23.05.0: IPQ40xx: When a mac address was seen on LAN, this mac address can't communicate on bridged devices

[StifflersMom]

Describe the bug

On my D-LINK DAP-2610 devices a Wifi Station can't communicate (eg. get an ip adress from a dhcp server on lan network), if the mac address was seen on the LAN port before.
I described this in #13218 before, but the steps to reproduce this were not clear.

OpenWrt version

r23497-6637af95aa

OpenWrt target/subtarget

ipq40xx/generic

Device

D-Link DAP 2610

Image kind

Official downloaded image

Steps to reproduce

Configure 2 APs as bridged to LAN, with DHCP server on the LAN network.
Let a wifi device (eg smartphone) talk to the network (eg using a ssid on AP 2)
Now switch to AP1 (the DAP-2610)

Actual behaviour

Packets don't reach the LAN for ~ 300 seconds, if the station was seen on LAN before

Expected behaviour

Packets should be always bridged.

Additional info

May be the issue starts with changed VLAN handling between 22.03 and 23.05 ...

Diffconfig

No response

Terms

• I am reporting an issue for OpenWrt, not an unsupported fork.

[brada4]

How the things are connected? ie like dhcp on 3rd party switch and openwrt's as basic antennas in workstations' places?

[StifflersMom]

The dhcp is connected to a 24 port POE+ switch, also both APs are connected to the 24port POE switch.
The same configuration works with 22.03.
If the smartphone doing the dhcp discover broadcast, you cant see the broadcast outside the AP. It doesn't reach the LAN.

[brada4]

cat /sys/class/net/br-lan/bridge/ageing_time - it will be in 100ths
brctl setageing br-lan 5 - set expiry interval smaller, too small means more frames will be broadcast to everybody around.

[StifflersMom]

This does not change anything. For notice: I have an ubiquiti mesh-ac with exactly the same configuration, there is no problem.
On the DAP-2610 the problem appears with removing of swconfig in 23.05

[brada4]

What did you get from sysfs and did brctl change the value? DSA offloads are expected to follow it to the point.
brctl showmacs br-lan - "non-local" should have rolling expire timers - once they expire next packet to MAC should reach all ports, I think default was 15s for ages on most unmanageable switches and soft switches.
(i am just trying to figure out if kernel on CPU or DSA switch config is to blame)
Are you in position to flip wired client between switches under 15s?

[StifflersMom]

root@DAP-UNTEN:/proc# cat /sys/class/net/br-lan/bridge/ageing_time
500

root@DAP-UNTEN:/proc# brctl showmacs br-lan
port no mac addr                is local?       ageing timer
  1     00:25:22:a6:ee:d8       no                 0.09
  1     04:d9:f5:cf:2c:ea       no                 0.54
  2     8c:7a:3d:c4:cb:0a       no                 4.41
  1     ec:ad:e0:79:98:40       yes                0.00
  1     ec:ad:e0:79:98:40       yes                0.00
  2     ec:ad:e0:79:98:48       yes                0.00
  2     ec:ad:e0:79:98:48       yes                0.00
  4     ee:ad:e0:79:98:40       yes                0.00
  4     ee:ad:e0:79:98:40       yes                0.00

If I do a tcpdump on the wifi:

tcpdump -n -i phy0-ap0 'ether host 8c:7a:3d:c4:cb:0a'
18:35:42.835968 8c:7a:3d:c4:cb:0a > ff:ff:ff:ff:ff:ff Null Unnumbered, xid, Flags [Response], length 6: 01 00
18:35:42.997264 IP6 :: > ff02::16: HBH ICMP6, multicast listener report v2, 4 group record(s), length 88
18:35:43.042253 IP 0.0.0.0.68 > 255.255.255.255.67: BOOTP/DHCP, Request from 8c:7a:3d:c4:cb:0a, length 310

8c:7a:3d:c4:cb:0a is the mac address of my mobile device.

