On Mon, Feb 23, 2009 at 11:38:57AM +0100, Helmut Schaa wrote:

Jouni, would you accept a patch for wpa_supplicant that adds some retries
to l2_packet_send when the network stack returns an error?

Changing l2_packet_send() to do some magic on errors does not sound like
the correct solution here. l2_packet_send() caller might be more
interested in doing something should the send fail. If I understood the
description correctly, it sounds like kernel code is telling user space
that everything is ready and it is even delivering a received frame into
user space. If sending of the reply to this frame is now failing, it
sounds much more like a kernel side issue that should really be fixed
instead of trying to come up with workarounds in user space
applications. If the driver/network stack is not ready to accept
packets, it should not really claim to be.

I did not fully understand the details of the issue and why noop_qdisc
is used at the point when a frame is delivered to user space, so I may
be missing some details here. Anyway, I would highly prefer the kernel
to avoid delivering EAPOL frames into user space if it is not ready to
transmit a response to them.

-- 
Jouni Malinen                                            PGP id EFC895FA