Re: [RFC PATCH net-next 3/3] net: dsa: never skip VLAN configuration

[Vladimir Oltean <vladimir.oltean@nxp.com>]

Hi Martin,

On Wed, Jul 06, 2022 at 09:57:40PM +0200, Martin Blumenstingl wrote:
> Hi Vladimir,
> 
> On Wed, Jul 6, 2022 at 6:45 PM Vladimir Oltean <vladimir.oltean@nxp.com> wrote:
> [...]
> > Somehow we should do something to make sure that the OpenWRT devices are
> > able to run the selftests, because there's a large number of DSA switches
> > intended for that segment and we should all be onboard (easily).
> 
> I could work on this on the OpenWrt side.
> But this would require me to get any test working at all...
> 
> I am struggling with not seeing any ping responses.
> Instead of adding VLANs and others to the mix I started with seemingly
> simple commands while connecting an Ethernet cable between lan1 and
> lan2 and another cable between lan3 and lan4:
> 1) give each port a unique MAC address, which is not the default on my
> device under test
> ip link set dev lan1 address 6a:88:f1:99:e1:81
> ip link set dev lan2 address 6a:88:f1:99:e1:82
> ip link set dev lan3 address 6a:88:f1:99:e1:83
> ip link set dev lan4 address 6a:88:f1:99:e1:84
> 
> 2) set up IP addresses on LAN1 and LAN2
> ip addr add 192.168.2.1/24 brd + dev lan1
> ip addr add 192.168.2.2/24 brd + dev lan2
> 
> 3) bring up the interfaces
> ip link set up dev lan1
> ip link set up dev lan2
> 
> 4) start tcpdump on LAN1
> tcpdump -i lan1 &
> 
> 5) start ping from LAN2 to LAN1
> ping -I lan2 192.168.2.1
> 
> result:
> PING 192.168.2.1 (192.168.2.1): 56 data bytes
> 20:02:01.522977 ARP, Request who-has 192.168.2.1 tell 192.168.2.2, length 46
> 20:02:02.569234 ARP, Request who-has 192.168.2.1 tell 192.168.2.2, length 46
> 20:02:03.609132 ARP, Request who-has 192.168.2.1 tell 192.168.2.2, length 46
> 20:02:05.524200 ARP, Request who-has 192.168.2.1 tell 192.168.2.2, length 46
> 20:02:06.569226 ARP, Request who-has 192.168.2.1 tell 192.168.2.2, length 46
> ...repeats...
> 
> So LAN1 can see the ARP request from the ping on LAN2.
> But I am not seeing Linux trying to send a reply.

It won't reply, you either need a network namespace or a VRF to do
loopback IP networking. A VRF is a bit more complicated to do, here's a
netns setup:

ip netns add ns0
ip link set lan2 netns ns0
ip -n ns0 link set lan2 up
ip -n ns0 addr add 192.168.2.2/24 dev lan2
ip netns exec ns0 tcpdump -i lan2 -e -n
ping 192.168.2.2

> I already verified that nftables doesn't have any rules active (if it
> was I think it would rather result in tcpdump not seeing the who-has
> request):
> # nft list tables
> #
> 
> # grep "" /proc/sys/net/ipv4/icmp_echo_ignore_*
> /proc/sys/net/ipv4/icmp_echo_ignore_all:0
> /proc/sys/net/ipv4/icmp_echo_ignore_broadcasts:1
> 
> # ps | grep netif
> 3014 root      1308 S    grep netif
> #
> 
> To make things worse: if I let OpenWrt's netifd configure LAN1 and
> LAN2 as single ports with above IP addresses ping works.
> Something is odd and I am not sure how to find out what's wrong.

I'm not familiar with OpenWrt, sorry, I don't know what netifd does.
Also, it's curious that this works, are you sure that the ARP responses
and ICMP replies actually exit through the Ethernet port? ethtool -S
should show if the physical counters increment.

> > I wonder, would it be possible to set up a debian chroot?
> 
> I'm thinking of packaging the selftests as OpenWrt package and also
> providing all needed dependencies as OpenWrt packages.
> I think (or I hope, not sure yet) the ping interval is just a matter
> of a busybox config option.

I think it depends on busybox version. At least the latest one
https://github.com/mirror/busybox/blob/master/networking/ping.c#L970
seems to support fractions of a second as intervals, I didn't see any
restriction to sub-second values. In any case, the iputils version
certainly does work.