The FB6000 series router provides BGP routing capabilities. The aim of the design is to make configurationm simple for a small ISP or corporate BGP user - defining key types of BGP peer with pre-set rules to minimise mistakes.

The key features supported are:-

A typical installation may have transit connections from which a complete internet routing table is received, peers which provide their own routes only, internal peers making an IBGP mesh, customers to which transit is provided and customer routes may be accepted. To make this set up simple the <peer> definition contains a type attribute. This allows simple BGP configuration such as:-

<bgp as="12345">
 <peer as="666" name="transit1" type="transit" ip="1.2.3.4"/>
 <peer as="777" name="transit2" type="transit" ip="2.3.4.5 2.3.4.6"/\>
 <peer type="internal" ip="5.6.7.8"/\>
</bgp>

This example has two transit providers, the second of which is actually two peer IP addresses, and one internal connection. Note that the peer AS is optional and unnecessary on internal type as it has to match ours.

The exact elements that apply are defined in the XML/XSD documentation for your software release.

The type attribute controls some of the behaviour of the session and some of the default settings as follows.

Each peer has a set of import and export rules which are applied to routes that are imported or exported from the peer. There are also named bgp-filter which can be used as import-filters or export-filters.

The objects import and export work in exactly the same way, checking the routes imported or exported against a set of rules and then possibly making changes to the attributes of the routes or even choosing to discard the route.

Each of these objects contain:-

The rules are considered in order. The first rule to match all of the matching attributes is used. If no rules match then the default actions from the import/export object are used.

In addition, the top level import/export has a prefix list. If present then this will limit the prefixes processed at a top level, dropping any that do not match the list without even considering the rules.

Specific well known communities are supported natively. Some of these are set automatically based on peer type and can be explicitly removed using the detag action. These rules are automatically checked for exporting routes unless overridden on the peer attributes.

The FireBrick allows black hole routes to be defined using the the blackhole object. Routing for such addresses is simply dropped with no ICMP error. Such routes can be marked for BGP announcement just like any other routes.

It is also possible for L2TP announced routes to be marked as black hole routes using the D filter. If L2TP is marked to BGP announce such routes they are set to be bgp="true" rathed than the bgp setting defined.

In order to ensure that your internal BGP network sees such routes as a black hole, and not simply as a route to the router than has the black hole defined (where the packets will be dropped), you can ensure all black hole routes are announced using a suitable community tag. In many cases an EBGP peer will even allow you to announce black hole routes to them with a suitable community tag.

The top level bgp object includes a blackhole-community attribute which can be set to a tag that is used to mark routes as a black hole within your network. Any route received on a BGP peer within that config object which includes the specified community is treated as a black hole route. It is installed in the BGP routes and propagated as normal but it is internally set as forwarding to nowhere and packets dropped as a black hole.

Each peer object also has a blackhole-community tag. If set then this is added to any black hole routes announced. If not set, then black hole routes are not sent on EBGP links. On IBGP links, if not set, the blackhole-community from the parent bgp is added if present. Black hole routes are always announced on IBGP (subject to normal rules for announcement).

To use this, define a suitable blackhole-community for your network, such as YourAS:666 and set in all bgp objects. For all EBGP peers, set the peer object blackhole-community attribute with the tag they expect for black hole routes.

It is unlikely you would want to announce a black hole route to an EBGP peer without an agreed tag as you will be drawing traffic from them only to be discarded. If you want to do this, you have to specify a blackhole-community to add, but this could simply be your own community tag for black holes.

Sometimes you may want to inject a blackhole route in to your network, but not actually blackhole within your network, simply ensuring that the routes propagete via BGP until edge routers where they are announced to transit/peering as blackhole community tagged routes (as above).

For thsi reason, blackhole route objects can be tagged no-fib which creates the routing in the routing table but does not impact packet forwarding. By defining a greyhole-community on your bgp settings, this will be used for IBGP to pass the route around as a blackhole route but with no-fib set.

