em0, VLAN and bpf(?) trouble w/RELENG_5

[Robert Watson]

On Mon, 8 Nov 2004 astesin at ukrtelecom.net wrote:

> The problem. From time to time, vlan0 stops passing packets at all. At
> this moments, Catalyst stops seeing MAC of vlan0 (it's the same MAC as
> em0) in the mantione VLAN (untagged VLAN is also configured at em0 and
> works fine!).  This means that `show mac-address-table vlan XX' command
> on Catalyst don't show the MAC. 
> 
> The problem can be easily repeated manually. It's enough just to issue a
> command like `trafshow -I vlan0' of `tcpdump -I vlan0' and voila! vlan0
> is out of business, no packets are going through. 

Hmm.  Could I get you to try/investigate a few things:

(1) If you run tcpdump on the em0 interface itself, does the same thing
    happen?

(2) When vlan0 wedges, do you still see traffic on em0, and can you
    generate traffic on em0 that's picked up by the switch?

(3) Do other vlan pseudo-interfaces wedge under similar circumstances?

(4) Could you try running with "debug.mpsafenet=0" in loader.conf (reboot
    for it to take effect) and see if that makes a difference?

(5) Does it matter whether you enter promiscuous mode using BPF -- i.e.,
    "tcpdump -p -i vlan0" vs w/o the -p flag?

(6) When vlan0 is in the wedged condition, how "no packets" is it?  Can
    you send packets but not receieve, or receive packets and not send?

Thanks!

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Principal Research Scientist, McAfee Research