Commit d31d38a0 authored Dec 15, 2017 by David S. Miller

Merge branch 'ipvlan-packet-scrub'



Mahesh Bandewar says:

====================
ipvlan: packet scrub

While crossing namespace boundary IPvlan aggressively scrubs packets.
This is creating problems. First thing is that scrubbing changes the
packet type in skb meta-data to PACKET_HOST. This causes erroneous
packet delivery when dev_forward_skb() has already marked the packet
type as OTHER_HOST.

On the egress side scrubbing just before calling dev_queue_xmit()
creates another set of problems. Scrubbing remove skb->sk so the
prio update gets missed and more seriously, socket back-pressure
fails making TSQ not function correctly.

The first patch in the series just reverts the earlier change which
was adding a mac-check, but that is unnecessary if packet_type that
dev_forward_skb() has set is honored. The second path removes two of
the scrubs which are causing problems described above.
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

parents 8ce38aeb c0d451c8

drivers/net/ipvlan/ipvlan_core.c

+2 −6

Original line number	Diff line number	Diff line
		@@ -315,17 +315,13 @@ static int ipvlan_rcv_frame(struct ipvl_addr addr, struct sk_buff *pskb,

		*pskb = skb;
		}
		ipvlan_skb_crossing_ns(skb, dev);

		if (local) {
		skb->pkt_type = PACKET_HOST;
		if (dev_forward_skb(ipvlan->dev, skb) == NET_RX_SUCCESS)
		success = true;
		} else {
		if (!ether_addr_equal_64bits(eth_hdr(skb)->h_dest,
		ipvlan->phy_dev->dev_addr))
		skb->pkt_type = PACKET_OTHERHOST;

		skb->dev = dev;
		ret = RX_HANDLER_ANOTHER;
		success = true;
		}
		@@ -590,7 +586,7 @@ static int ipvlan_xmit_mode_l2(struct sk_buff skb, struct net_device dev)
		return NET_XMIT_SUCCESS;
		}

		ipvlan_skb_crossing_ns(skb, ipvlan->phy_dev);
		skb->dev = ipvlan->phy_dev;
		return dev_queue_xmit(skb);
		}