BGP is the routing protocol that runs the Internet. It is an increasingly popular protocol for use in the data center as it lends itself well to the rich interconnections in a Clos topology. Specifically, BGP:
- Does not require the routing state to be periodically refreshed, unlike OSPF.
- Is less chatty than its link-state siblings. For example, a link or node transition can result in a bestpath change, causing BGP to send updates.
- Is multi-protocol and extensible.
- Has many robust vendor implementations.
- Is very mature as a protocol and comes with many years of operational experience.
RFC 7938 provides further details of the use of BGP within the data center.
Autonomous System Number (ASN)
One of the key concepts in BGP is an autonomous system number or ASN. An autonomous system is defined as a set of routers under a common administration. Because BGP was originally designed to peer between independently managed enterprises and/or service providers, each such enterprise is treated as an autonomous system, responsible for a set of network addresses. Each such autonomous system is given a unique number called its ASN. ASNs are handed out by a central authority (ICANN). However, ASNs between 64512 and 65535 are reserved for private use. Using BGP within the data center relies on either using this number space or using the single ASN you own.
The ASN is central to how BGP builds a forwarding topology. A BGP route advertisement carries with it not only the originator’s ASN, but also the list of ASNs that this route advertisement has passed through. When forwarding a route advertisement, a BGP speaker adds itself to this list. This list of ASNs is called the AS path. BGP uses the AS path to detect and avoid loops.
ASNs were originally 16-bit numbers, but were later modified to be 32-bit. FRRouting supports both 16-bit and 32-bit ASNs, but most implementations still run with 16-bit ASNs.
eBGP and iBGP
When BGP is used to peer between autonomous systems, the peering is referred to as external BGP or eBGP. When BGP is used within an autonomous system, the peering used is referred to as internal BGP or iBGP. eBGP peers have different ASNs while iBGP peers have the same ASN.
The heart of the protocol is the same when used as eBGP or iBGP, however, there is a key difference in the protocol behavior between use as eBGP and iBGP: an iBGP speaker does not forward routing information learned from one iBGP peer to another iBGP peer to prevent loops. eBGP prevents loops using the AS_Path attribute.
All iBGP speakers need to be peered with each other in a full mesh. In a large network, this requirement can quickly become unscalable. The most popular method to scale iBGP networks is to introduce a route reflector.
In a two-tier Clos network, the leaf (or tier 1) switches are the only ones connected to end stations. The spines themselves do not have any routes to announce; they are merely reflecting the routes announced by one leaf to the other leaves. Therefore, the spine switches function as route reflectors while the leaf switches serve as route reflector clients.
In a three-tier network, the tier 2 nodes (or mid-tier spines) act as both route reflector servers and route reflector clients. They act as route reflectors because they announce the routes learned from the tier 1 nodes to other tier 1 nodes and to tier 3 nodes. They also act as route reflector clients to the tier 3 nodes, receiving routes learned from other tier 2 nodes. Tier 3 nodes act only as route reflectors.
In the following illustration, tier 2 node 2.1 is acting as a route reflector server, announcing the routes between tier 1 nodes 1.1 and 1.2 to tier 1 node 1.3. It is also a route reflector client, learning the routes between tier 2 nodes 2.2 and 2.3 from the tier 3 node, 3.1.
Configuring route-reflector-client Requires Specific Order
When configuring a router to be a route reflector client, you must specify the FRRouting configuration in a specific order; otherwise, the router will not be a route reflector client.
You must run the
net add bgp neighbor <IPv4/IPV6> route-reflector-client command after the
net add bgp neighbor <IPV4/IPV6> activate command; otherwise, the
route-reflector-client command is ignored. For example:
A cluster consists of route reflectors (RRs) and their clients and is used in iBGP environments where multiple sets of route reflectors and their clients are configured. Configuring a unique ID per cluster (on the route reflector server and clients) prevents looping as a route reflector does not accept routes from another that has the same cluster ID. Additionally, because all route reflectors in the cluster recognize updates from peers in the same cluster, they do not install routes from a route reflector in the same cluster; this reduces the number of updates that need to be stored in BGP routing tables.
To configure a cluster ID on a route reflector, run the
net add bgp cluster-id (<ipv4>|<1-4294967295>) command. You can enter the cluster ID as an IP address or as a 32-bit quantity.
