-layer subsystems such as TCP or the speaker's current Receive Window size, and there is no existing BGP mechanism for such a blocked connection to be recognized. Hence BGP implementations are not able to handle this situation in a consistent fashion.¶
The primary issue that arises when a BGP speaker is unable to send a BGP message to a remote speaker is that the affected speaker may end up operating with outdated routing information. Failure of the BGP speaker to send (and thus the remote speaker to receive) BGP messages on a single BGP session can negatively impact the ability of an entire autonomous system (or even a group of autonomous systems) to converge.¶
SendHoldTimer
BGP speakers are implemented following a conceptual model "BGP Finite State Machine" (FSM), which is outlined in Section 8 of [RFC4271].
This specification adds a BGP timer, SendHoldTimer, and updates the BGP FSM as indicated in the following subsections.¶
The following optional session attributes for each connection are added to the list in Section 8 of [RFC4271] appearing just prior to "The optional session attributes support different features of the BGP functionality that have implications for the BGP FSM state transitions":¶
NEW¶
SendHoldTime determines how long a BGP speaker will stay in the Established state before the TCP connection is dropped because no BGP messages can be transmitted to its peer.
A BGP speaker can configure the value of the SendHoldTime for each peer independently.¶
SendHoldTimer_Expires
Another timer event is added to Section 8.1.3 of [RFC4271] as follows:¶
NEW¶
The following changes are made to Section 8.2.2 of [RFC4271].¶
In "OpenConfirm State", the handling of Event 26 is revised as follows:¶
OLD¶
If the local system receives a KEEPALIVE message (KeepAliveMsg (Event 26)), the local system:¶
NEW¶
If the local system receives a KEEPALIVE message (KeepAliveMsg (Event 26)), the local system:¶
The following paragraph is added to Section 8.2.2 of [RFC4271] in "Established State", after the paragraph that ends "unless the negotiated HoldTime value is zero":¶
NEW¶
If the
SendHoldTimer_Expires(Event 29) occurs, the local system:¶
- (optionally) sends a NOTIFICATION message with the BGP Error Code "Send Hold Timer Expired" if the local system can determine that doing so will not delay the following actions in this paragraph,¶
- logs an error message in the local system with the BGP Error Code "Send Hold Timer Expired",¶
- releases all BGP resources,¶
- sets the ConnectRetryTimer to zero,¶
- drops the TCP connection,¶
- increments the ConnectRetryCounter by 1,¶
- (optionally) performs peer oscillation damping if the DampPeerOscillations attribute is set to TRUE, and¶
- changes its state to Idle.¶
Each time the local system sends a BGP message, it restarts the
SendHoldTimerunless theSendHoldTimevalue is zero or the negotiatedHoldTimevalue is zero, in which case theSendHoldTimeris stopped.¶The
SendHoldTimeris stopped following any transition out of the Established state as part of the "release all BGP resources" action.¶
Section 10 of [RFC4271] summarizes BGP timers.
This document adds another optional BGP timer: SendHoldTimer.¶
NEW¶
SendHoldTimeis an FSM attribute that stores the initial value for theSendHoldTimer. IfSendHoldTimeis non-zero, then it MUST be greater than the value ofHoldTime; see Section 6 of [RFC9687] for suggested default values.¶
If the local system does not send any BGP messages within the period specified in SendHoldTime, then a NOTIFICATION message with the "Send Hold Timer Expired" Error Code MAY be sent and the BGP connection MUST be closed.
Additionally, an error MUST be logged in the local system, indicating the "Send Hold Timer Expired" Error Code.¶
Due to the relative rarity of the failure mode that this specification is designed to address, and also the fact that network operators may be unfamiliar with the formal specification of BGP fault detection mechanisms such as HoldTimer, it is likely that a large number of operators will be unaware of the need for an additional mechanism such as SendHoldTimer.¶
Accordingly, it is RECOMMENDED that implementations of this specification enable SendHoldTimer by default, without requiring additional configuration of the BGP-speaking device.¶
The default value of SendHoldTime for a BGP connection SHOULD be the greater of:¶
Implementations MAY make the value of SendHoldTime configurable, either globally or on a per-peer basis, within the constraints set out in Section 4.4.¶
The subcode for NOTIFICATION message "Send Hold Timer Expired" is set to 0 and is not used; no additional data is to be appended to the end of a "Send Hold Timer Expired" NOTIFICATION message.¶
When the local system recognizes that a remote speaker has not processed any BGP messages for the duration of the SendHoldTime, it is likely that the local system will not be able to inform the remote peer through a NOTIFICATION message as to why the connection is being closed.
This document suggests that an attempt to send a NOTIFICATION message with the "Send Hold Timer Expired" Error Code still be made, if doing so will not delay closing the BGP connection.
Meanwhile, an error message is logged in the local system.¶
Other mechanisms can be used as well, for example, BGP speakers SHOULD provide this reason ("Send Hold Timer Expired") as part of their operational state (for example, bgpPeerLastError in the BGP MIB [RFC4273]).¶
This specification does not change BGP's security characteristics. Implementing the BGP SendHoldTimer as specified in this document will enhance network resilience by terminating connections with malfunctioning or overwhelmed remote peers.¶
IANA has registered value 8 for "Send Hold Timer Expired" in the "BGP Error (Notification) Codes" registry within the "Border Gateway Protocol (BGP) Parameters" registry group.¶
The authors would like to thank William McCall, Theo de Raadt, John Heasley, Nick Hilliard, Jeffrey Haas, Tom Petch, Susan Hares, Keyur Patel, Ben Maddison, Claudio Jeker, and John Scudder for their helpful review of this document.¶