Re: [PATCH v3 net] net, ip6_tunnel: enhance tunnel locate with link check

[William Dauchy <w.dauchy@criteo.com>]

On Thu, Feb 13, 2020 at 10:34:19AM +0100, Nicolas Dichtel wrote:
> > mtu = ETH_DATA_LEN;
> > if (t->parms.link) {
> > 	tdev = __dev_get_by_index(t->net, t->parms.link);
> > 	if (tdev && tdev->mtu < mtu)
> Why this second condition? Why not allowing more than ETH_DATA_LEN (1500)?
> ip_tunnels do:
>         if (tdev) {
> 
>                 hlen = tdev->hard_header_len + tdev->needed_headroom;
> 
>                 mtu = min(tdev->mtu, IP_MAX_MTU);
> 
>         }
> which seems better.

I wrongly mixed the two codes in my head as I wanted to take the lowest
of the two values; will correct that; sorry for the confusion.

> Note also that you patch ip6_tnl_dev_init_gen(), but ip6_tnl_link_config() is
> called later and may adjust the mtu. I would suggest to take care of link mtu in
> ip6_tnl_link_config().

agreed, I overlooked it. Unsure whether I can put it in CAP_XMIT
condition as well.

> hard_header_len is not set for ipv4 tunnels, but for ipv6 tunnels:
>         dev->hard_header_len = LL_MAX_HEADER + t_hlen;
> 
> This is not the real value, I don't think you can calculate the real mtu based
> on this.

understood, I indeed did not noticed hard_header_len was not set in
ip_tunnel.

Best,
-- 
William