Hello Ondrej, Thank you for prompt answer.
First remark - suggested setup is that both BGP1 and BGP2 is part of OSPF domain, do not propagate BGP routes, but both propagate default route, locally configured as 'unreachable'.
I don't find Bird propagates unreachable 0/0. I made a few tests and only valid 0/0 routes where basis for announcing default gw (0/0) via OSPF. Of course - I may be mistaken in my configs. But in short - if I have valid 0/0 route in my table (fe. set by static route): # birdc sh route for 0.0.0.0/0 BIRD 1.6.3 ready. 0.0.0.0/0 via 195.x.x.x on enp2s0f6.3113 [static1 2018-10-10] * (200) I can export it: # birdc sh route export ospf1 | grep 0.0/0 0.0.0.0/0 via 195.x.x.x on enp2s0f6.3113 [static1 2018-10-10] * (200) If there is no such valid entry - nothing is exprted. This is why I am worried what would happen if fe. enp2s0f6.3113 gets down? I suppose it will set static route as unreacheable and propagation of 0/0 will be stopped. Let's KIS = Keep It Simple. Let's leave for a while other parts of discussion. The key question is - how to make Bird announcing 0/0 for OSPF ALWAYS? Without any need to relly somethning else which can get unrecheable? This is something which we may call a bit cisco style like: "Propagate default gw ALWAYS" = unconditionally. As I understand you correctly - there is no such mechanizm so far? Best wishes, --- Zboj (Piotr Marciniak) zboj /at\ mnc dot pl -----Oryginalna wiadomość----- From: Ondrej Zajicek Sent: Thursday, October 11, 2018 2:20 PM To: Piotr Marciniak Cc: bird-users@network.cz Subject: Re: OSPF generate default route - again On Wed, Oct 10, 2018 at 04:07:01PM +0200, Piotr Marciniak wrote:
Hello,
I’ve spent some time trying to find any option to make our border BGP router to announce itself as defaul router by OSPF to our network switches. No luck.
This discussion was rised fe. 5 years ago but with no answer (https://bird-users.network.narkive.com/AD6KOk8b/ospf-generate-default-route). All docs end examples I found touch only redistribute aspect of the problem. I think it is not optimal for such implementations.
Hello Thanks for your comments, i agree that BIRD needs mechanism that could be used to generate the default route based on received BGP routes. Our inability to already deliver it is generally a result of consecutive generalization and moving goalposts. The idea started as simple aggregation protocol, developed over time to conditional route protocol, then to conditional route extension to static protocol and finally to conditional extensions to filters, leaving behind several batches of partially written code, postponed several time when more pressing issues appeared. I still have some notes / minor objections to your mail (below)
OK – let draw simple example.
Step 1.
Let’s have just 2 border backbone BGP routers with a few peers and full tables. Let’s name them just simple BGP1 and BGP2. Of course they talk one to another by iBGP as well. In fact – they know everything about Internet, so they do not need any 0/0 routes – external or static. If any peer fails – they will just recalculate tables and everything will work fine. ,,, Step 7.
Theoretically we may imagine that in my example OSPF10 will act as ‘default originate always’ router and forwards packets by static route to BGP1... What a waste. How to secure redundancy supported by BGP2?
There are several cases how BGP1 and BGP2 can be connected. They can be connected directly (in L3 sense), through intermediate routers having full BGP table by IBGP, or through intermediate OSPF-only routers. Third case is problematic in many ways even with proper default route generator based on BGP state, so i will ignore that case, and the first two cases are generally the same, so for simplicity i assume that BGP1 and BGP2 are directly connected. First remark - suggested setup is that both BGP1 and BGP2 is part of OSPF domain, do not propagate BGP routes, but both propagate default route, locally configured as 'unreachable'. In such setup if e.g. BGP1 (as router) fails, traffic get redirected to BGP2 by OSPF. If both BGP1 and BGP2 are running, but on BGP1 fails upstream BGP session, then BGP1 still originates default route and attracts local traffic, but it has full BGP table received from BGP2 by iBGP, so received local traffic is forwarded to BGP2 and delivered to Internet. While there may be an additional hop, there is no blackholing or broken routing, so it not a big issue that default route is originated unconditionally. While in this setup a conditional default route is not a necessary feature, there are other setups where it is required for redundancy, e.g. if BGP border routers use BGP only to propagate their range, do not keep full BGP table and only have default routes to their upstream eBGP neighbors. Second remark - you seems to mix two independent issues - whether a default route is originated conditionally or unconditionally and whether it is originated by an external protocol (e.g. static) or originated internally by OSPF. While i agree with usefulness of conditional origination, i do not see a reason for doing it internally by OSPF instead of externally as regular route. I see several reasons for do it by e.g. conditional static route: 1) Default route may be necessary outside of OSPF, e.g. in example above with low-end BGP speakers having conditional default route in FIB. 2) No need to do it independently for OSPF, RIP, Babel and other IGPs. 3) Fits more in BIRD design, uniform way to handle it like other routes, for e.g. attribute setting. 4) More general tool, could be used to generate not only default route but also other conditional routes for other purposes.
This announcement should be based on simple rule that Bird may know from other protocols (like BGP but not only)
BTW, what kind of conditions would you expect to be able to express in the mentioned simple rule? -- Elen sila lumenn' omentielvo Ondrej 'Santiago' Zajicek (email: santiago@crfreenet.org) OpenPGP encrypted e-mails preferred (KeyID 0x11DEADC3, wwwkeys.pgp.net) "To err is human -- to blame it on a computer is even more so."