On Fri, Oct 07, 2005 at 04:35:51PM +0200, Olivier Mehani wrote: > On Fri, 7 Oct 2005 16:09:28 +0200 > Lio Goehrs <[EMAIL PROTECTED]> wrote: > > > The address space can be given by one of the provider. > > But then, I understand that the route to these addresses will go > through the address-providing ISP. Correct ? >
No. You need provider independent address space for such setups plus a AS number. At least for IPv4 it goes this way. In IPv6 land it is no longer possible to get provider independent address space and so multihoming is broken and this makes IPv6 unusable in the real life. > Or is the very role of bgpd to tell the _other_ provider that the > adresses are also reachable through his routers, which will then > propagate the information to the whole internet ? > The role of bgp is just to exchange routing information and selecting the best path. So yes that's the role of bgpd. > (I absolutely don't know about BGP, thought it was time I started > getting information ;)) > > Morevover, I guess not every provider accepts BGP information from its > clients. And what prevents me from sending crafted BGP packects saying > that I can route to a specific address space I actually don't own ? > Getting a bgp session from a provider is normaly the smallest problem. OK most will refuse to do that for a private customer but for business customers with fat pipes this is mostly no porblem. Address spoofing is a known problem at that's why the upstream providers should filter what you send to them. It is possible to hijack address room at least for part of the internet. As an example it happend once through missconfiguration that a small customer started to announce a /8 as individual /24 networks. This resulted in a major internet outage because some backbone cisco routers started to reload because of memory shortage. -- :wq Claudio