RE: [PATCH net v3] tun: add missing rx stats accounting in tun_xdp_act

[wangyunjian <wangyunjian@huawei.com>]

> -----Original Message-----
> From: Willem de Bruijn [mailto:willemdebruijn.kernel@gmail.com]
> Sent: Thursday, January 18, 2024 10:44 PM
> To: wangyunjian <wangyunjian@huawei.com>; Willem de Bruijn
> <willemdebruijn.kernel@gmail.com>; jasowang@redhat.com;
> kuba@kernel.org; davem@davemloft.net
> Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org; xudingke
> <xudingke@huawei.com>
> Subject: RE: [PATCH net v3] tun: add missing rx stats accounting in tun_xdp_act
> 
> wangyunjian wrote:
> > > -----Original Message-----
> > > From: Willem de Bruijn [mailto:willemdebruijn.kernel@gmail.com]
> > > Sent: Wednesday, January 17, 2024 11:42 PM
> > > To: wangyunjian <wangyunjian@huawei.com>;
> > > willemdebruijn.kernel@gmail.com; jasowang@redhat.com;
> > > kuba@kernel.org; davem@davemloft.net
> > > Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org; xudingke
> > > <xudingke@huawei.com>; wangyunjian <wangyunjian@huawei.com>
> > > Subject: Re: [PATCH net v3] tun: add missing rx stats accounting in
> > > tun_xdp_act
> > >
> > > Yunjian Wang wrote:
> > > > The TUN can be used as vhost-net backend, and it is necessary to
> > > > count the packets transmitted from TUN to vhost-net/virtio-net.
> > > > However, there are some places in the receive path that were not
> > > > taken into account when using XDP. The commit 8ae1aff0b331
> > > > ("tuntap: split out XDP logic") only includes dropped counter for
> > > > XDP_DROP, XDP_ABORTED, and invalid XDP actions. It would be
> > > > beneficial to also include new accounting for successfully
> > > > received bytes using dev_sw_netstats_rx_add and introduce new
> > > > dropped counter for XDP errors
> > > on XDP_TX and XDP_REDIRECT.
> > >
> > > From the description it is clear that these are two separate changes
> > > wrapped into one patch. I should have flagged this previously.
> >
> > Do I need to split these two modifications into 2 patches?
> > 1. only fix dropped counter
> > 2. add new accounting for successfully received bytes
> > Or:
> > Only fix dropped counter?
> 
> It's definitely good to fix both.
> 
> It might be a bit pedantic, but two separate patches is more correct.
> 
> The second fix, add missing byte counter, goes back to the original introduction
> of XDP for tun, so has a different tag:
> 
> Fixes: 761876c857cb ("tap: XDP support")

OK, I will update it to 2 patches.
Thanks

> 
> >
> > >
> > > Ack on returning the error counter that was previously present and
> > > matches the Fixes tag.
> > >
> > > For the second change, I had to check a few other XDP capable
> > > drivers to verify that it is indeed common to count such packets.
> > >
> > > > Fixes: 8ae1aff0b331 ("tuntap: split out XDP logic")
> > > > Signed-off-by: Yunjian Wang <wangyunjian@huawei.com>
> > > > ---
> > > > v3: update commit log and code
> > > > v2: add Fixes tag
> > > > ---
> > > >  drivers/net/tun.c | 14 +++++++++-----
> > > >  1 file changed, 9 insertions(+), 5 deletions(-)
> > > >
> > > > diff --git a/drivers/net/tun.c b/drivers/net/tun.c index
> > > > afa5497f7c35..0704a17e74e1 100644
> > > > --- a/drivers/net/tun.c
> > > > +++ b/drivers/net/tun.c
> > > > @@ -1625,18 +1625,15 @@ static struct sk_buff
> > > > *__tun_build_skb(struct tun_file *tfile,  static int
> > > > tun_xdp_act(struct tun_struct *tun, struct
> > > bpf_prog *xdp_prog,
> > > >  		       struct xdp_buff *xdp, u32 act)  {
> > > > -	int err;
> > > > +	unsigned int datasize = xdp->data_end - xdp->data;
> > > > +	int err = 0;
> > > >
> > > >  	switch (act) {
> > > >  	case XDP_REDIRECT:
> > > >  		err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
> > > > -		if (err)
> > > > -			return err;
> > > >  		break;
> > > >  	case XDP_TX:
> > > >  		err = tun_xdp_tx(tun->dev, xdp);
> > > > -		if (err < 0)
> > > > -			return err;
> > > >  		break;
> > > >  	case XDP_PASS:
> > > >  		break;
> > > > @@ -1651,6 +1648,13 @@ static int tun_xdp_act(struct tun_struct
> > > > *tun,
> > > struct bpf_prog *xdp_prog,
> > > >  		break;
> > > >  	}
> > > >
> > > > +	if (err < 0) {
> > > > +		act = err;
> > > > +		dev_core_stats_rx_dropped_inc(tun->dev);
> > > > +	} else if (act == XDP_REDIRECT || act == XDP_TX) {
> > > > +		dev_sw_netstats_rx_add(tun->dev, datasize);
> > > > +	}
> > > > +
> > >
> > > Let's avoid adding yet another branch and just do these operations
> > > in the case statements, like XDP_DROP.
> >
> > Fix it like this?
> 
> Perhaps avoid computing datasize is all paths, when it is not used in common
> XDP_PASS, high performance XDP_DROP and a few others. Not sure whether
> (all) compilers would optimze that.
> 
>         dev_core_stats_rx_dropped_inc(tun->dev, xdp,
> 
                                      xdp->data_end - xdp->data);
OK, I agree.

> 
> 
> > --- a/drivers/net/tun.c
> > +++ b/drivers/net/tun.c
> > @@ -1625,18 +1625,25 @@ static struct sk_buff *__tun_build_skb(struct
> > tun_file *tfile,  static int tun_xdp_act(struct tun_struct *tun, struct
> bpf_prog *xdp_prog,
> >                        struct xdp_buff *xdp, u32 act)  {
> > +       unsigned int datasize = xdp->data_end - xdp->data;
> >         int err;
> >
> >         switch (act) {
> >         case XDP_REDIRECT:
> >                 err = xdp_do_redirect(tun->dev, xdp, xdp_prog);
> > -               if (err)
> > +               if (err) {
> > +                       dev_core_stats_rx_dropped_inc(tun->dev);
> >                         return err;
> > +               }
> > +               dev_sw_netstats_rx_add(tun->dev, datasize);
> >                 break;
> >         case XDP_TX:
> >                 err = tun_xdp_tx(tun->dev, xdp);
> > -               if (err < 0)
> > +               if (err < 0) {
> > +                       dev_core_stats_rx_dropped_inc(tun->dev);
> >                         return err;
> > +               }
> > +               dev_sw_netstats_rx_add(tun->dev, datasize);
> >                 break;
> >         case XDP_PASS:
> >
> > >
> > > >  	return act;
> > > >  }
> > > >
> > > > --
> > > > 2.41.0
> > > >
> > >
> > >
> >
> 
>