Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7;
MIME-Version: 1.0
References: <20230915104156.3406380-1-make_ruc2021@163.com>
 <CANn89iJyktWcztc76Pw16MP-k-DfSjstW+WFgRxwUat7p25CGw@mail.gmail.com>
 <16461255-c2c0-2ffd-f031-5b7a1f67bf7e@mojatatu.com> <CANn89i++j0-QJ1WE=RO4_ucN9k-DgqK52jLSTcz_s_DmFiAnFw@mail.gmail.com>
In-Reply-To: <CANn89i++j0-QJ1WE=RO4_ucN9k-DgqK52jLSTcz_s_DmFiAnFw@mail.gmail.com>
From:   Jamal Hadi Salim <jhs@mojatatu.com>
Date:   Fri, 15 Sep 2023 18:55:05 -0400
Message-ID: <CAM0EoMmzCBewv4hfWQrNb+gaO_+aR7jnMbsQScN+FEchTdWXmw@mail.gmail.com>
Subject: Re: [PATCH] net: sched: drr: dont intepret cls results when asked to drop
To:     Eric Dumazet <edumazet@google.com>
Cc:     Pedro Tammela <pctammela@mojatatu.com>,
        Ma Ke <make_ruc2021@163.com>, xiyou.wangcong@gmail.com,
        jiri@resnulli.us, davem@davemloft.net, kuba@kernel.org,
        pabeni@redhat.com, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org, Victor Nogueira <victor@mojatatu.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Fri, Sep 15, 2023 at 11:06=E2=80=AFAM Eric Dumazet <edumazet@google.com>=
 wrote:
>
> On Fri, Sep 15, 2023 at 5:03=E2=80=AFPM Pedro Tammela <pctammela@mojatatu=
.com> wrote:
> >
> > On 15/09/2023 09:55, Eric Dumazet wrote:
> > > On Fri, Sep 15, 2023 at 12:42=E2=80=AFPM Ma Ke <make_ruc2021@163.com>=
 wrote:
> > >>
> > >> If asked to drop a packet via TC_ACT_SHOT it is unsafe to
> > >> assume res.class contains a valid pointer.
> > >>
> > >> Signed-off-by: Ma Ke <make_ruc2021@163.com>
> > >> ---
> > >>   net/sched/sch_drr.c | 2 ++
> > >>   1 file changed, 2 insertions(+)
> > >>
> > >> diff --git a/net/sched/sch_drr.c b/net/sched/sch_drr.c
> > >> index 19901e77cd3b..2b854cb6edf9 100644
> > >> --- a/net/sched/sch_drr.c
> > >> +++ b/net/sched/sch_drr.c
> > >> @@ -309,6 +309,8 @@ static struct drr_class *drr_classify(struct sk_=
buff *skb, struct Qdisc *sch,
> > >>          *qerr =3D NET_XMIT_SUCCESS | __NET_XMIT_BYPASS;
> > >>          fl =3D rcu_dereference_bh(q->filter_list);
> > >>          result =3D tcf_classify(skb, NULL, fl, &res, false);
> > >> +       if (result =3D=3D TC_ACT_SHOT)
> > >> +               return NULL;
> > >>          if (result >=3D 0) {
> > >>   #ifdef CONFIG_NET_CLS_ACT
> > >>                  switch (result) {
> > >> --
> > >> 2.37.2
> > >>
> > >
> > >   I do not see a bug, TC_ACT_SHOT is handled in the switch (result) j=
ust fine
> > > at line 320 ?
> >
> > Following the code path (with CONFIG_NET_CLS_ACT=3Dn in mind), it looks
> > like there are a couple of places which return TC_ACT_SHOT before
> > calling any classifiers, which then would cause some qdiscs to look int=
o
> > a uninitialized 'struct tcf_result res'.
> > I could be misreading it... But if it's the problem the author is tryin=
g
> > to fix, the obvious way to do it would be:
> >         struct tcf_result res =3D {};
>
> CONFIG_NET_CLS_ACT=3Dn, how come TC_ACT_SHOT could be used ?
>
> Can we get rid of CONFIG_NET_CLS_ACT, this seems obfuscation to me at
> this point.

The problem is the verdict vs return code are intermixed - not saying
this was fixing anything useful.
We discussed this in the past after/during commit
caa4b35b4317d5147b3ab0fbdc9c075c7d2e9c12
Victor worked on a patch to resolve that. Victor, maybe revive that
patch and post as RFC?


cheers,
jamal