Received-SPF: pass (google.com: domain of linux-kernel+bounces-145409-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Mon, 15 Apr 2024 11:00:32 -0400
From: Willem de Bruijn <willemdebruijn.kernel@gmail.com>
To: Alexander Duyck <alexander.duyck@gmail.com>, 
 Willem de Bruijn <willemdebruijn.kernel@gmail.com>
Cc: Richard Gobert <richardbgobert@gmail.com>, 
 davem@davemloft.net, 
 edumazet@google.com, 
 kuba@kernel.org, 
 pabeni@redhat.com, 
 dsahern@kernel.org, 
 aleksander.lobakin@intel.com, 
 netdev@vger.kernel.org, 
 linux-kernel@vger.kernel.org
Message-ID: <661d41106f996_c0c829445@willemb.c.googlers.com.notmuch>
In-Reply-To: <CAKgT0UfB+3DTjK7vq1uvG-2xtA53pw03ygJhwSG8j1bPtmYU8A@mail.gmail.com>
References: <20240412152120.115067-1-richardbgobert@gmail.com>
 <20240412152120.115067-2-richardbgobert@gmail.com>
 <661ad1136bc10_3be9a7294c2@willemb.c.googlers.com.notmuch>
 <CAKgT0UfB+3DTjK7vq1uvG-2xtA53pw03ygJhwSG8j1bPtmYU8A@mail.gmail.com>
Subject: Re: [PATCH net v1 1/2] net: gro: add flush check in
 udp_gro_receive_segment
Precedence: bulk
Mime-Version: 1.0
Content-Type: text/plain;
 charset=utf-8
Content-Transfer-Encoding: quoted-printable

Alexander Duyck wrote:
> On Sat, Apr 13, 2024 at 11:38=E2=80=AFAM Willem de Bruijn
> <willemdebruijn.kernel@gmail.com> wrote:
> >
> > Richard Gobert wrote:
> > > GRO-GSO path is supposed to be transparent and as such L3 flush che=
cks are
> > > relevant to all flows which call skb_gro_receive. This patch uses t=
he same
> > > logic and code from tcp_gro_receive but in the relevant flow path i=
n
> > > udp_gro_receive_segment.
> > >
> > > Fixes: 36707061d6ba ("udp: allow forwarding of plain (non-fragliste=
d) UDP GRO packets")
> > > Signed-off-by: Richard Gobert <richardbgobert@gmail.com>
> >
> > Reviewed-by: Willem de Bruijn <willemb@google.com>
> >
> > > ---
> > >  net/ipv4/udp_offload.c | 13 ++++++++++++-
> > >  1 file changed, 12 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/net/ipv4/udp_offload.c b/net/ipv4/udp_offload.c
> > > index 3498dd1d0694..1f4e08f43c4b 100644
> > > --- a/net/ipv4/udp_offload.c
> > > +++ b/net/ipv4/udp_offload.c
> > > @@ -471,6 +471,7 @@ static struct sk_buff *udp_gro_receive_segment(=
struct list_head *head,
> > >       struct sk_buff *p;
> > >       unsigned int ulen;
> > >       int ret =3D 0;
> > > +     int flush;
> > >
> > >       /* requires non zero csum, for symmetry with GSO */
> > >       if (!uh->check) {
> > > @@ -528,7 +529,17 @@ static struct sk_buff *udp_gro_receive_segment=
(struct list_head *head,
> > >                               skb_gro_postpull_rcsum(skb, uh,
> > >                                                      sizeof(struct =
udphdr));
> > >
> > > -                             ret =3D skb_gro_receive(p, skb);
> > > +                             flush =3D NAPI_GRO_CB(p)->flush;
> > > +
> > > +                             if (NAPI_GRO_CB(p)->flush_id !=3D 1 |=
|
> > > +                                 NAPI_GRO_CB(p)->count !=3D 1 ||
> > > +                                 !NAPI_GRO_CB(p)->is_atomic)
> > > +                                     flush |=3D NAPI_GRO_CB(p)->fl=
ush_id;
> > > +                             else
> > > +                                     NAPI_GRO_CB(p)->is_atomic =3D=
 false;
> > > +
> > > +                             if (flush || skb_gro_receive(p, skb))=

> > > +                                     ret =3D 1;
> >
> > UDP_L4 does not have the SKB_GSO_TCP_FIXEDID that uses is_atomic as
> > input.
> >
> > And I still don't fully internalize the flush_id logic after staring
> > at it for more than one coffee.
> =

> The flush_id field is there to indicate the difference between the
> current IPv4 ID of the previous IP header. It is meant to be used in
> conjunction with the is_atomic for the frame coalescing. Basically
> after the second frame we can decide the pattern either incrementing
> IPv4 ID or fixed, so on frames 3 or later we can decide to drop the
> frame if it doesn't follow that pattern.
> =

> > But even ignoring those, the flush signal of NAPI_GRO_CB(p)->flush
> > set the network layer must be followed, so ACK. Thanks for the fix.
> =

> I'm not sure about the placement of this code though. That is the one
> thing that seems off to me. Specifically this seems like it should be
> done before we start the postpull, not after. It should be something
> that can terminate the flow before we attempt to aggregate the UDP
> headers.

In principle agreed that we should conclude the flush checks before
doing prep for coalescing.

In practice it does not matter? NAPI_GRO_CB(skb)->csum will be ignored
if the packet gets flushed.=