The following example configures a cluster ID on a route reflector in IP address format:
The following example configures a cluster ID on a route reflector as a 32-bit quantity:
If a BGP node hears a prefix p from multiple peers, it has all the information necessary to program the routing table to forward traffic for that prefix p through all of these peers; BGP supports equal-cost multipathing (ECMP).
To perform ECMP in BGP, you may need to configure
net add bgp bestpath as-path multipath-relax (if you are using eBGP).
In Cumulus Linux, the BGP
maximum-paths setting is enabled by default, so multiple routes are already installed. The default setting is 64 paths.
BGP for Both IPv4 and IPv6
Unlike OSPF, which has separate versions of the protocol to announce IPv4 and IPv6 routes, BGP is a multi-protocol routing engine, capable of announcing both IPv4 and IPv6 prefixes. It supports announcing IPv4 prefixes over an IPv4 session and IPv6 prefixes over an IPv6 session. It also supports announcing prefixes of both these address families over a single IPv4 session or over a single IPv6 session.
The following example shows a basic BGP configuration. The rest of this chapter discusses how to configure other BGP features, such as unnumbered interfaces to route maps.
Enable the BGP and Zebra daemons (
bgpd), then enable the FRRouting service and start it, as described in Configuring FRRouting.
Identify the BGP node by assigning an ASN and
Specify where to disseminate routing information:
For an iBGP session, the
remote-asis the same as the local AS:
Specifying the IP address of the peer allows BGP to set up a TCP socket with this peer, but it does not distribute any prefixes to it, unless it is explicitly told that it must with the
activatecommand. You must specify the
activatecommand for each address family that is being announced by the BGP session.
Specify BGP session properties:
If this is a route reflector client, it can be specified as follows:It is node switchRR, the route reflector, on which the peer is specified as a client.
Specify which prefixes to originate:
BGP Unnumbered Interfaces
Unnumbered interfaces are interfaces without unique IP addresses. In BGP, you configure unnumbered interfaces using extended next-hop encoding (ENHE), which is defined by RFC 5549. BGP unnumbered interfaces provide a means of advertising an IPv4 route with an IPv6 next-hop. Prior to RFC 5549, an IPv4 route could be advertised only with an IPv4 next-hop.
BGP unnumbered interfaces are particularly useful in deployments where IPv4 prefixes are advertised through BGP over a section without any IPv4 address configuration on links. As a result, the routing entries are also IPv4 for destination lookup and have IPv6 next-hops for forwarding purposes.
BGP and Extended Next-hop Encoding
When enabled and active, BGP makes use of the available IPv6 next-hops for advertising any IPv4 prefixes. BGP learns the prefixes, calculates the routes and installs them in IPv4 AFI to IPv6 AFI format. However, ENHE in Cumulus Linux does not install routes into the kernel in IPv4 prefix to IPv6 next-hop format. For link-local peerings enabled by dynamically learning the other end's link-local address using IPv6 neighbor discovery router advertisements, an IPv6 next-hop is converted into an IPv4 link-local address and a static neighbor entry is installed for this IPv4 link-local address with the MAC address derived from the link-local address of the other end.
Cumulus Linux 3.7.2 and later also supports advertising IPv4 prefixes with IPv6 next-hop addresses while peering over IPv6 global unicast addresses. See RFC 5549 Support with Global IPv6 Peers below.
Configure BGP Unnumbered Interfaces
To configure a BGP unnumbered interface, you must enable IPv6 neighbor discovery router advertisements. The
interval you specify is measured in seconds and defaults to 10 seconds.
In Cumulus Linux 3.7.1 and earlier, extended next-hop encoding is sent only for the link-local address peerings (as shown below). In Cumulus Linux 3.7.2 and later, extended next-hop encoding can be sent for the both link-local and global unicast address peerings (see RFC 5549 Support with Global IPv6 Peers).
These commands create the following configuration in the
For an unnumbered configuration, you can use a single command to configure a neighbor and attach it to a peer group (making sure to substitute for the interface and peer group below):
Manage Unnumbered Interfaces
All the relevant BGP commands show IPv6 next-hops and/or the interface name for any IPv4 prefix:
FRRouting RIB commands are also modified:
The following commands show how the IPv4 link-local address 169.254.0.1 is used to install the route and static neighbor entry to facilitate proper forwarding without having to install an IPv4 prefix with IPv6 next-hop in the kernel:
You can use the following command to display more neighbor information:
How traceroute Interacts with BGP Unnumbered Interfaces
Every router or end host must have an IPv4 address to complete a
traceroute of IPv4 addresses. In this case, the IPv4 address used is that of the loopback device.
