bpf: Fix dev's rx stats for bpf_redirect_peer traffic

[ Upstream commit 024ee93 ]

Traffic redirected by bpf_redirect_peer() (used by recent CNIs like Cilium)
is not accounted for in the RX stats of supported devices (that is, veth
and netkit), confusing user space metrics collectors such as cAdvisor [0],
as reported by Youlun.

Fix it by calling dev_sw_netstats_rx_add() in skb_do_redirect(), to update
RX traffic counters. Devices that support ndo_get_peer_dev _must_ use the
@tstats per-CPU counters (instead of @lstats, or @dstats).

To make this more fool-proof, error out when ndo_get_peer_dev is set but
@tstats are not selected.

  [0] Specifically, the "container_network_receive_{byte,packet}s_total"
      counters are affected.

Fixes: 9aa1206 ("bpf: Add redirect_peer helper")
Reported-by: Youlun Zhang <zhangyoulun@bytedance.com>
Signed-off-by: Peilin Ye <peilin.ye@bytedance.com>
Co-developed-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Nikolay Aleksandrov <razor@blackwall.org>
Link: https://lore.kernel.org/r/20231114004220.6495-6-daniel@iogearbox.net
Signed-off-by: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>

net/core/dev.c

@@ -10054,6 +10054,14 @@ static int netdev_do_alloc_pcpu_stats(struct net_device *dev)
 {
 	void __percpu *v;
 
+	/* Drivers implementing ndo_get_peer_dev must support tstat
+	 * accounting, so that skb_do_redirect() can bump the dev's
+	 * RX stats upon network namespace switch.
+	 */
+	if (dev->netdev_ops->ndo_get_peer_dev &&
+	    dev->pcpu_stat_type != NETDEV_PCPU_STAT_TSTATS)
+		return -EOPNOTSUPP;
+
 	switch (dev->pcpu_stat_type) {
 	case NETDEV_PCPU_STAT_NONE:
 		return 0;

net/core/filter.c

@@ -2489,6 +2489,7 @@ int skb_do_redirect(struct sk_buff *skb)
 			     net_eq(net, dev_net(dev))))
 			goto out_drop;
 		skb->dev = dev;
+		dev_sw_netstats_rx_add(dev, skb->len);
 		return -EAGAIN;
 	}
 	return flags & BPF_F_NEIGH ?