[xback]

Could you test again by building the master branch?
It contains this commit: link

If this works, I can backport it to 23.05 in my staging tree as a 2nd test then

[robimarko]

Assisted learning has been enabled on ipq40xx from the start because of roaming issues like these.

That patch wont help as ipq40xx has a modified qca8k copy in 23.05(In master we are reusing generic qca8k finally).

Kind of only idea I have is that maybe we need to disable learning on all ports by default and then let DSA enable it via bridge port attributes(this is the preffered way)

[StifflersMom]

I rebuilt my build environment (sounds funny) and after serval hours of warming up my home with compiling, I managed to build the firmware from the master branch. I did not patch anything and:
It works without any problems with the master branch.

[raenye]

This is the same as #11650, right? not particular to ipq40xx.

[siwind]

I build the openwrt-master branch using in my ipq40x device, the problem is still there.

I build the recent master branch(openwrt kernel is 6.1x ), the problem is there.
but build the firmwear with the previous version(kernel is 5.15.x), it works well without any problem.

[StifflersMom]

I build the openwrt-master branch using in my ipq40x device, the problem is still there.

I have installed an actual snapshot on my DAP-2610 and the problem is not present. Actually, it is only at the 23.0.5 branch.
What is your device model?

[siwind]

My router runs in my home with latest build openwrt main branch, and an AP connect to it's lan port without dns service.
My mobile devices(such as phone) cann't get IP address from AP device, but can get IP address from my router.

The router is:

root@OpenWrt:~# opkg print-architecture
arch all 1
arch noarch 1
arch arm_cortex-a7_neon-vfpv4 10
root@OpenWrt:~# opkg list-installed | grep ath
ath10k-board-qca4019 - 20230804-1
ath10k-firmware-qca4019-ct - 2020-11-08-1
kmod-ath - 6.1.62+6.5-1
kmod-ath10k-ct - 6.1.62+2023-06-05-fadd0768-1
ucode-mod-math - 2023-11-07-a6e75e02-1
root@OpenWrt:~# uname -a
Linux OpenWrt 6.1.62 #0 SMP Fri Nov 17 12:12:09 2023 armv7l GNU/Linux
root@OpenWrt:~# 

The AP device reports these error message:

[57953.642035] mt7530-mdio mdio-bus:1f: port 1 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[57953.659874] mt7530-mdio mdio-bus:1f: port 1 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[57953.677653] mt7530-mdio mdio-bus:1f: port 0 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[57953.695417] mt7530-mdio mdio-bus:1f: port 0 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[57966.400706] br-lan: received packet on lan3 with own address as source address (addr:e4:ea:83:af:98:8f, vlan:0)
[57978.908513] mt7530-mdio mdio-bus:1f: port 2 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[57978.926500] mt7530-mdio mdio-bus:1f: port 0 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[57978.944387] mt7530-mdio mdio-bus:1f: port 1 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[57978.962189] mt7530-mdio mdio-bus:1f: port 0 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[58023.594927] mt7530-mdio mdio-bus:1f: port 0 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[58029.249282] mt7530-mdio mdio-bus:1f: port 1 failed to delete 60:67:20:8f:bc:e4 vid 0 from fdb: -2
[58541.432047] br-lan: received packet on lan3 with own address as source address (addr:e4:ea:83:af:98:8f, vlan:0)
[58899.618941] br-lan: received packet on lan3 with own address as source address (addr:e4:ea:83:af:98:8f, vlan:0)
[59256.301008] br-lan: received packet on lan3 with own address as source address (addr:e4:ea:83:af:98:8f, vlan:0)
[59381.726279] br-lan: received packet on lan3 with own address as source address (addr:e4:ea:83:af:98:8f, vlan:0)

=========
But if my router(IPQ40xx) runs previous openwrt version (the kenel is 5.xx), then any thing goes well.

Please see here : #11650