Yunjian Wang wrote:
> There are few places on the receive path where packet receives and packet
> drops were not accounted for. This patch fixes that issue.
>
> Fixes: 8ae1aff0b331 ("tuntap: split out XDP logic")
Before this commit this_cpu_inc(tun->pcpu_stats->rx_dropped) would
get called, so that is indeed a regression in that patch. Please add
that to the commit message.
This commit also adds new accounting of successfully received bytes
with dev_sw_netstats_rx_add. I don't know off the top of my head if
other devices account XDP_TX and XDP_REDIRECT in that counter. Either
way, good to be explicit about such subtle details. Oddly, before
8ae1aff0b331 those, too, would be attributed to rx_dropped, because
after the out: label.
> Signed-off-by: Yunjian Wang <[email protected]>
> ---
> v2: add Fixes tag
> ---
> drivers/net/tun.c | 12 ++++++++----
> 1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/net/tun.c b/drivers/net/tun.c
> index afa5497f7c35..232e5319ac77 100644
> --- a/drivers/net/tun.c
> +++ b/drivers/net/tun.c
> @@ -1626,17 +1626,14 @@ 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;
>
> 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;
err is uninitialized in this case?
> @@ -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);
> + }
> +
> return act;
> }
>
> --
> 2.41.0
>