[quagga-dev 1400] Re: link-detect issues (not quite working)

Paul Jakma paul at clubi.ie
Wed Aug 11 21:53:52 BST 2004

On Wed, 11 Aug 2004, Eric S. Johnson wrote:

> paul at clubi.ie said:
>> I question the wisdom of wanting to remove the connected route.
> Well, if ospf can provide a better (IE working) path...
> But my further investigation has me confused, as a better path 
> *should have* come from ospf and is not anyway. Once I have a 
> plausible demo of the problem, ill post more.

The connected network will *always* be routed out the local 
interface, if the interface is still 'UP'. What you want to do, 
presumably is to remove the connected route for so long as the link 
is not running, and obviously reinstall it if the connected link 
comes back, so that if an alternate path exists to that 
connected-but-not-running network on that link you can access it via 
the alternate path.

The problem is, will the reply ever get back to you? The remote host 
can not know that you have lost your link, presuming the packet has a 
source address in the connected-but-not-running network, the remote 
host will use its connected route to reach you.

So, unless the original packet is originated from an INADDR_ANY bound 
socket[1], removing the connected route simply is a pretty poor 
solution. TCP connections will not survive such an event. It wont 
work if you're trying to ensure connectivity to a service listening 
on or published as being available on the IP on that 
connected-but-not-running network.

Ie removing the connected route is just a very *poor* solution in 
general, and will not buy you much.

If you want an address that is independent of link-failures, assign a 
*seperate* /32 address to loopback.

> (did I see something in other messages about netlink messages getting
> lost or delayed?)

They can do with large numbers of interfaces and/or high rate of 
netlink messages with the typical receive buffer size(s). Hasso's 
patch goes a long toward fixing it, but you wont know whether you 
need a bigger receive buffer size until zebra starts losing messages.

> E

Paul Jakma	paul at clubi.ie	paul at jakma.org	Key ID: 64A2FF6A
Moneyliness is next to Godliness.
 		-- Andries van Dam

More information about the Quagga-dev mailing list