[1399] in linux-net channel archive
ARP broken in 1.3.42
daemon@ATHENA.MIT.EDU (Paul Gortmaker)
Sun Nov 19 02:02:54 1995
From: Paul Gortmaker <gpg109@rsphy1.anu.edu.au>
To: kuznet@ms2.inr.ac.ru
Date: Sat, 18 Nov 1995 22:44:52 +1100 (EST)
Cc: iialan@www.linux.org.uk (Alan Cox), linux-net@vger.rutgers.edu
Hi,
ARP is broken (rather seriously) in '42. I can no longer ifconfig
one ethercard down and bring up another without arp complaining.
For example, if I drop eth0 and bring eth1 online, I get a set of
these:
arp_cache_bind: <NFS server's IP in reverse-hex> backlogged
arp_find: <NFS server's IP in reverse-hex> backlogged
arp_rcv: <NFS server's IP in reverse-hex> backlogged
When I switch from ethN to ethM, I ifconfig ethM up (with same IP, routes
etc.) and then immediately ifconfig ethN down (all in the same script).
All cards are on the same piece of cable.
Any further traffic results in more of the find/rcv messages.
If you wait for everything to timeout, you then get the cache_bind
one again as well. The NFS server does have the new hw_addr of eth1
in its ARP table, but no traffic flows.
If I *then* drop eth1 and go to eth2, I get:
arp_device_event: bug
which also appears on another machine when I drop a dialup SLIP link.
(simple SLIP link - no proxy ARP or IP fwd involved)
Let me know if you need more details.
Paul.
