Received-SPF: pass (google.com: domain of linux-kernel+bounces-187668-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Date: Thu, 23 May 2024 10:52:25 -0400
From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
To: Chengen Du <chengen.du@canonical.com>, 
 Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Cc: Paolo Abeni <pabeni@redhat.com>, 
 davem@davemloft.net, 
 edumazet@google.com, 
 kuba@kernel.org, 
 alexandre.ferrieux@orange.com, 
 netdev@vger.kernel.org, 
 linux-kernel@vger.kernel.org
Message-ID: <664f5829d6485_1b5d2429420@willemb.c.googlers.com.notmuch>
In-Reply-To: <CAPza5qfam1DBffBJWjOrMsSW8J+6ans7apeX=64YVCixBWX-tQ@mail.gmail.com>
References: <20240520070348.26725-1-chengen.du@canonical.com>
 <664b97e8abe7a_12b4762946f@willemb.c.googlers.com.notmuch>
 <CAPza5qcGyfcUYOoznci4e=1eaScVTgkzAhXfKSG3bTzC=aOwew@mail.gmail.com>
 <eaf33ba66cbdc639b0209b232f892ec8a52a1f21.camel@redhat.com>
 <664ca1651b66_14f7a8294cb@willemb.c.googlers.com.notmuch>
 <CAPza5qfZ8JPkt4Ez1My=gfpT7VfHo75N01fLQdFaojBv2whi8w@mail.gmail.com>
 <664e3be092d6a_184f2f29441@willemb.c.googlers.com.notmuch>
 <92edf27b-a2b9-4072-b8a4-0d7fde303151@orange.com>
 <CAPza5qfam1DBffBJWjOrMsSW8J+6ans7apeX=64YVCixBWX-tQ@mail.gmail.com>
Subject: Re: [PATCH] af_packet: Handle outgoing VLAN packets without hardware
 offloading
Precedence: bulk
Mime-Version: 1.0
Content-Type: text/plain;
 charset=utf-8
Content-Transfer-Encoding: 7bit

Chengen Du wrote:
> Hi Willem, all,
> 
> Thank you for highlighting the QinQ and L2.5 issues.
> These are areas I am not very familiar with, and I appreciate your guidance.
> 
> To address the QinQ and L2.5 issues, the third approach seems like a
> promising solution.
> If I understand correctly, in the QinQ scenario, we need to preserve
> the link layer header because it includes two VLAN tags.
> For the L2.5 issue, we can adjust by pulling mac_len instead of
> skb_network_offset.
> In summary, we may need to retain the link layer header to enable
> receivers to parse different protocol scenarios.
> 
> Although this approach can resolve all issues, it requires the
> receiver's cooperation to implement the parsing logic of the link
> layer header.

I was about to bring up the same.

Existing BPF filters may expect L3 headers. A change like this will
have to be opt-in through a socket option.

Since I see no better option to support all L2.5 protocols, if this
is something that tcpdump/pcap would use, we should do it.

> I am concerned that this implementation may take time and adding it
> directly into the kernel could place time pressure on existing users.
> 
> I would like to propose some ideas and welcome your thoughts on them.
> Firstly, I suggest we address the VLAN issue using the second
> approach, as it appears to be a bug and can be resolved without
> affecting current users.

Agreed. Let me respond in more detail to your original patch.

> Secondly, we could introduce the link layer header preservation via a
> new packet socket flag.
> Receivers could implement and test this by enabling the socket flag.
> This way, we avoid disrupting existing receiver behavior while
> providing the capability to parse more complex protocols.