bgp.session.md

(plugin-bgp-session)=

BGP Sessions Plugin

The bgp.session plugin (initially contributed by Stefano Sasso) implements numerous BGP session nerd knobs attributes, from BGP session protection to AS path manipulation.

.. contents:: Table of Contents
   :depth: 2
   :local:
   :backlinks: none

Supported BGP Attributes

The plugin adds the following BGP session attributes:

• bgp.allowas_in is an attribute that controls whether a BGP router accepts one or more copies of its own AS number in incoming update. It takes an integer value between 1 and 10. A true value sets it to 1.

• bgp.as_override is a boolean attribute that controls whether a BGP router replaces peer AS with its own AS.

• bgp.bfd is a boolean attribute that enables BFD with BGP neighbors.

• bgp.default_originate is a boolean attribute that controls whether a BGP router advertises a default route to its neighbor(s).

• bgp.gtsm is an integer attribute that enables the Generic TTL Security Mechanism (GTSM). A true value sets it to 1 (the remote router can be up to one hop away).

• bgp.passive is a boolean attribute that makes a node act as a passive BGP peer¹ on an interface or on all applicable sessions (when specified as a node parameter).

• bgp.password is a string attribute that specifies the MD5 or TCP-AO password used on EBGP sessions.

• bgp.remove_private_as is a boolean/string/list attribute that describes the desired removal of private autonomous system(s) from the AS path. See section for more details.

• bgp.rs is a boolean attribute specifying that a BGP router is a BGP route server. See section for more details.

• bgp.tcp_ao is an attribute that enables TCP-AO on a BGP session. The attribute value true enables TCP-AO with the HMAC-SHA1-96 algorithm; you can specify the desired algorithm as a string value of bgp.tcp_ao parameter.

• bgp.timers is a dictionary of BGP session timers. It has three elements:

  • bgp.timers.keepalive -- keepalive timer in seconds
  • bgp.timers.hold -- hold timer in seconds
  • bgp.timers.min_hold -- minimum hold timer accepted from the remote node. It is used only on devices where the minimum hold timer can be specified per BGP neighbor.

BGP session attributes can be specified at the global, node, link, or interface (node-to-link attachment) level. The following table describes where you could apply individual attributes:

BGP session attribute	Global	Node	Link	Interface
allowas_in	❌	❌	❌	✅
as_override	❌	❌	❌	✅
bfd	✅	✅	✅	✅
default_originate	❌	✅	❌	✅
gtsm	✅	✅	✅	✅
passive	❌	✅	❌	✅
password	✅	✅	✅	✅
remove_private_as	❌	❌	❌	✅
rs	❌	✅	❌	✅
tcp_ao	✅	✅	✅	✅
timers	✅	✅	✅	✅

Platform Support

The plugin implements generic BGP session features for the following platforms:

(bgp-session-platforms)=

Operating system	default originate	BGP timers	BFD	Passive peer
Arista EOS	✅	✅	✅	✅
Aruba AOS-CX	✅	✅	✅	❌
Cisco IOSv/IOSvL2	✅	✅	✅	✅
Cisco IOS-XE2	✅	✅	✅	✅
Cisco Nexus OS	✅	✅	❌	✅
Cumulus Linux	✅	✅	✅	✅
FRR	✅	✅	✅	✅
Junos3	❌	✅	✅	✅
Mikrotik RouterOS 7	✅	❌	❌	❌
Nokia SR Linux	✅	✅	✅	✅
Nokia SR OS	✅	❌	❌	❌
VyOS	✅	❌	❌	❌

Notes:

• Arista EOS supports TCP-AO only when running as a virtual machine
• netlab always configures HMAC-SHA1-96 as the cryptographic algorithm on IOS XE

(bgp-session-security)= BGP session security features are available on these platforms:

Operating system	password	GTSM	TCP-AO
Arista EOS	✅	✅	✅
Aruba AOS-CX	✅	✅	❌
Cisco IOSv/IOSvL2	✅	✅	❌
Cisco IOS-XE2	✅	✅	✅
Cisco Nexus OS	✅	✅	❌
Cumulus Linux	✅	✅	❌
FRR	✅	✅	❌
Junos3	✅	❌	❌
Mikrotik RouterOS 7	✅	❌	❌
Nokia SR Linux	✅	❌	❌
Nokia SR OS	✅	❌	✅

BGP session security features are also available on these daemons:

Operating system	password	GTSM	TCP-AO
BIRD	✅	❌	❌

(bgp-session-as-path)= The plugin implements AS-path-mangling nerd knobs for the following platforms:

Operating system	Allow AS in	AS override	Remove private AS	Route Server	Route Server Client
Arista EOS	✅	✅	✅	❌	✅
Aruba AOS-CX	✅	❌	✅	❌	❌
Bird	❌	❌	❌	✅	✅
Cisco IOSv/IOSvL2	✅	✅	✅	✅	✅
Cisco IOS-XE2	✅	✅	✅	✅	✅
Cisco Nexus OS	✅	✅	✅	❌	❌
Cumulus Linux	✅	✅	✅	✅	✅
FRR	✅	✅	✅	✅	✅
Junos3	❌	✅	❌	❌	❌
Mikrotik RouterOS 7	✅	✅	❌	❌	❌
Nokia SR Linux	✅	✅	✅	❌	✅
Nokia SR OS	✅	✅	❌	❌	❌
VyOS	✅	✅	❌	❌	❌

(bgp-session-apply)=

Applying BGP Session Attributes to IBGP Sessions

The plugin applies BGP session attributes to EBGP sessions -- either all EBGP sessions a node has if the attribute has been specified on a global or node level, or all EBGP sessions on a link where the BGP session attribute has been specified.

