Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
In-Reply-To: <20180806.134116.1016660959593702560.davem@davemloft.net>
References: <1533556020-20778-1-git-send-email-laoar.shao@gmail.com> <20180806.134116.1016660959593702560.davem@davemloft.net>
From:   Yafang Shao <laoar.shao@gmail.com>
Date:   Tue, 7 Aug 2018 09:40:42 +0800
Message-ID: <CALOAHbBU4ntM7YBuD1QAXybmJkOa2K10WksxXDF91ci=TFc+pw@mail.gmail.com>
Subject: Re: [PATCH net-next] tcp: fix the calculation of sysctl_max_tw_buckets
 in tcp_sk_init()
To:     David Miller <davem@davemloft.net>
Cc:     Eric Dumazet <edumazet@google.com>,
        yanhaishuang@cmss.chinamobile.com, netdev <netdev@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, Aug 7, 2018 at 4:41 AM, David Miller <davem@davemloft.net> wrote:
> From: Yafang Shao <laoar.shao@gmail.com>
> Date: Mon,  6 Aug 2018 19:47:00 +0800
>
>> tcp_hashinfo.ehash_mask is always an odd number, which is set in function
>> alloc_large_system_hash(). See bellow,
>>       if (_hash_mask)
>>               *_hash_mask = (1 << log2qty) - 1; <<< always odd number
>>
>> Hence the local variable 'cnt' is a even number, as a result of that it is
>> no difference to do the incrementation here.
>>
>> Maybe the compiler could also optimize it, but this code is a little ugly.
>>
>> Fix: fee83d09 ("ipv4: Namespaceify tcp_max_syn_backlog knob")
>
> The correct tag is "Fixes: "

OK

>
>> @@ -2543,7 +2543,7 @@ static int __net_init tcp_sk_init(struct net *net)
>>       net->ipv4.sysctl_tcp_tw_reuse = 2;
>>
>>       cnt = tcp_hashinfo.ehash_mask + 1;
>> -     net->ipv4.tcp_death_row.sysctl_max_tw_buckets = (cnt + 1) / 2;
>> +     net->ipv4.tcp_death_row.sysctl_max_tw_buckets = cnt / 2;
>>       net->ipv4.tcp_death_row.hashinfo = &tcp_hashinfo;
>
> This is completely harmless, and does no harm.
>

It is harmless, meanwhile, it is useless and a little ugly.

By the way,  why do we init sysctl_max_tw_buckets to
(tcp_hashinfo.ehash_mask + 1) / 2 or (tcp_hashinfo.ehash_mask + 2) / 2
?
Per my pespective, this default value is a little huge, sometimes we
have to set this value to a lower value with sysctl, otherwise it may
easily cause tcp port exhaustion issue.
So why not just init it to a lower value ?

> You aren't "fixing" anything.