Even if ENHE is not used in the data center, link addresses are not typically advertised. This is because:
- Link addresses take up valuable FIB resources. In a large Clos environment, the number of such addresses can be quite large.
- Link addresses expose an additional attack vector for intruders to use to either break in or engage in DDOS attacks.
Assigning an IP address to the loopback device is essential.
Advanced: How Next-hop Fields Are Set
This section describes how the IPv6 next-hops are set in the MP_REACH_NLRI (multiprotocol reachable NLRI) initiated by the system, which applies whether IPv6 prefixes or IPv4 prefixes are exchanged with ENHE. There are two main aspects to determine — how many IPv6 next-hops are included in the MP_REACH_NLRI (since the RFC allows either one or two next-hops) and the values of the next-hop(s). This section also describes how a received MP_REACH_NLRI is handled as far as processing IPv6 next-hops.
- Whether peering to a global IPv6 address or link-local IPv6 address, the determination whether to send one or two next-hops is as follows:
- If reflecting the route, two next-hops are sent only if the peer has
nexthop-local unchangedconfigured and the attribute of the received route has an IPv6 link-local next-hop; otherwise, only one next-hop is sent.
- Otherwise (if it is not reflecting the route), two next-hops are sent if explicitly configured (
nexthop-local unchanged) or the peer is directly connected (that is, either peering is on link-local address or the global IPv4 or IPv6 address is directly connected) and the route is either a local/self-originated route or the peer is an eBGP peer.
- In all other cases, only one next-hop gets sent, unless an outbound route map adds another next-hop.
- If reflecting the route, two next-hops are sent only if the peer has
route-mapcan impose two next-hops in scenarios where Cumulus Linux only sends one next-hop — by specifying
set ipv6 nexthop link-local.
- For all routes to eBGP peers and self-originated routes to iBGP peers, the global next-hop (first value) is the peering address of the local system. If the peering is on the link-local address, this is the global IPv6 address on the peering interface, if present; otherwise, it is the link-local IPv6 address on the peering interface.
For other routes to iBGP peers (eBGP to iBGP or reflected), the global next-hop will be the global next-hop in the received attribute.
If this address is a link-local IPv6 address, it is reset so that the link-local IPv6 address of the eBGP peer is not passed along to an iBGP peer, which most likely is on a different link.
route-mapand/or the peer configuration can change the above behavior. For example,
route-mapcan set the global IPv6 next-hop or the peer configuration can set it to self — which is relevant for iBGP peers. The route map or peer configuration can also set the next-hop to unchanged, which ensures the source IPv6 global next-hop is passed around — which is relevant for eBGP peers.
- Whenever two next-hops are being sent, the link-local next-hop (the second value of the two) is the link-local IPv6 address on the peering interface unless it is due to
route-maphas set the link-local next-hop.
- Network administrators cannot set martian values for IPv6 next-hops in
route-map. Also, global and link-local next-hops are validated to ensure they match the respective address types.
- In a received update, a martian check is imposed for the IPv6 global next-hop. If the check fails, it gets treated as an implicit withdraw.
- If two next-hops are received in an update and the second next-hop is not a link-local address, it gets ignored and the update is treated as if only one next-hop was received.
- Whenever two next-hops are received in an update, the second next-hop is used to install the route into
zebra. As per the previous point, it is already assured that this is a link-local IPv6 address. Currently, this is assumed to be reachable and is not registered with NHT.
route-mapspecifies the next-hop as
peer-address, the global IPv6 next-hop as well as the link-local IPv6 next-hop (if it's being sent) is set to the peering address. If the peering is on a link-local address, the former could be the link-local address on the peering interface, unless there is a global IPv6 address present on this interface.
- When using iBGP unnumbered with IPv6 Link Local Addresses (the default), FRR rewrites the BGP next hop to be the adjacent link. This is similar behavior to eBGP next hops. However, iBGP route advertisement rules do not change and a full mesh or route reflectors is still required.
The above rules imply that there are scenarios where a generated update has two IPv6 next-hops, and both of them are the IPv6 link-local address of the peering interface on the local system. If you are peering with a switch or router that is not running Cumulus Linux and expects the first next-hop to be a global IPv6 address, a route map can be used on the sender to specify a global IPv6 address. This conforms with the recommendations in the Internet draft draft-kato-bgp-ipv6-link-local-00.txt, "BGP4+ Peering Using IPv6 Link-local Address".
