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: <20220301144453.snstwdjy3kmpi4zf@begin> <CA+FuTSfi1aXiBr-fOQ+8XJPjCCTnqTicW2A3OUVfNHurfDL3jA@mail.gmail.com>
 <20220301150028.romzjw2b4aczl7kf@begin>
In-Reply-To: <20220301150028.romzjw2b4aczl7kf@begin>
From:   Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Date:   Tue, 1 Mar 2022 10:14:18 -0500
Message-ID: <CA+FuTSeZw228fsDj+YoSpu5sLaXsp+uR+N+qHrzZ4e3yMWhPKw@mail.gmail.com>
Subject: Re: [PATCH] SO_ZEROCOPY should rather return -ENOPROTOOPT
To:     Samuel Thibault <samuel.thibault@labri.fr>,
        Willem de Bruijn <willemdebruijn.kernel@gmail.com>,
        willemb@google.com, davem@davemloft.net, kuba@kernel.org,
        linux-kernel@vger.kernel.org,
        Network Development <netdev@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Mar 1, 2022 at 10:00 AM Samuel Thibault
<samuel.thibault@labri.fr> wrote:
>
> Willem de Bruijn, le mar. 01 mars 2022 09:51:45 -0500, a ecrit:
> > On Tue, Mar 1, 2022 at 9:44 AM Samuel Thibault <samuel.thibault@labri.fr> wrote:
> > >
> > > ENOTSUPP is documented as "should never be seen by user programs", and
> > > is not exposed in <errno.h>, so applications cannot safely check against
> > > it. We should rather return the well-known -ENOPROTOOPT.
> > >
> > > Signed-off-by: Samuel Thibault <samuel.thibault@labri.fr>
> > >
> > > diff --git a/net/core/sock.c b/net/core/sock.c
> > > index 4ff806d71921..6e5b84194d56 100644
> > > --- a/net/core/sock.c
> > > +++ b/net/core/sock.c
> > > @@ -1377,9 +1377,9 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
> > >                         if (!(sk_is_tcp(sk) ||
> > >                               (sk->sk_type == SOCK_DGRAM &&
> > >                                sk->sk_protocol == IPPROTO_UDP)))
> > > -                               ret = -ENOTSUPP;
> > > +                               ret = -ENOPROTOOPT;
> > >                 } else if (sk->sk_family != PF_RDS) {
> > > -                       ret = -ENOTSUPP;
> > > +                       ret = -ENOPROTOOPT;
> > >                 }
> > >                 if (!ret) {
> > >                         if (val < 0 || val > 1)
> >
> > That should have been a public error code. Perhaps rather EOPNOTSUPP.
> >
> > The problem with a change now is that it will confuse existing
> > applications that check for -524 (ENOTSUPP).
>
> They were not supposed to hardcord -524...
>
> Actually, they already had to check against EOPNOTSUPP to support older
> kernels, so EOPNOTSUPP is not supposed to pose a problem.

Which older kernels returned EOPNOTSUPP on SO_ZEROCOPY?

There is prior art in changing this error code when it leaks to
userspace, such as commit 2230a7ef5198 ("drop_monitor: Use correct
error code") and commit 4a5cdc604b9c ("net/tls: Fix return values to
avoid ENOTSUPP").

I certainly wrote code in the past that explicitly checks for 524
(ENOTSUPP). But do not immediately see public code that does this.
Indeed, udpgso_bench_tx checks for both these codes.

So it's probably fine. Note that there is some risk. But no more than
with those prior commits.