[Planetlab-devel] Re: [PL #45504] VSYS script for adjusting txqueuelen parameter on tap interface

[Thom Haddow]

Hello,

Sapan Bhatia wrote:
 > Hi Thom,
 >
 > Did you get a chance to test the latest version of your script? I
 > noticed that I missed updating the path to netblocks.conf in one of the
 > scripts when I updated it on planetlab-04.cs.princeton.edu
 > <http://planetlab-04.cs.princeton.edu>. I have done so now.

When I tried it on Friday that was breaking it, but I've just tried again
and it's back to an almost-working state.

 > I have also tested the script in my own slice and it appears to be
 > working fine. As soon as you tell me that you are satisfied with it, I
 > can deploy it on the beta nodes, and then in a few days PlanetLab-wide.

I've found a couple more problems:

Firstly, using vif_down to bring down the interface causes the process that
opened the device (via fd_tuntap) to hang briefly and we get the following
kernel error:

Message from syslogd at planetlab-04 at Oct 20 15:58:45 ...
  kernel: unregister_netdevice: waiting for tap622-0 to become free. Usage 
count = 6

Closing the process that opens the device seems to bring the interface down 
cleanly. I maybe need to do something extra in the tun alloc code to make it 
handle this better but I'm not sure what this would be. Possibly it's better 
just to remove the vif_down script and keep closing-the-FD as the only means of 
bringing down the interface?

Secondly, the code wasn't quite functioning properly yet - it wasn't correctly
passing the name back over the control socket. fd_tuntap.c line 48 currently reads:

	vec.iov_base = &vif_name;

Whereas it should actually be

	vec.iov_base = vif_name;

I'm guessing you already spotted this if you had the code working properly? 
Either way, this will need to be changed so we're not left to just guess
the name of the new interface.

cheers,

Thom