[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZAm+ovPoEFfESOQI@nanopsycho>
Date: Thu, 9 Mar 2023 12:10:26 +0100
From: Jiri Pirko <jiri@...nulli.us>
To: Jianguo Wu <wujianguo106@....com>
Cc: netdev <netdev@...r.kernel.org>,
"edumazet@...gle.com >> Eric Dumazet" <edumazet@...gle.com>,
davem@...emloft.net, daniel@...earbox.net,
Florian Westphal <fw@...len.de>
Subject: Re: [PATCH net-next] ipvlan: Make skb->skb_iif track skb->dev for
l3s mode
Thu, Mar 09, 2023 at 03:03:36AM CET, wujianguo106@....com wrote:
>From: Jianguo Wu <wujianguo@...natelecom.cn>
>
>For l3s mode, skb->dev is set to ipvlan interface in ipvlan_nf_input():
> skb->dev = addr->master->dev
>but, skb->skb_iif remain unchanged, this will cause socket lookup failed
>if a target socket is bound to a interface, like the following example:
>
> ip link add ipvlan0 link eth0 type ipvlan mode l3s
> ip addr add dev ipvlan0 192.168.124.111/24
> ip link set ipvlan0 up
>
> ping -c 1 -I ipvlan0 8.8.8.8
> 100% packet loss
>
>This is because there is no match sk in __raw_v4_lookup() as sk->sk_bound_dev_if != dif(skb->skb_iif).
>Fix this by make skb->skb_iif track skb->dev in ipvlan_nf_input().
>
>Fixes: c675e06a98a4 ("ipvlan: decouple l3s mode dependencies from other modes").
>
>Signed-off-by: Jianguo Wu <wujianguo@...natelecom.cn>
>---
> drivers/net/ipvlan/ipvlan_l3s.c | 1 +
> 1 file changed, 1 insertion(+)
>
>diff --git a/drivers/net/ipvlan/ipvlan_l3s.c b/drivers/net/ipvlan/ipvlan_l3s.c
>index 943d26cbf39f..71712ea25403 100644
>--- a/drivers/net/ipvlan/ipvlan_l3s.c
>+++ b/drivers/net/ipvlan/ipvlan_l3s.c
>@@ -101,6 +101,7 @@ static unsigned int ipvlan_nf_input(void *priv, struct sk_buff *skb,
> goto out;
>
> skb->dev = addr->master->dev;
>+ skb->skb_iif = skb->dev->ifindex;
I think we should have a macro to replace skb->dev which would handle
this too. It's on multiple places.
Anyway, this patch looks fine to me.
Reviewed-by: Jiri Pirko <jiri@...dia.com>
> len = skb->len + ETH_HLEN;
> ipvlan_count_rx(addr->master, len, true, false);
> out:
>--
>1.8.3.1
>
Powered by blists - more mailing lists