netdev - Re: [PATCH v7] xfrm: add SA information to the offloaded packet when if

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <Z1lv/+VBldgUYTGw@gauss3.secunet.de>
Date: Wed, 11 Dec 2024 11:57:03 +0100
From: Steffen Klassert <steffen.klassert@...unet.com>
To: Jakub Kicinski <kuba@...nel.org>
CC: Feng Wang <wangfe@...gle.com>, Leon Romanovsky <leon@...nel.org>,
	<netdev@...r.kernel.org>, <antony.antony@...unet.com>, <pabeni@...hat.com>
Subject: Re: [PATCH v7] xfrm: add SA information to the offloaded packet when
 if_id is set

On Tue, Dec 10, 2024 at 07:20:48PM -0800, Jakub Kicinski wrote:
> On Tue, 10 Dec 2024 10:38:34 +0100 Steffen Klassert wrote:
> > > This patch was done based on our previous discussion.  I did the
> > > changes we agreed on.  
> > 
> > there is still no real packet offload support for netdev sim.
> > And as said, this is at most the second best option.
> > 
> > You need to prove that this works. I want a complete API,
> > but I also want a working one.
> > 
> > The easiest way to prove that this is implemented correctly
> > is to upstream your driver. Everyting else is controversial
> > and complicated.
> 
> Yes, I don't have full context but FWIW offload changes accompanied 
> by just netdevsim modifications raise a red flag:

Actually, we have the mlx5 driver that supports packet offload.
When that was implemented, xfrm interfaces were no usecase.
Because of that, we forgot to care for the xfrm interface ID as
a lookup key. The problem is that users can still configure
policies/states for xfrm interfaces and packet offload.
The driver then just don't know if a packet was routed via
a xfrm interface, and if so, via which one. So it might happen
that a wrong policy/state is applied to a packet.

My idea was to fix that by supporting xfrm interfaces
with packet offload in the stack. But after looking
a bit closer to the code yesterday, I noticed that we
might not need any stack changes to get this right.

I think the driver should do the following:

If xfrm interfaces are not supported:

- Reject policies and states that have xfrm interfaces
  configured.

If xfrm interfaces are supported:

- Accept policies and states that have xfrm interfaces
  configured.

- On TX: Use the drivers xdo_dev_offload_ok function
  to check if the packet came via the correct xfrm
  interface and/or amend that info by adding the
  xfrm state to the packets secpath.

- On RX: Driver traveses through the list of registered
  xfrm interfaces and match these against the used SAs
  xfrm interface ID.

By doing that, no stack changes should be required.