If you want to apply BGP session attributes to IBGP sessions, you must set the bgp.session.apply global/node parameter -- a list or a dictionary of BGP neighbor types to which the BGP session attributes are applied. The default behavior is equivalent to bgp.session.apply: [ ebgp ]; if you want to apply the session attributes to IBGP and EBGP sessions, specify bgp.session.apply: [ ibgp, ebgp ] as a global or node attribute.

IBGP sessions are not associated with links/interfaces and, therefore, cannot inherit the BGP session parameters specified on links or interfaces -- only the parameters specified on the global or node level are applied to IBGP sessions. For example, to use MD5 password secret on all IBGP and EBGP sessions in your lab, use the following global settings:

bgp.session.apply: [ ibgp, ebgp ]
bgp.password: secret

If you don't want to apply all node-level BGP session parameters to IBGP sessions, you could use the dictionary format of bgp.session.apply parameter: the BGP neighbor types are the dictionary keys, and the values are lists of parameters you want to apply to IBGP and EBGP sessions (a missing value means apply all parameters).

For example, the node settings from the following example cause the BGP password to be applied to IBGP and EBGP sessions, while all other BGP session parameters (including the default route origination) are used only on EBGP sessions:

bgp.password: secret

nodes:
  core:
    bgp.default_originate: True
    bgp.session.apply:
      ibgp: [ password ]
      ebgp:

The following example is even more specific: default route origination and a few AS-related session parameters (but not MD5 password) are applied to EBGP sessions, while the MD5 passwords are applied only to IBGP sessions:

bgp.password: secret
bgp.session.apply:
  ibgp: [ password ]
  ebgp: [ default_originate, allowas_in, as_override ]

(bgp-session-remove-private-as)=

Remove Private AS Options

This plugin accepts the following values of the bgp.remove_private_as attribute.

• True (or on⁴): enable simple removal of private AS on egress updates
• all: remove all instances of private AS numbers in egress updates
• replace: replace all instances of private AS number with the router's own AS number in egress updates
• ingress: remove private AS numbers from ingress updates
• ingress-replace: replace all private AS numbers with the peer AS number in ingress updates

You can specify the bgp.remove_private_as attribute as a boolean value (equivalent to on), as a single string value, or as a list of string values in case you have to specify egress and ingress behavior.

The implementations of the neighbor remove-private-as command vary widely across supported network devices, as described in the following table:

Operating system	on	all	replace	ingress	ingress-replace
Arista EOS	✅	✅	✅	✅	✅
Cisco IOSv/IOSvL2	✅	❗	❗	❌	❌
Cisco IOS-XE	✅	❗	❗	❌	❌
Cumulus Linux	✅	✅	✅	❌	❌
FRR	✅	✅	✅	❌	❌
Nokia SR Linux	✅	✅	✅	❌	❌

Notes:

• Cisco IOSv/IOS-XE do not accept all or replace option when the local BGP AS number is private.

(bgp-session-route-server)=

BGP Route Servers

The bgp.rs attribute set on a node or on an interface indicates that the node acts like a BGP route server on the specified interface or all interfaces.

A route server does not add its BGP AS into the AS path; its EBGP neighbors must accept AS paths with 'incorrect' first AS. Most devices reject what seem to be malformed AS paths and have to be configured to ignore the mismatch in the first AS. netlab configures that setting on all route server EBGP neighbors and rejects the lab topology if a route server EBGP neighbor is not capable of being a route server client.

Introducing a BGP route server to an IP subnet also prunes the full mesh of EBGP sessions established over that subnet. netlab removes all EBGP sessions from the subnet unless one of the EBGP neighbors is a route server and the other neighbor is a route server client. The following table summarizes the EBGP session pruning:

Node A	Node B	EBGP session state
Route server	Regular BGP router	retained
Route server	Route server	removed
Regular BGP router	Regular BGP route	removed

Please note that an IBGP session between route servers in the same autonomous system is retained and might generate a warning unless you also run an IGP between the route servers. To remove the IBGP session between route servers, set the bgp.sessions.ipv4: [ ebgp ] and bgp.sessions.ipv6: [ ebgp ] node attributes on route server nodes.

Test Topology

The following test topology illustrates a subset of plugin capabilities. You might also want to explore the topologies used in ipSpace.net BGP labs and the BGP session plugin test topology.

---
defaults:
  device: eos

module: [ bgp, vrf, bfd ]
plugin: [ bgp.session ]

vrfs:
  red:
  blue:

bgp.timers:
  hold: 10
  keepalive: 3

bgp.bfd: True

nodes:
  y1:
    bgp.as: 65001
  y2:
    bgp.as: 65002

links:
- y1:
    bgp.default_originate: True
  y2:
  bgp.password: TestPassword
- y1:
    vrf: red
  y2:
    vrf: red
    bgp.allowas_in: True
    bgp.default_originate: True
- y1:
    vrf: blue
  y2:
    vrf: blue
    bgp.as_override: True

Footnotes

1. A passive BGP peer waits for its neighbor to initiate the conversation and never tries to establish a TCP session with its neighbor. ↩

2. Includes Cisco CSR 1000v, Cisco Catalyst 8000v, Cisco IOS-on-Linux (IOL), and IOL Layer-2 image. ↩ ↩² ↩³

3. Includes vMX, vSRX, vPTX, vJunos-switch, and vJunos-router ↩ ↩² ↩³

4. You have to quote the on string, otherwise the Python YAML parser treats it as a boolean. For example, you have to specify bgp.remove_private_as: [ 'on', ingress ] ↩