Re: [PATCH net 1/5] net: sched: cls_bpf: Undo tcf_bind_filter in case of an error

From: Jamal Hadi Salim <hadi@mojatatu.com>
To: Simon Horman <simon.horman@corigine.com>
Cc: Victor Nogueira <victor@mojatatu.com>,
	netdev@vger.kernel.org, jhs@mojatatu.com,
	 xiyou.wangcong@gmail.com, jiri@resnulli.us, davem@davemloft.net,
	 edumazet@google.com, kuba@kernel.org, pabeni@redhat.com,
	 pctammela@mojatatu.com, kernel@mojatatu.com
Subject: Re: [PATCH net 1/5] net: sched: cls_bpf: Undo tcf_bind_filter in case of an error
Date: Tue, 4 Jul 2023 16:55:25 -0400	[thread overview]
Message-ID: <CAAFAkD-WppW_Gf+Dfm=SSr62PNQwwngwXe2=XKo52AkWD=sSPA@mail.gmail.com> (raw)
In-Reply-To: <ZKSFrSW2zJZYelNj@corigine.com>

On Tue, Jul 4, 2023 at 4:48 PM Simon Horman <simon.horman@corigine.com> wrote:
>
> On Tue, Jul 04, 2023 at 12:14:52PM -0300, Victor Nogueira wrote:
> > If cls_bpf_offload errors out, we must also undo tcf_bind_filter that
> > was done in cls_bpf_set_parms.
> >
> > Fix that by calling tcf_unbind_filter in errout_parms.
> >
> > Fixes: eadb41489fd2 ("net: cls_bpf: add support for marking filters as hardware-only")
> >
>
> nit: no blank line here.
>
> > Signed-off-by: Victor Nogueira <victor@mojatatu.com>
> > Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
> > Reviewed-by: Pedro Tammela <pctammela@mojatatu.com>
> > ---
> >  net/sched/cls_bpf.c | 8 ++++++--
> >  1 file changed, 6 insertions(+), 2 deletions(-)
> >
> > diff --git a/net/sched/cls_bpf.c b/net/sched/cls_bpf.c
> > index 466c26df853a..4d9974b1b29d 100644
> > --- a/net/sched/cls_bpf.c
> > +++ b/net/sched/cls_bpf.c
> > @@ -409,7 +409,7 @@ static int cls_bpf_prog_from_efd(struct nlattr **tb, struct cls_bpf_prog *prog,
> >  static int cls_bpf_set_parms(struct net *net, struct tcf_proto *tp,
> >                            struct cls_bpf_prog *prog, unsigned long base,
> >                            struct nlattr **tb, struct nlattr *est, u32 flags,
> > -                          struct netlink_ext_ack *extack)
> > +                          bool *bound_to_filter, struct netlink_ext_ack *extack)
> >  {
> >       bool is_bpf, is_ebpf, have_exts = false;
> >       u32 gen_flags = 0;
> > @@ -451,6 +451,7 @@ static int cls_bpf_set_parms(struct net *net, struct tcf_proto *tp,
> >       if (tb[TCA_BPF_CLASSID]) {
> >               prog->res.classid = nla_get_u32(tb[TCA_BPF_CLASSID]);
> >               tcf_bind_filter(tp, &prog->res, base);
> > +             *bound_to_filter = true;
> >       }
> >
> >       return 0;
> > @@ -464,6 +465,7 @@ static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
> >  {
> >       struct cls_bpf_head *head = rtnl_dereference(tp->root);
> >       struct cls_bpf_prog *oldprog = *arg;
> > +     bool bound_to_filter = false;
> >       struct nlattr *tb[TCA_BPF_MAX + 1];
> >       struct cls_bpf_prog *prog;
> >       int ret;
>
> Please use reverse xmas tree - longest line to shortest - for
> local variable declarations in Networking code.
>

I think Ed's tool is actually wrong on this Simon.
The rule I know of is: initializations first then declarations -
unless it is documented elsewhere as not the case.

cheers,
jamal

> > @@ -505,7 +507,7 @@ static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
> >       prog->handle = handle;
> >
> >       ret = cls_bpf_set_parms(net, tp, prog, base, tb, tca[TCA_RATE], flags,
> > -                             extack);
> > +                             &bound_to_filter, extack);
> >       if (ret < 0)
> >               goto errout_idr;
> >
> > @@ -530,6 +532,8 @@ static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
> >       return 0;
> >
> >  errout_parms:
> > +     if (bound_to_filter)
> > +             tcf_unbind_filter(tp, &prog->res);
> >       cls_bpf_free_parms(prog);
> >  errout_idr:
> >       if (!oldprog)
>
> --
> pw-bot: changes-requested
>