linux-kernel - Re: B53 DSA switch problem on Banana Pi-R1 on Fedora 26

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <8ffb6309-3d08-3043-43d9-49aa3a686bf6@wiesinger.com>
Date:   Mon, 28 May 2018 07:32:10 +0200
From:   Gerhard Wiesinger <lists@...singer.com>
To:     Florian Fainelli <f.fainelli@...il.com>,
        Andrew Lunn <andrew@...n.ch>, initramfs@...r.kernel.org
Cc:     linux-kernel@...r.kernel.org
Subject: Re: B53 DSA switch problem on Banana Pi-R1 on Fedora 26 -
 systemd-networkd problem

On 27.05.2018 22:31, Florian Fainelli wrote:
> Le 05/27/18 à 12:01, Gerhard Wiesinger a écrit :
>> On 24.05.2018 08:22, Gerhard Wiesinger wrote:
>>> On 24.05.2018 07:29, Gerhard Wiesinger wrote:
>>>> After some analysis with Florian (thnx) we found out that the current
>>>> implementation is broken:
>>>>
>>>> https://patchwork.ozlabs.org/patch/836538/
>>>> https://github.com/torvalds/linux/commit/c499696e7901bda18385ac723b7bd27c3a4af624#diff-a2b6f8d89e18de600e873ac3ac43fa1d
>>>>
>>>>
>>>> Florians comment:
>>>>
>>>> c499696e7901bda18385ac723b7bd27c3a4af624 ("net: dsa: b53: Stop using
>>>> dev->cpu_port incorrectly") since it would result in no longer setting
>>>> the CPU port as tagged for a specific VLAN. Easiest way for you right
>>>> now is to just revert it, but this needs some more thoughts for a proper
>>>> upstream change. I will think about it some more.
>>> Can confirm 4.14.18-200.fc26.armv7hl works, 4.15.x should be broken.
>>>
>>> # Kernel 4.14.x ok
>>> https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/log/drivers/net/dsa/b53?h=v4.14.43
>>>
>>> # Kernel 4.15.x should be NOT ok
>>> https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/log/drivers/net/dsa/b53?h=v4.15.18
>>>
>> Kernel 4.14.18-300.fc27.armv7hl works well so far, even with FC28
>> update. Florian send me a patch to try for 4.16.x
> So does my patch make 4.16 work correctly for you now? If so, can I just
> submit it and copy you?
>
>> I got the  commands below to work with manual script commands.
>> Afterwards I wrote systemd-networkd config where I've a strage problem
>> when IPv6 sends a multicast broadcast from another machine to the bridge
>> this will be sent back via the network interface, but with the source
>> MAC of the bridge of the other machine. dmesg from the other machine:
>> [117768.330444] br0: received packet on lan0 with own address as source
>> address (addr:a0:36:9f:ab:cd:ef, vlan:0)
>> [117768.334887] br0: received packet on lan0 with own address as source
>> address (addr:a0:36:9f:ab:cd:ef, vlan:0)
>> [117768.339281] br0: received packet on lan0 with own address as source
>> address (addr:a0:36:9f:ab:cd:ef, vlan:0)
>>
>> And: If I just enter this command after e.g. a systemd-network restart
>> everything is fine forever:
>> # Not OK (dmesg message above is triggered on a remote computer, whole
>> switching network gets unstable, ssh terminals close, packet loss, etc.)
>> systemctl restart systemd-networkd
>> # OK again when this command is entered
>> bridge vlan add dev wan vid 102 pvid untagged
>>
>> brctl show, ip link, bridge vlan, bridge link commands, etc. look all
>> the same, also /sys/class/net/br0/bridge, /sys/class/net/br1/bridge
>> settings
>>
>> Systemd config correct?
>> Any ideas?
> You should not have eth0.101 and eth0.102 to be enslaved in a bridge at
> all, this is what is causing the bridge to be confused. Remember what I
> wrote to you before, with the current b53 driver that does not have any
> tagging enabled the lanX interfaces and brX interfaces are only used for
> control and should not be used for passing any data. The only network
> device that will be passing data is eth0, which is why we need to set-up
> VLAN interfaces to pop/push the VLAN id accordingly.
>
> I have no idea why manual vs. systemd does not work but you can most
> certainly troubleshoot that by comparing the bridge/ip outputs.

So is that then the correct structure?

br1
- lan1 (with VID 101)
- lan2 (with VID 101)
- lan3 (with VID 101)
- lan4 (with VID 101)

brlan
- eth0.101
- wlan0 (currently not active, could be optimized without bridge but for 
future comfort)

br2
- wan (with VID 102) (could be optimized without bridge but for future 
comfort)
- future1

brwan
- eth0.102 (could be optimized without bridge but for future comfort)
- future2

Ad systemd vs. manual config: As I said I didn't find any difference in 
the bridge/ip outputs. As they are broken (see other message) maybe 
something else is broken, too.

Thnx.

Ciao,
Gerhard