Received-SPF: pass (google.com: domain of linux-kernel+bounces-59942-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Message-ID: <fe9f9fa2-8308-422f-b47f-951b0d7875a4@gmail.com>
Date: Fri, 9 Feb 2024 12:33:58 -0800
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH] net: remove check before __cgroup_bpf_run_filter_skb
Content-Language: en-US
To: Stanislav Fomichev <sdf@google.com>, Oliver Crumrine <ozlinuxc@gmail.com>
Cc: bpf@vger.kernel.org, linux-kernel@vger.kernel.org, thinker.li@gmail.com
References: <ngc7klapduckb67tsymb3blu2wlmdsjo4pa4gbaivgxezbwzxp@v7akqu7gbwl4>
 <ZcV1GgitdBUIcKJT@google.com>
 <3htegzrugq4xwlizizsaku6g2pzwhndcnxxxmji4fvblisiuro@icvcsa3mky3w>
 <ZcZ2ObDxRwZ-hKLb@google.com>
From: Kui-Feng Lee <sinquersw@gmail.com>
In-Reply-To: <ZcZ2ObDxRwZ-hKLb@google.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit


On 2/9/24 11:00, Stanislav Fomichev wrote:
> On 02/08, Oliver Crumrine wrote:
>> On Thu, Feb 08, 2024 at 04:43:06PM -0800, Stanislav Fomichev wrote:
>>> The check is here to make sure we only run this hook on non-req sockets.
>>> Dropping it would mean we'd be running the hook on the listeners
>>> instead. I don't think we want that.
>>
>> You are correct that we don't want to run the code on listeners. However
>> the check for that is in the function this macro calls,
>> __cgroup_bpf_run_filter_skb (the check is on line 1367 of
>> kernel/bpf/cgroup.c, for 6.8.0-rc3). The check doesn't need to be done
>> twice, so it can be removed in this macro.
> 
> Maybe we should instead remove "(!sk || !sk_fullsock(sk))" check from
> __cgroup_bpf_run_filter_skb? BPF_CGROUP_RUN_PROG_INET_EGRESS makes
> care of all those corner conditions. We just need to add those checks to
> BPF_CGROUP_RUN_PROG_INET_INGRESS.
> 
> Let me also CC Kui-Feng, he was touching this part recently in commit
> 223f5f79f2ce ("bpf, net: Check skb ownership against full socket.").
> 

Adding those checks in BPF_CGROUP_RUN_PROG_INET_INGRESS makes sense
to me if the point here is saving CPU cycles during runtime.