This is useful where injecting a black hole is needed, but you want to ensure internal routing to externally blackholed routes.

The top level bgp object includes a dead-end-community attribute which can be set to a tag that is used to mark routes as a dead end within your network. Any route received on a BGP peer within that config object which includes the specified community is treated as a dead end route. It is installed in the BGP routes and propagated as normal but it is internally set as forwarding to nowhere and icmp errors generated (rate limited as usual). Any route installed as network are announced with this community. Note, this is not set automatically on a nowhere route, allowing a route to be announced to get to this FireBrick to be propagated via IBGP.

The effect of this is that your network can include one (or more) source of top level network routes which, within your network, are installed as dead ends at each point. Without this these would be announced to your internal network so traffic is sent to the originating router and it has to handle all dead end traffic. Using this system you can ensure dead end traffic is handled at your borders instead.

The BGP specification is clear that receipt of a path attribute that we understand but is in some way wrong should cause the BGP session to be shut down. This has a problem if the attribute is one that is not known to intermediate routers in the internet which means a bad content is propagated to multiple routers on the internet and they will drop their session. This can cause a major problem in the internet.

To work around this have, by default, ignore-bad-optional-partial set to true. The effect is that if a path attribute we understand is wrong, and it is optional, and trhe router that sent it to us did not understand or check it (partial bit is set), we ignore the specific route rather than dropping the whole BGP session.

The network element defines a prefix that is to be announced by BGP by default, and tagged with any dead-end-community, but otherwise treated the same as a nowhere route.

Subnet and route elements used for normal set-up of internal routing can be announced by BGP using the bgp attribute with the same values as the well known community tags, please true meaning simply announce with no tags, and false meaning the same as no-advertise.

Many other objects in the configuration which can cause routes to be inserted have a bgp attribute which can be set to control whether the routes are announced, or not.

The FB6000 has an aggressive route feasibility test that confirms not only routability of each next-hop but also that it is answering ARP/ND requests. Whenever a next-hop is infeasible then all routes using that next-hop are removed. When it becomes feasible the routes are re-applied. This goes beyond the normal BGP specification and minimises any risk of announcing a black hole route.

The web control pages have diagnostics allowing routing to be show, either for a specific target IP (finding the most specific route which applies), or for a specified prefix. This lists the routes that exist in order, and indicates if they are supressed (e.g. route feasibility has removed the route). There are command line operation to show routing as well.

It is also possible, using the command line, to confirm what routes are imported from or exported to any peer.

The diagnostics also allow ping and traceroute which can be useful to confirm correct routing.

On router shutdown/reboot (e.g. for software load) all established BGP sessions are closed cleanly. Before the sessions are closed all outgoing routes are announced with a lower priority (high MED, low localpref, prefix stuffed) and then a delay allows these to propagate. This is a configurable option per peer and the maximum delay of all active peers is used as the delay. Setting to zero will not do the low priority announcement. A special case of setting this delay to a negative value on a peer causes routes to be specifically withdrawn before the delay rather than announced low priority.

The FireBrick supports RFC5082 standard TTL security. Simply setting ttl-security="1" on the peer settings causes all of the BGP control packets to have a TTL of 255 and expects all received packets to be TTL 255 as well.

You can configure multiple hops as well, setting ttl-security="2" for example still sends TTL 255 but accepts 254 or 255. This works up to 127.

You can also configure a non standard forced TTL mode by setting a negative TTL security (-1 to -128) which forces a specific TTL on sending packets but does not check received packets. For example, setting ttl-security="-1" causes a TTL of 1 on outgoing packets. This simulates the behaviour of some other routers in IBGP mode. Using -2, -3, etc, will simulate the behaviour of such routers in EBGP multi-hop mode. This is non standard as RFCs recommend a much higher TTL and BGP does not require TTLs to be set differently.

Without ttl-security set (or set to 0) the RFC recommended default TTL us used on all sent packets and not checked on received packets.