Depends on what you mean by “any prefixes learned by the bgp peers”. I think most upstreams would manually configure route filters to control what BGP advertisements to accept, and maybe also an ACL based on source IP. Otherwise there’s too much risk a customer would advertise routes for non owned blocks or bogons to you, either maliciously or by mistake, and you would automatically install these routes and pass them on. Boom, you’re trying to advertise 8.0.0.0/0 or 10.0.0.0/0 and also allowing source address spoofing from those IPs, because your customer did something stupid.
Without an LOA and manual configuration, just advertising a route via BGP to another ASN should not cause those routes to be accepted. Of course your upstream probably has these same rules, so your customer would have to give you an LOA that also goes to your upstream authorizing them to advertise those blocks. From: Dennis Burgess Sent: Monday, January 12, 2015 11:46 AM To: af@afmug.com Subject: Re: [AFMUG] BCP38 Basically ,any IPs that SHOULD be sourced from your network. But yes, the idea behind BCP38 is to block src address packets originating from your network that SHOULD NOT. So yes, you should already have those rules to not all traffic from your network if it’s coming from a IP that should not come from your network, and yes that would include any customer originated traffic. An example, customer has 4 /19s and two /22s, plus has about 30 BGP peers for customer traffic. The 5 prefixes would be allowed out, plus any prefixes learned by the bgp peers. If there were two upstream on the same router, both would have a line, if the SRC address is ! (not) customer prefixes, including the 5 prefixes they use, then it would be dropped on egress of the upstream ports. An example of this is add action=drop chain=forward out-interface=ether17-internet src-address-list=!Inside-IPs The inside_ips list include the local prefixes and the customer prefixes. Dennis Burgess, CTO, Link Technologies, Inc. den...@linktechs.net – 314-735-0270 – www.linktechs.net From: Af [mailto:af-boun...@afmug.com] On Behalf Of Ken Hohhof Sent: Monday, January 12, 2015 10:55 AM To: af@afmug.com Subject: Re: [AFMUG] BCP38 Yeah, I’m missing what the big deal is here. If you’re talking about your border router to your upstream, why would you allow outbound traffic with source IPs outside your IP blocks? Allow your IPs, block the rest. If you’re talking about other routers within your network and are wanting to stop the traffic at the source, it could get more complicated since I assume we all use some private IP space within our networks for various purposes mostly management addresses on network equipment. Dennis mentions customer IPs, if you route customer blocks those would also be allowed, based on an LOA. From: Dennis Burgess Sent: Monday, January 12, 2015 10:43 AM To: af@afmug.com Subject: Re: [AFMUG] BCP38 Very simple. In MT we do an address list of all valid subnets behind the core routers, this would include any prefixes that you own or use, plus any BGP prefixes learned from your customers. Then a simple, out interface (internet) drop if its not SRCed from that list. Not exactly IP tables, but there ya go.. Dennis Burgess, CTO, Link Technologies, Inc. den...@linktechs.net – 314-735-0270 – www.linktechs.net From: Af [mailto:af-boun...@afmug.com] On Behalf Of Sean Heskett Sent: Monday, January 12, 2015 10:25 AM To: af@afmug.com Subject: Re: [AFMUG] BCP38 Hey Mike, Would you be willing to post an iptables statement that would drop this traffic? Thanks, Sean On Monday, January 12, 2015, Mike Hammett <af...@ics-il.net> wrote: http://www.bcp38.info/index.php/Main_Page Make sure you implement this in your networks. Drop all outbound traffic to your upstream that is not from valid public IP space. ----- Mike Hammett Intelligent Computing Solutions http://www.ics-il.com