- Interface-based peering with separate IPv4 and IPv6 sessions is not supported.
- In Cumulus Linux 3.7.1 and earlier, ENHE is sent for IPv6 link-local peerings only. In Cumulus Linux 3.7.2 and later, ENHE can also be also sent for IPv6 GUA peerings (see below).
- If an IPv4 /30 or /31 IP address is assigned to the interface, IPv4 peering is used over IPv6 link-local peering.
If the default router lifetime in the generated IPv6 route advertisements (RA) is set to 0, the receiving FRRouting instance drops the RA if it is on a Cumulus Linux 2.5.z switch. To work around this issue, either:
- Explicitly configure the switch to advertise a router lifetime of 0, unless a value is specifically set by the operator — with the assumption that the host is running Cumulus Linux 3.y.z version of FRRouting. When hosts see an IPv6 RA with a router lifetime of 0, they do not make that router a default router.
- Use the
sysctlon the host —
net.ipv6.conf.all.accept_ra_defrtr. However, this requires applying this setting on all hosts, which might mean many hosts, especially if FRRouting is run on the hosts.
RFC 5549 Support with Global IPv6 Peers (Cumulus Linux 3.7.2 and later)
RFC 5549 defines the method used for BGP to advertise IPv4 prefixes with IPv6 next-hops. The RFC does not make a distinction between whether the IPv6 peering and next-hop values should be global unicast addresses (GUA) or link-local addresses. Cumulus Linux 3.7.1 and earlier only supports advertising IPv4 prefixes using link-local IPv6 next-hop addresses via BGP unnumbered peering. Cumulus Linux 3.7.2 supports advertising IPv4 prefixes with IPv6 global unicast and link-local next-hop addresses, with either unnumbered or numbered BGP.
When BGP peering uses IPv6 global addresses and IPv4 prefixes are being advertised and installed, IPv6 route advertisements are used to derive the MAC address of the peer so that FRR can create an IPv4 route with a link-local IPv4 next-hop address (defined by RFC 3927). This is required to install the route into the kernel. These route advertisement settings are configured automatically when FRR receives an update from a BGP peer using IPv6 global addresses that contain an IPv4 prefix with an IPv6 nexthop, and the enhanced-nexthop capability has been negotiated.
Configure RFC 5549 Support with Global IPv6 Peers
To enable advertisement of IPv4 prefixes with IPv6 next-hops over global IPv6 peerings, add the
extended-nexthop capability to the global IPv6 neighbor statements on each end of the BGP sessions.
The above commands create the following configuration in the
Ensure that the IPv6 peers are activated under the IPv4 unicast address family; otherwise, all peers are activated in the IPv4 unicast address family by default.
no bgp default ipv4-unicast is configured, you need to explicitly activate the IPv6 neighbor under the IPv4 unicast address family as shown below:
The above commands create the following configuration in the
Show IPv4 Prefixes Learned with IPv6 Next-Hops
To show IPv4 prefixes learned with IPv6 next-hops, you can run
net show bgp ipv4 unicast commands.
Cumulus Linux supports both BGP add-path RX and BGP add-path TX.
BGP add-path RX
BGP add-path RX allows BGP to receive multiple paths for the same prefix. A path identifier is used so that additional paths do not override previously advertised paths. No additional configuration is required for BGP add-path RX.
To view the existing capabilities, run
net show bgp neighbor. The existing capabilities are listed in the subsection Add Path, below Neighbor capabilities:
The example output above shows that additional BGP paths can be sent and received (TX and RX are advertised). It also shows that the BGP neighbor, fe80::4638:39ff:fe00:5c, supports both.
To view the current additional paths, run
net show bgp <network>. The example output shows an additional path that has been added by the TX node for receiving. Each path has a unique AddPath ID.
BGP add-path TX
AddPath TX allows BGP to advertise more than just the bestpath for a prefix. Consider the following topology:
In this topology:
- r1 and r2 are in AS 100
- r3 and r4 are in AS 300
- r5 and r6 are in AS 500
- r7 is in AS 700
- r8 is in AS 800
- r7 learns 22.214.171.124/32 from r1, r2, r3, r4, r5, and r6. Among these r7 picks the path from r1 as the bestpath for 126.96.36.199/32
The example below configures the r7 session to advertise the bestpath learned from each AS. In this case, this means a path from AS 100, a path from AS 300, and a path from AS 500. The
net show bgp 188.8.131.52/32 from r7 has "bestpath-from-AS 100" so the user can see what the bestpath is from each AS:
The output below shows the result on r8:
The example below shows the results if r7 is configured to advertise all paths to r8:
The output below shows the result on r8:
Fast Convergence Design Considerations
Cumulus Networks strongly recommends the following use of addresses in the design of a BGP-based data center network:
- Set up BGP sessions only using interface-scoped addresses. This allows BGP to react quickly to link failures.
