netdev - Re: [PATCH net-next v3 2/3] net/macsec: Add MACsec skb extension Rx Data path support

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

Message-ID: <20220621122641.3cba3d38@kernel.org>
Date:   Tue, 21 Jun 2022 12:26:41 -0700
From:   Jakub Kicinski <kuba@...nel.org>
To:     Lior Nahmanson <liorna@...dia.com>
Cc:     Paolo Abeni <pabeni@...hat.com>,
        "edumazet@...gle.com" <edumazet@...gle.com>,
        "davem@...emloft.net" <davem@...emloft.net>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        Raed Salem <raeds@...dia.com>, Jiri Pirko <jiri@...dia.com>,
        Ben Ben Ishay <benishay@...dia.com>
Subject: Re: [PATCH net-next v3 2/3] net/macsec: Add MACsec skb extension Rx
 Data path support

On Tue, 21 Jun 2022 12:39:23 +0000 Lior Nahmanson wrote:
> > Shooting from the hip a little bit, but macsec being a tightly bound L2 upper
> > maybe metadata dst is a workable solution for carrying the sci and offload
> > status between upper and lower? The range of values should be well known
> > and limited.  
> 
> Under the assumption that by skb_metadata you meant metadata_dst,

Can you show me in my email where I said skb_metadata?

> I think there are few reasons why i think is better to use skb extensions:
> 
> 1. Unlike skb extension, the metadata_dst deallaction is handled directly by the allocator.
> Since the sci and offloaded fields are shared between the MACsec driver and the offload driver
> (in our case mlx5 driver), for Rx, the metadata_dst allocation is done in the mlx5 driver,
> while the dealloction should be done in the MACsec driver.
> This is undesired behavior.

You allocate metadata skb once and then attach it to the skbs.

> 2. medadata_dst is attached to the skb using skb_dst_set(), which set the slow_gro bit.
> So there is no gain regarding slow_gro flow.
> 
> 3. metadata_dst allocation require much more memory than needed for MACsec use case
> (mainly because struct dst_entry which seems redundant for this case).