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: <1623058534-78782-1-git-send-email-chengshuyi@linux.alibaba.com>
 <CANn89iLNf+73MsPH7O7wX3PrN26FVLcjw_SmsN6jNwnjrYg4KQ@mail.gmail.com>
 <0e938649-986d-ce79-e3c4-1f29bdcb64e0@linux.alibaba.com> <CANn89iKnT5Ebk5vovFJKHY4Fe7ERkN3ak_Nkqyc=vXL=VWvyPg@mail.gmail.com>
 <258e3c94-f479-509c-a4b0-5a881779dd14@linux.alibaba.com>
In-Reply-To: <258e3c94-f479-509c-a4b0-5a881779dd14@linux.alibaba.com>
From:   Eric Dumazet <edumazet@google.com>
Date:   Thu, 10 Jun 2021 12:04:37 +0200
Message-ID: <CANn89iL8bkiJdDH110U5uD4mSJM7v9fU8xFa9BXAu4wCf+y13Q@mail.gmail.com>
Subject: Re: [PATCH net-next] net: tcp: Updating MSS, when the sending window
 is smaller than MSS.
To:     Shuyi Cheng <chengshuyi@linux.alibaba.com>
Cc:     David Miller <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>,
        Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
        David Ahern <dsahern@kernel.org>,
        netdev <netdev@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        kernel-janitors@vger.kernel.org,
        Mao Wenan <wenan.mao@linux.alibaba.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Jun 10, 2021 at 8:00 AM Shuyi Cheng
<chengshuyi@linux.alibaba.com> wrote:

> Thank you very much for your reply!
>
> Maybe it's not clear enough that I described it. The scenario where the
> above problem occurs is precisely because the tcp server sets the size
> of RCVBUFF to be smaller after the connection is established. Here is a
> sample code that caused the problem.
>
> # default tcp_rmem is 87380

Except that this value is overridden at connection establishment.

tcp_rmem[1] is only a floor value, say if you want a reasonable value
even if MSS == 100

> tcpServerSocket= socket.socket(socket.AF_INET, socket.SOCK_STREAM)
> tcpServerSocket.bind(server_addr)
> tcpServerSocket.listen()
> while True:
>      connection,client_addr = tcpServerSocket.accept()
>      # Shrink rmem
>      connection.setsockopt(socket.SOL_SOCKET, socket.SO_RCVBUF, 16*1024)
>
> Therefore, when the developer calls the sock_setsockopt function to
> reset RCVBUF, we can use sock to determine the TCP state. When in the
> connected state, it is not allowed to set RCVBUF smaller than mss.
>

Sure, but the application can _also_ set SO_RCVBUF before listen() or connect()

We can not have assumptions about SO_RCVBUF values and socket states.
Otherwise we would have to add some sk_rcvbuf adjustments every time
the socket state is changed.