- Use of next-hop-self: Every BGP node says that it knows how to forward traffic to the prefixes it is announcing. This reduces the requirement to announce interface-specific addresses and thereby reduces the size of the forwarding table.
When you configure BGP for the neighbors of a given interface, you can specify the interface name instead of its IP address. All the other
neighbor command options remain the same.
This is equivalent to BGP peering to the link-local IPv6 address of the neighbor on the given interface. The link-local address is learned via IPv6 neighbor discovery router advertisements.
Consider the following example configuration in the
You create the above configuration with the following NCLU commands:
By default, Cumulus Linux sends IPv6 neighbor discovery router advertisements. Cumulus Networks recommends you adjust the interval of the router advertisement to a shorter value (
net add interface <interface> ipv6 nd ra-interval <interval>) to address scenarios when nodes come up and miss router advertisement processing to relay the neighbor’s link-local address to BGP. The
interval is measured in seconds and defaults to 10 seconds.
Peer Groups to Simplify Configuration
When a switch has many peers to connect to, the amount of redundant configuration becomes overwhelming. For example, repeating the
next-hop-self commands for even 60 neighbors makes for a very long configuration file. To address this problem, you can use
Instead of specifying properties of each individual peer, FRRouting allows you to define one or more peer groups and associate all the attributes common to that peer session to a peer group. A peer needs to be attached to a peer group only once, when it then inherits all address families activated for that peer group.
After you attach a peer to a peer group, you need to associate an IP address with the peer group. The following example shows how to define and use peer groups:
BGP peer-group restrictions have been replaced with update-groups, which dynamically examine all peers and group them if they have the same outbound policy.
Configure BGP Dynamic Neighbors
BGP dynamic neighbor provides BGP peering to a group of remote neighbors within a specified range of IPv4 or IPv6 addresses for a BGP peer group. You can configure each range as a subnet IP address.
You configure dynamic neighbors using the
bgp listen range <IP address> peer-group <GROUP> command. After you configure the dynamic neighbors, a BGP speaker can listen for, and form peer relationships with, any neighbor in the IP address range and mapped to a peer group.
To limit the number of dynamic peers, specify the limit in the
bgp listen limit command (the default value is 100):
Collectively, a sample configuration for IPv4 looks like this:
These commands produce an IPv4 configuration that looks like this:
Configure BGP Peering Relationships across Switches
A BGP peering relationship is typically initiated with the
neighbor x.x.x.x remote-as [internal|external] command.
Specifying internal signifies an iBGP peering; that is, the neighbor only creates or accepts a connection with the specified neighbor if the remote peer AS number matches this BGP AS number.
Specifying external signifies an eBGP peering; that is, the neighbor will only create a connection with the neighbor if the remote peer AS number does not match this BGP AS number.
You can make this distinction using the
neighbor command or the
In general, use the following syntax with the
Some example configurations follow.
Configure MD5-enabled BGP Neighbors
The following sections outline how to configure an MD5-enabled BGP neighbor. Each process assumes that FRRouting is used as the routing platform, and consists of two switches (
AS 65011 and
AS 65020), connected by the link 10.0.0.100/30, with the following configurations:
To manually configure an MD5-enabled BGP neighbor:
SSH into leaf01.
Configure the password for the neighbor:
Confirm the configuration has been implemented with the
net show bgp summarycommand:
SSH into spine01.
Configure the password for the neighbor:
Confirm the configuration has been implemented with the
net show bgp summarycommand:
The MD5 password configured against a BGP listen-range peer-group (used to accept and create dynamic BGP neighbors) is not enforced. This means that connections are accepted from peers that do not specify a password.
Configure eBGP Multihop
The eBGP Multihop option lets you use BGP to exchange routes with an external peer that is more than one hop away.
