two bugs and a patch

[Nickolai Zeldovich]

>                   It seems to be that it might be a problem if
> rxi_DecongestionEvent() can run concurrently for same peer, that might
> cause problems. I just can't figure out if its happening.

Actually after some more thought, I think this might be yet another
result of my peer structs being corrupted and the refcount might be
getting set to zero that way..
 
> >                   In case you have multi-homed fileservers, the
> > patch attached might help you.
> If I tried a little harder to test the patches they might work.

Thanks; I actually do have multi-homed fileservers. It seems to be
running ok so far with this patch; we'll see if it crashes less now.

-- kolya