Hello,
On Wed, Apr 28, 2021 at 02:25:26AM +0200, Alexander Bluhm wrote:
> On Wed, Apr 28, 2021 at 03:19:47AM +0300, Vitaliy Makkoveev wrote:
> > The code not only breaks loop but also cleans the queue. Should this be
> > kept?
>
this is a good point
> In IPv6 nd6_cache_lladdr() we have neither queue nor loop, just a
> single mbuf on hold. But we have that discard logic, too.
>
> if (ln->ln_hold) {
> struct mbuf *n = ln->ln_hold;
> ln->ln_hold = NULL;
> /*
> * we assume ifp is not a p2p here, so just
> * set the 2nd argument as the 1st one.
> */
> ifp->if_output(ifp, n, rt_key(rt), rt);
> if (ln->ln_hold == n) {
> /* n is back in ln_hold. Discard. */
> m_freem(ln->ln_hold);
> ln->ln_hold = NULL;
> }
> }
>
> So I guess there is a reason to clean up.
>
I would say the la->la_mq queue should be empty once we dispatch all
packets to wire. If it is not empty, then something went wrong and packets
got recirculated back to la->la_mq. I think dropping them is better
plan than keeping them in queue. There is not much we can do in this
case than to drop them.
thanks and
regards
sashan