To establish a connection between two eBGP peers that are not directly connected:
Confirm the configuration with the
net show bgp neighbor <ip>command:
cumulus@leaf02:mgmt-vrf:~$ net show bgp neighbor 10.0.0.11 BGP neighbor is 10.0.0.11, remote AS 65011, local AS 65012, external link Hostname: leaf01 BGP version 4, remote router ID 10.0.0.11 BGP state = Established, up for 00:02:54 Last read 00:00:00, Last write 00:00:00 Hold time is 9, keepalive interval is 3 seconds Neighbor capabilities: 4 Byte AS: advertised and received AddPath: IPv4 Unicast: RX advertised IPv4 Unicast and received Route refresh: advertised and received(old & new) Address Family IPv4 Unicast: advertised and received Hostname Capability: advertised (name: leaf02,domain name: n/a) received (name: leaf01,domain name: n/a) Graceful Restart Capability: advertised and received Remote Restart timer is 120 seconds Address families by peer: none Graceful restart informations: End-of-RIB send: IPv4 Unicast End-of-RIB received: IPv4 Unicast Message statistics: Inq depth is 0 Outq depth is 0 Sent Rcvd Opens: 1 1 Notifications: 0 0 Updates: 2868 2872 Keepalives: 60 60 Route Refresh: 0 0 Capability: 0 0 Total: 2929 2933 Minimum time between advertisement runs is 0 seconds For address family: IPv4 Unicast Update group 2, subgroup 4 Packet Queue length 0 Community attribute sent to this neighbor(all) 9 accepted prefixes Connections established 1; dropped 0 Last reset never External BGP neighbor may be up to 255 hops away. Local host: 10.0.0.12, Local port: 40135 Foreign host: 10.0.0.11, Foreign port: 179 Nexthop: 10.0.0.12 Nexthop global: :: Nexthop local: :: BGP connection: non shared network BGP Connect Retry Timer in Seconds: 10 Estimated round trip time: 1 ms Read thread: on Write thread: on
Configure BGP TTL Security
The steps below show how to configure BGP TTL security on Cumulus Linux using a leaf (
leaf01) and spine (
spine01) for the example output:
SSH into leaf01 and configure it for TTL security:
SSH into spine01 and configure it for TTL security:
Confirm the configuration with the
show ip bgp neighborcommand:
cumulus@spine01:mgmt-vrf:~$ net show bgp neighbor swp1 BGP neighbor on swp1: fe80::4638:39ff:fe00:5b, remote AS 65011, local AS 65020, external link Hostname: leaf01 BGP version 4, remote router ID 10.0.0.11 BGP state = Established, up for 00:10:45 Last read 00:00:03, Last write 00:00:03 Hold time is 9, keepalive interval is 3 seconds Neighbor capabilities: 4 Byte AS: advertised and received AddPath: IPv4 Unicast: RX advertised IPv4 Unicast and received Extended nexthop: advertised and received Address families by peer: IPv4 Unicast Route refresh: advertised and received(old & new) Address Family IPv4 Unicast: advertised and received Hostname Capability: advertised (name: spine01,domain name: n/a) received (name: leaf01,domain name: n/a) Graceful Restart Capabilty: advertised and received Remote Restart timer is 120 seconds Address families by peer: none Graceful restart informations: End-of-RIB send: IPv4 Unicast End-of-RIB received: IPv4 Unicast Message statistics: Inq depth is 0 Outq depth is 0 Sent Rcvd Opens: 46 2 Notifications: 41 0 Updates: 38 34 Keepalives: 49334 49331 Route Refresh: 0 0 Capability: 0 0 Total: 49459 49367 Minimum time between advertisement runs is 0 seconds For address family: IPv4 Unicast Update group 1, subgroup 1 Packet Queue length 0 Community attribute sent to this neighbor(all) 3 accepted prefixes Connections established 2; dropped 1 Last reset 00:17:37, due to NOTIFICATION sent (Hold Timer Expired) External BGP neighbor may be up to 1 hops away. Local host: fe80::4638:39ff:fe00:5c, Local port: 35564 Foreign host: fe80::4638:39ff:fe00:5b, Foreign port: 179 Nexthop: 10.0.0.21 Nexthop global: fe80::4638:39ff:fe00:5c Nexthop local: fe80::4638:39ff:fe00:5c BGP connection: shared network BGP Connect Retry Timer in Seconds: 10 Read thread: on Write thread: on
Configure Graceful BGP Shutdown
To reduce packet loss during planned maintenance of a router or link, you can configure graceful BGP shutdown, which forces traffic to route around the node.
