Re: NFSroot and new IP routing code [was: NFS root filesystem broken in 2.1.36]

David S. Miller (davem@jenolan.rutgers.edu)
Mon, 28 Apr 1997 11:25:08 -0400


From: okir@monad.swb.de (Olaf Kirch)
Date: Sun, 27 Apr 1997 18:34:17 +0200 (MET DST)

As ANK updated the IP routing stuff, he seems to have changed some
things in the way you have to configure an interface and route. As
a result, the old interface handling in nfsroot breaks as well (the
problem is that the kernel ARPs for the NFS server, but drops the
ARP replies somewhere because it won't find a matching route).

Unfortunately, I don't understand ANKs new code too well. I've
spent the better part of two days on trying to figure out what I've
been missing, but haven't succeeded so far.

[ BTW: Alexey is back, but is still catching up on email backlog, so
he should be able to help us with this stuff again soon. ]

I think I've found it, when you add the route add RTF_MAGIC to the
flags, see what that does. I am not certain if this is it, but
specifically in nfsroot.c:root_dev_chg_route() change:

route.rt_flags = RTF_UP;

to

route.rt_flags = (RTF_UP | RTF_MAGIC);

I think this will fix it, because if RTF_MAGIC is not set then during
fib_create(), fib_autopublish() will not get called, that call is what
allows ARP responses to be recognized.

---------------------------------------------////
Yow! 11.26 MB/s remote host TCP bandwidth & ////
199 usec remote TCP latency over 100Mb/s ////
ethernet. Beat that! ////
-----------------------------------------////__________ o
David S. Miller, davem@caip.rutgers.edu /_____________/ / // /_/ ><