Received-SPF: pass (google.com: domain of linux-kernel+bounces-151648-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Message-ID: <d59a08c3-eaec-4d19-83bf-b085b4c56b04@gmail.com>
Date: Fri, 19 Apr 2024 17:17:21 +0200
Precedence: bulk
MIME-Version: 1.0
Subject: Re: [PATCH net v1 2/2] net: gro: add p_off param in *_gro_complete
To: Willem de Bruijn <willemdebruijn.kernel@gmail.com>, davem@davemloft.net,
 edumazet@google.com, kuba@kernel.org, pabeni@redhat.com, dsahern@kernel.org,
 aleksander.lobakin@intel.com, netdev@vger.kernel.org,
 linux-kernel@vger.kernel.org
References: <20240412152120.115067-1-richardbgobert@gmail.com>
 <20240412152120.115067-3-richardbgobert@gmail.com>
 <661ad2e8a7e95_3be9a7294a5@willemb.c.googlers.com.notmuch>
 <97a01bf9-99d2-4368-9ebd-1e1194c1d7fd@gmail.com>
 <662025813539e_c86472944@willemb.c.googlers.com.notmuch>
 <6736050b-1410-453e-8afa-55e4c8f34033@gmail.com>
 <6621692cbedd1_f648a29469@willemb.c.googlers.com.notmuch>
From: Richard Gobert <richardbgobert@gmail.com>
In-Reply-To: <6621692cbedd1_f648a29469@willemb.c.googlers.com.notmuch>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

Willem de Bruijn wrote:
> Richard Gobert wrote:
>> Willem de Bruijn wrote:> Since this INDIRECT_CALL_INET needs to know about the definitions of
>>> udp4_lib_lookup_skb and udp6_lib_lookup_skb anyway, we can just get
>>> rid of the whole udp_lookup_t type and function pointer passing?
>>>
>>> Or move the entire lookup to udp4_gro_complete/udp6_gro_complete and
>>> pass the sk to udp_gro_complete.
>>
>> This sounds like a really good idea, I like it. Although I think it may be
>> more relevant to net-next instead of this bug fix. I can post a
>> complementing patch to net-next later if that's OK with you
> 
> That's fine.
> 
> This patch as is is already very intrusive for a net patch. For a fix,
> could we do something smaller? Like add the network offset into
> NAPI_GRO_CB in inet_gro_callback and use that in udp4_lib_lookup_skb? 
> (and same for IPv6 of course).
> 

Yes, I think that's a good idea. I also think a smaller patch for net will
turn out to be more reliable.

I'll take the first commit in the net-next patch series and use it to fix
the UDP bug discussed here. I tested that version thoroughly for all
protocols supported in GRO, and it doesn't add any overhead for the TCP
common case.