[quagga-users 4063] FW: OSPFD loses all neighbours after shutdown of one interface

[Daniel Niasoff]

Hi All,



I have a server acting as a firewall running RedHat Enterprise Linux 3
(2.4.21) and Quagga 0.99.0 CVS.



It has two physical Gigabit Interfaces which each have about 10
sub-interfaces carrying traffic from different vlans.



It is connected to one internal server network, 4 DMZ's, and three
external networks. There are dual paths to either network, via eth0 or
eth1.



The problem that I am having is that it works fine until I shutdown the
port on the switch connected to eth0 (I do a "shut" command on the
interface).



After issuing the shut command, the network reconverges after about 40
seconds but and about 1/2 a minute after that all adjacencies are lost
and all ospf routes disappear from the network.



When the system goes into error state I get loads of
"Neighbor[192.168.3.252] state Init is less than Exchange" and all
neighbours are in the "Init/DROther" stage.



This error occurs on earlier versions of quagga as well.



I am attaching a log file, conf files, and routes pre and post the
error state.



Many thanks



Daniel



P.S. I did initially get loads of "can't setsockopt IP_ADD_MEMBERSHIP
(fd 6, addr 192.168.71.1, ifindex 19, AllDRouters): No buffer space
available; perhaps a kernel limit on # of multicast group memberships
has been exceeded?" errors but this has been resolved by running a "echo
200 > /proc/sys/net/ipv4/igmp_max_memberships" command. (thanks Andy
:-))

[Vladislav Bogdanov]

Daniel Niasoff wrote:

Hi

I think it may be the same problem as I noted in [quagga-users 4055]. I
don't know exactly from where it raises (the reason could be as ospfd as
rhEL3 kernel), but there are too many similarities. We both use rhEL3,
we both have multipath routing, equal-cost in your case and not-equal in
mine.
Can you simply plug second cord out of the box (or shutdown the port on
the switch) and see what's happened with OSPF hellos on vlans on another
phisical interface ("tcpdump -i eth1.x 'proto ospf'" should help)? If
I'm right, your box should stop sending them after some time. At least
some messages in your log make me think so (compared to my situation).
And would you downgrade to 0.98.2 in order to compare logging output of
two setups?

40 seconds is a default dead-time for ospf interface, so I think it
could be definitly the same problem. Init/DRother state means that
router receives hellos from neighbours but can't synchronize with them.
In my case it happens because hellos don't leave the box.

Best,
Vladislav

Post by Daniel Niasoff
Hi All,
I have a server acting as a firewall running RedHat Enterprise Linux 3
(2.4.21) and Quagga 0.99.0 CVS.
It has two physical Gigabit Interfaces which each have about 10
sub-interfaces carrying traffic from different vlans.
It is connected to one internal server network, 4 DMZ’s, and three
external networks. There are dual paths to either network, via eth0 or
eth1.
The problem that I am having is that it works fine until I shutdown
the port on the switch connected to eth0 (I do a “shut” command on the
interface).
After issuing the shut command, the network reconverges after about 40
seconds but and about 1/2 a minute after that all adjacencies are lost
and all ospf routes disappear from the network.
When the system goes into error state I get loads of
“Neighbor[192.168.3.252] state Init is less than Exchange” and all
neighbours are in the “Init/DROther” stage.
This error occurs on earlier versions of quagga as well.
I am attaching a log file, conf files, and routes pre and post the
error state.
Many thanks
Daniel
P.S. I did initially get loads of “can't setsockopt IP_ADD_MEMBERSHIP
(fd 6, addr 192.168.71.1, ifindex 19, AllDRouters): No buffer space
available; perhaps a kernel limit on # of multicast group memberships
has been exceeded?” errors but this has been resolved by running a
“echo 200 > /proc/sys/net/ipv4/igmp_max_memberships” command. (thanks
Andy J)