ports/147897: Openbgp nexthop invalid after em0 resetting

[Paulo Fragoso]

>Number:         147897
>Category:       ports
>Synopsis:       Openbgp nexthop invalid after em0 resetting
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    freebsd-ports-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          sw-bug
>Submitter-Id:   current-users
>Arrival-Date:   Wed Jun 16 01:30:01 UTC 2010
>Closed-Date:
>Last-Modified:
>Originator:     Paulo Fragoso
>Release:        FreeBSD 7.2-RELEASE
>Organization:
Nlink Comercio e Servicos Ltda
>Environment:
FreeBSD * 7.2-RELEASE-p7 FreeBSD 7.2-RELEASE-p7 #1: Thu Apr  8 10:19:05 BRT 2010     paulo@*:/usr/obj/usr/src/sys/KERNEL1  amd64

>Description:
After a hardware reset on em0 this interface and its vlan are marked with nexthop invalid:

Jun 11 09:05:42 * kernel: em0: watchdog timeout -- resetting
Jun 11 09:05:42 * kernel: em0: link state changed to DOWN
Jun 11 09:05:42 * kernel: vlan1: link state changed to DOWN
Jun 11 09:05:43 * kernel: em0: link state changed to UP
Jun 11 09:05:43 * kernel: vlan1: link state changed to UP 

{root}:/home/paulo:2# bgpctl show interfaces
Interface      Nexthop state  Flags          Link state
tun0           ok             UP             invalid
vlan1          invalid        UP             down
lo0            ok             UP             invalid
bge1           ok             UP             Ethernet, active, 1000 MBit/s
bge0           ok             UP             Ethernet, active, 100 MBit/s
em1            ok             UP             Ethernet, active, 1000 MBit/s
em0            invalid        UP             Ethernet, no carrier

All sessions are closed after that and this machine lost all routes after some time.
>How-To-Repeat:
Install openbgp from ports and configure it to open two session, one by em0 and other by vlan.

Try to make a hardware reseting on em0 changing net.inet.ip.fastforwarding to 1 any generating heavy traffic through interface.

When to happen a watchdog timeout with a reset those interfaces change to nexthop state invalid on OpenBGP and don't change back to ok.

Now I'm runnig OpenBGP 4.7, but this problem happened with 4.6 too.

{root}:~:2# pkg_info |grep openbgp
openbgpd-4.7.20100410 Free implementation of the Border Gateway Protocol, Version
>Fix:
I don't know how to fix this, but there is a workaround, forcing changing state for this interface:

ifconfig em0 down
ifconfig em0 up

after an up by the hand nexthop state changes to ok.

Or, of course, restarting bdpd daemon all back to normal work.

>Release-Note:
>Audit-Trail:
>Unformatted: