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: <20200730210728.2051-1-f.fainelli@gmail.com> <CANn89iJETzud8PK7eTj=rXMSCjBtnmcSq1y0qF7EVK8b5M_vXA@mail.gmail.com>
 <2347a342-f0b0-903c-ebb6-6e95eb664864@gmail.com>
In-Reply-To: <2347a342-f0b0-903c-ebb6-6e95eb664864@gmail.com>
From:   Eric Dumazet <edumazet@google.com>
Date:   Thu, 30 Jul 2020 14:32:46 -0700
Message-ID: <CANn89iJ=x8eYs9+cGYuyScMo7AD3JZqr6Jp1oZfJg41fHej8JQ@mail.gmail.com>
Subject: Re: [PATCH net] tcp: Export tcp_write_queue_purge()
To:     Florian Fainelli <f.fainelli@gmail.com>
Cc:     netdev <netdev@vger.kernel.org>,
        "David S. Miller" <davem@davemloft.net>,
        Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>,
        Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
        Jakub Kicinski <kuba@kernel.org>,
        open list <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Jul 30, 2020 at 2:24 PM Florian Fainelli <f.fainelli@gmail.com> wrote:
>
> On 7/30/20 2:16 PM, Eric Dumazet wrote:
> > On Thu, Jul 30, 2020 at 2:07 PM Florian Fainelli <f.fainelli@gmail.com> wrote:
> >>
> >> After tcp_write_queue_purge() got uninlined with commit ac3f09ba3e49
> >> ("tcp: uninline tcp_write_queue_purge()"), it became no longer possible
> >> to reference this symbol from kernel modules.
> >>
> >> Fixes: ac3f09ba3e49 ("tcp: uninline tcp_write_queue_purge()")
> >> Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
> >> ---
> >>  net/ipv4/tcp.c | 1 +
> >>  1 file changed, 1 insertion(+)
> >>
> >> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> >> index 6f0caf9a866d..ea9d296a8380 100644
> >> --- a/net/ipv4/tcp.c
> >> +++ b/net/ipv4/tcp.c
> >> @@ -2626,6 +2626,7 @@ void tcp_write_queue_purge(struct sock *sk)
> >>         tcp_sk(sk)->packets_out = 0;
> >>         inet_csk(sk)->icsk_backoff = 0;
> >>  }
> >> +EXPORT_SYMBOL(tcp_write_queue_purge);
> >>
> >>  int tcp_disconnect(struct sock *sk, int flags)
> >>  {
> >> --
> >> 2.17.1
> >>
> >
> > Hmmm.... which module would need this exactly ?
>
> None in tree unfortunately, and I doubt it would be published one day.
> For consistency one could argue that given it used to be accessible, and
> other symbols within net/ipv4/tcp.c are also exported, so this should
> one be. Not going to hold that line of argumentation more than in this
> email, if you object to it, that would be completely fine with me.

:)

>
> >
> > How come it took 3 years to discover this issue ?
>
> We just upgraded our downstream kernel from 4.9 to 5.4 and this is why
> it took so long.

It is not because TCP used an inline function in the past that it
means we have to keep
the equivalent function available for all possible out-of-tree modules.

Sorry, we can not accept that out-of-tree modules use TCP stack like that.

You will have to carry this change locally. Or even better get rid of it.