linux-kernel - Re: [PATCH net-next] net: init sk

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <45bfd7ce-ecfa-8a4d-b1ed-8914a6b90093@gmail.com>
Date:   Tue, 24 Apr 2018 05:37:13 -0700
From:   Eric Dumazet <eric.dumazet@...il.com>
To:     Yafang Shao <laoar.shao@...il.com>
Cc:     David Miller <davem@...emloft.net>,
        Alexei Starovoitov <alexei.starovoitov@...il.com>,
        netdev@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH net-next] net: init sk_cookie for inet socket

On 04/24/2018 04:47 AM, Yafang Shao wrote:

> 
> Could you pls. explain the issue to me ?

Just run a synflood test on your host, it will definitely show the atomic
consuming most cpu cycles in inet_reqsk_alloc(), because of huge contention
on a cache line shared by all cpus.

Performance is reduced from ~5 Mpps to ~3.8 Mpps with 16 RX queues on my host.

atomic64_inc_return(&sock_net(sk)->cookie_gen) was not meant to be used in the
normal case (when a socket cookie is not ever requested/needed)