Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210914143515.106394-1-yan2228598786@gmail.com>
In-Reply-To: <20210914143515.106394-1-yan2228598786@gmail.com>
From:   Cong Wang <xiyou.wangcong@gmail.com>
Date:   Mon, 20 Sep 2021 09:54:02 -0700
Message-ID: <CAM_iQpWLPvSmZD4CTmzSoor04xfdkvZuDhF=_CCaumT7XiaN7g@mail.gmail.com>
Subject: Re: [PATCH] tcp: tcp_drop adds `SNMP` and `reason` parameter for tracing
To:     Zhongya Yan <yan2228598786@gmail.com>
Cc:     Eric Dumazet <edumazet@google.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Brendan Gregg <brendan.d.gregg@gmail.com>,
        Linux Kernel Network Developers <netdev@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Jakub Kicinski <kuba@kernel.org>,
        Ingo Molnar <mingo@redhat.com>,
        David Miller <davem@davemloft.net>,
        Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
        dsahern@kernel.org, Yonghong Song <yhs@fb.com>, 2228598786@qq.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, Sep 14, 2021 at 7:38 AM Zhongya Yan <yan2228598786@gmail.com> wrote=
:
>
> When we wanted to trace the use of the `tcp_drop(struct sock *sk, struct =
sk_buff *skb)` function, we didn't know why `tcp` was deleting `skb'. To so=
lve this problem, I updated the function `tcp_drop(struct sock *sk, struct =
sk_buff *skb, int field, const char *reason)`.
> This way you can understand the reason for the deletion based on the prom=
pt message.
> `field`: represents the SNMP-related value
> `reason`: represents the reason why `tcp` deleted the current `skb`, and =
contains some hints.
> Of course, if you want to know more about the reason for updating the cur=
rent function, you can check: https://www.brendangregg.com/blog/2018-05-31/=
linux-tcpdrop.html

I think you fail to explain why only TCP needs it? This should
be useful for all kinds of drops, not just TCP, therefore you should
consider extending net/core/drop_monitor.c instead of just tcp_drop().

Also, kernel does not have to explain it in strings, those SNMP
counters are already available for user-space, so kernel could
just use SNMP enums and let user-space interpret them. In many
cases, you are just adding strings for those SNMP enums.

Thanks.