Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221029130957.1292060-1-imagedong@tencent.com>
 <20221029130957.1292060-3-imagedong@tencent.com> <CANn89i+mK2VG2VnxPduSREHp70gJLP7oNp3SUoMW-YaSd7jJRA@mail.gmail.com>
In-Reply-To: <CANn89i+mK2VG2VnxPduSREHp70gJLP7oNp3SUoMW-YaSd7jJRA@mail.gmail.com>
From:   Menglong Dong <menglong8.dong@gmail.com>
Date:   Mon, 31 Oct 2022 11:08:19 +0800
Message-ID: <CADxym3a2qsM_2s_thwv93QLr-KP4eDcqPaK6LPUGnhenraKZ_w@mail.gmail.com>
Subject: Re: [PATCH net-next 2/9] net: tcp: add 'drop_reason' field to struct tcp_skb_cb
To:     Eric Dumazet <edumazet@google.com>
Cc:     kuba@kernel.org, davem@davemloft.net, pabeni@redhat.com,
        yoshfuji@linux-ipv6.org, dsahern@kernel.org, imagedong@tencent.com,
        kafai@fb.com, asml.silence@gmail.com, keescook@chromium.org,
        linux-kernel@vger.kernel.org, netdev@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hello,

On Sat, Oct 29, 2022 at 11:23 PM Eric Dumazet <edumazet@google.com> wrote:
>
> On Sat, Oct 29, 2022 at 6:11 AM <menglong8.dong@gmail.com> wrote:
> >
> > From: Menglong Dong <imagedong@tencent.com>
> >
> > Because of the long call chain on processing skb in TCP protocol, it's
> > hard to pass the drop reason to the code where the skb is freed.
> >
> > Therefore, we can store the drop reason to skb->cb, and pass it to
> > kfree_skb_reason(). I'm lucky, the struct tcp_skb_cb still has 4 bytes
> > spare space for this purpose.
>
> No, we have needs for this space for future use.
>
> skb->cb[] purpose is to store semi-permanent info, for skbs that stay
> in a queue.
>

May I use it for a while? Or, can I make it a union with
some field, such as 'header'? As the 'drop_reason' field will
be set only if it is going to be freed.

> Here, you need a state stored only in the context of the current context.
> Stack variables are better.

It's hard to get the drop reason through stack variables,
especially some functions in TCP protocol, such as:

  tcp_rcv_synsent_state_process
  tcp_timewait_state_process
  tcp_conn_request
  tcp_rcv_state_process

And it will mess the code a little up, just like what
you comment in this series:

https://lore.kernel.org/netdev/20220617100514.7230-1-imagedong@tencent.com/

I hope to complete this part, or I think I can't move
ahead :/

Thanks!
Menglong Dong