Open cagney opened 6 months ago
On Tue, 21 May 2024, The Libreswan Vault wrote:
Subject: Re: [libreswan/libreswan] make probing nic_offload really on-demand
Closed #1701 as completed via 9a13acb.
This seems wrong?
commit 9a13acb10937b28e823a4cd436f3fad53dbc4281 (HEAD -> main, origin/main) Author: Andrew Cagney @.***> Date: Tue May 21 15:20:04 2024 -0400
ipsec globalstatus: only list algs approved by ike_alg
fix #1701 in ipsec stats use ike_alg to select algorithm stats to be shown
I don't see this being related to nic_offload ?
The complication is:
* Policy Packet nic offload requires us to poke an IPsec policy
* hole that allows IKE packets. This installs one IPsec policy
* per interface (that supports offload), this function is called
* for port 500.
which happens when the interface is detected. If probing the interface is to be made on-demand then so too is poking this hole.
I've been playing probing when adding connections. I find the behaviour a little weird - announcing that an interface supports offload better fits with adding the interface and not the connection. With that in mind, I think adding knobs to:
ipsec start
invokes addconn
which triggers a listen unconditionally; the only way round this is to invoke pluto
directly, oopsmight be better.
moving to 5.2; #1753 provides an immediate solution
per:
1697
in 4.x and 5.0 the code probes the kernel to see if it supports nic_offload while adding an interface; which effectively means during startup (when
ipsec listen
is run).The code should instead probe nic_offload on an interface when a connection tries to orient to it.