To configure graceful BGP shutdown for the current node, run the
net add bgp graceful-shutdown command:
When configured, the
graceful-shutdown community is added to all paths from eBGP peers and the
local-pref for that route is set to 0. An example configuration is shown below:
To disable graceful shutdown for the current node, run the
net del bgp graceful-shutdown command:
BGP Advertisement Best Practices
Limiting the exchange of routing information at various parts in the network is a best practice you should follow. The following image illustrates one way you can do so in a typical Clos architecture:
Multiple Routing Tables and Forwarding
You can run multiple routing tables (one for in-band/data plane traffic and one for out-of-band/management plane traffic) on the same switch using management VRF (multiple routing tables and forwarding).
BGP and static routing (IPv4 and IPv6) are supported within a VRF context. For more information, refer to Virtual Routing and Forwarding - VRF.
BGP Community Lists
You can use community lists to define a BGP community to tag one or more routes. You can then use the communities to apply route policy on either egress or ingress.
The BGP community list can be either standard or expanded. The standard BGP community list is a pair of values (such as 100:100) that can be tagged on a specific prefix and advertised to other neighbors or applied on route ingress. Alternately, it can be one of four BGP default communities:
- internet: a BGP community that matches all routes
- local-AS: a BGP community that restrict routes to your confederation's sub-AS
- no-advertise: a BGP community that isn't advertised to anyone
- no-export: a BGP community that isn't advertised to the eBGP peer
An expanded BGP community list takes a regular expression of communities matches the listed communities.
When the neighbor receives the prefix, it examines the community value and takes action accordingly, such as permitting or denying the community member in the routing policy.
Here is an example of a standard community list filter:
You can apply the community list to a route map to define the routing policy:
Additional Default Settings
Other settings not discussed in detail in this chapter that are enabled by default, include the following:
bgp deterministic-med, which ensures path ordering no longer impacts bestpath selection.
bgp show-hostname, which displays the hostname in show command output.
bgp network import-check, which enables the advertising of the BGP network in IGP.
Configure BGP Neighbor maximum-prefixes
The maximum number of route announcements, or prefixes, allowed by a BGP neighbor can be configured using the FRR
To troubleshoot BGP, you can view the summary of neighbors to which the switch is connected and see information about these connections. The following example shows sample command output:
To determine if the sessions above are iBGP or eBGP sessions, look at the ASNs.
It is also useful to view the routing table as defined by BGP:
To show a more detailed breakdown of a specific neighbor, run the
net show bgp neighbor <neighbor> command:
To see details of a specific route, such as from where it is received and where it is sent, run the
net show bgp <ip address/prefix> command:
The above example shows that the routing table prefix seen by BGP is 10.0.0.11/32, that this route is advertised to two neighbors, and that it is not heard by any neighbors.
Log Neighbor State Changes
To log the changes that a neighbor goes through so that you can troubleshoot issues associated with that neighbor, run the
log-neighbor-changes command, which is enabled by default.
The output is sent to the specified log file, usually
/var/log/frr/bgpd.log, and looks like this:
Troubleshoot Link-local Addresses
To verify that
frr learned the neighboring link-local IPv6 address via the IPv6 neighbor discovery router advertisements on a given interface, run the
show interface <if-name> command. If
ipv6 nd suppress-ra is not enabled on both ends of the interface, then
Neighbor address(s): has the other end's link-local address. That is the address that BGP uses when BGP is enabled on that interface.
IPv6 route advertisements (RAs) are automatically enabled on an interface with IPv6 addresses; the
no ipv6 nd suppress-ra command is not needed for BGP unnumbered.
vtysh to verify the configuration:
Instead of the IPv6 address, the peering interface name is displayed in the
show ip bgp summary command and wherever else applicable:
Most of the
net show commands can take the interface name instead of the IP address.
Enable Read-only Mode
As BGP peers are established and updates are received, prefixes might be installed in the RIB and advertised to BGP peers even though the information from all peers has not yet been received and processed. Depending on the timing of the updates, prefixes might be installed and propagated through BGP, and then immediately withdrawn and replaced with new routing information. Read-only mode minimizes this BGP route churn in both the local RIB and with BGP peers.
Enable read-only mode to reduce CPU and network usage when you restart the BGP process, or when you issue the
ip bgp command. Because intermediate best paths are possible for the same prefix as peers get established and start receiving updates at different times, read-only mode is particularly useful in topologies where BGP learns a prefix from many peers and the network has a high number of prefixes.
To enable read-only mode, run the
net add bgp update-delay <max-delay in 0-3600 seconds> [<establish-wait in 1-3600 seconds>] command.
The following example command enables read-only mode, sets the
max-delay timer to 300 seconds and the
establish-wait timer to 90 seconds.
The default value for max-delay is 0, which disables read-only mode.
establish-wait option is optional; however, if specified, the
establish-wait option must be shorter than the
Read-only mode begins as soon as the first peer reaches its established state and the
max-delay timer starts, and continues until either of the following two conditions are met:
All the configured peers (except the shutdown peers) have sent an explicit EOR (End-Of-RIB) or an implicit EOR. The first keep-alive after BGP has reached the established state is considered an implicit EOR.
If you specify the
establish-waitoption, BGP only considers peers that have reached the established state from the moment the
max-delaytimer starts until the
The minimum set of established peers for which EOR is expected are the peers that are established during the
establish-wait window,not necessarily all the configured neighbors.
- The timer reaches the configured
While in read-only mode, BGP does not run best-path or generate any updates to its peers.
To show information about the state of the update delay, run the
show bgp summary command.
Apply a Route Map for Route Updates
There are two ways you can apply route maps for BGP:
- By filtering routes from BGP into Zebra
- By filtering routes from Zebra into the Linux kernel
Filter Routes from BGP into Zebra
You can apply a route map on route updates from BGP to Zebra. All the applicable match operations are allowed, such as match on prefix, next-hop, communities, and so on. Set operations for this attach-point are limited to metric and next-hop only. Any operation of this feature does not affect BGPs internal RIB.
Both IPv4 and IPv6 address families are supported. Route maps work on multi-paths; however, the metric setting is based on the best path only.
To apply a route map to filter route updates from BGP into Zebra, run the following command:
Filter Routes from Zebra into the Linux Kernel
To apply a route map to filter route updates from Zebra into the Linux kernel, run the following command:
Converge Quickly On Link Failures
In the Clos topology, we recommend that you only use interface addresses to set up peering sessions. This means that when the link fails, the BGP session is torn down immediately, triggering route updates to propagate through the network quickly. This requires the following commands be enabled for all links:
ttl-security hops <hops>.
ttl-security hops specifies how many hops away the neighbor is. For example, in a Clos topology, every peer is at most 1 hop away.
Here is an example:
Converge Quickly On Soft Failures
It is possible that the link is up, but the neighboring BGP process is hung or has crashed. If a BGP process crashes, the FRRouting
watchquagga daemon, which monitors the various FRRouting daemons, will attempt to restart it. If the process is also hung,
watchquagga will attempt to restart the process. BGP itself has a keepalive timer that is exchanged between neighbors. By default, this keepalive timer is set to three seconds. This time can be increased to a higher number, which decreases CPU load, especially in the presence of a lot of neighbors.
keepalive-time is the periodicity with which the keepalive message is sent.
hold-time specifies how many keepalive messages can be lost before the connection is considered invalid. It is usually set to three times the keepalive time, and defaults to nine seconds. The following example shows how to change these timers:
The following snippet shows that the default values have been modified for this neighbor:
A BGP process attempts to connect to a peer after a failure (or on startup) every
connect-time seconds. By default, this is 10 seconds. To modify this value, run the following command:
You must specify this command for each neighbor.
BGP by default chooses stability over fast convergence. This is very useful when routing for the Internet. For example, unlike link-state protocols, BGP typically waits for a duration of
advertisement-interval seconds between sending consecutive updates to a neighbor. This ensures that an unstable neighbor flapping routes are not propagated throughout the network. By default, this is set to zero seconds for both eBGP and iBGP sessions, which allows for very fast convergence. You can modify this as follows:
The following output shows the modified value:
See this IETF draft for more details on the use of this value.
Caveats and Errata
ttl-security does not cause the hardware to be programmed with the relevant information. This means that frames will come up to the CPU and be dropped there. It is recommended that you use the
net add acl command to explicitly add the relevant entry to hardware.
For example, you can configure a file, such as
/etc/cumulus/acl/policy.d/01control_plane_bgp.rules, with a rule like this for TTL:
BGP Dynamic Capabilities not Supported
Dynamic capabilities, which enable BGP to renegotiate a new feature for an already established peer, are not supported in Cumulus Linux.
- Bidirectional forwarding detection (BFD) and BGP
- Wikipedia entry for BGP (includes list of useful RFCs)
- FRR BGP documentation
- IETF draft discussing BGP use within data centers
- RFC 1657, Definitions of Managed Objects for the Fourth Version of the Border Gateway Protocol (BGP-4) using SMIv2