[Cake] (no subject)

[Cake] (no subject)

[Mark Captur]

[-- Attachment #1: Type: text/plain, Size: 327 bytes --]

I am using cake on latest lede nightly. I'm using diffserv 4 which creates
4 tins bulk, best effort, video and voice.

Is there a way to change dscp markings on in comming traffic to place it in
te video tin. More specifically i would like all incoming traffic with
source port 23000 to be set to dscp class CS4.

Thanks,
Mark

[-- Attachment #2: Type: text/html, Size: 411 bytes --]

[Cake] dscp marking ingress traffic (was (no subject))

[John Sager]

I've wondered that, and at the moment I don't think there is. The ingress
qdisc stuff happens before the packets hit any iptables hooks so you can't
use iptables rules to do it. None of the actions that can be attached to a
filter on the ingress qdisc seem to offer what you want.

I use packet marks and connmarks to classify outgoing traffic using iptables
rules. The connmarks are then transferred back to incoming packets using the
connmark action, followed by the mirred action to send them to ifb0 where
the real downstream qdiscs - htb and fq_codel - are applied. This works well
to partition my upstream & downstream bandwidth between different traffic
classes, but I can't yet see how to adapt this to work with cake.

John


On 07/12/17 13:34, Mark Captur wrote:
> I am using cake on latest lede nightly. I'm using diffserv 4 which creates 4
> tins bulk, best effort, video and voice.
> 
> Is there a way to change dscp markings on in comming traffic to place it in
> te video tin. More specifically i would like all incoming traffic with
> source port 23000 to be set to dscp class CS4.
> 
> Thanks,
> Mark
> 

Re: [Cake] dscp marking ingress traffic (was (no subject))

[Georgios Amanakis]

You can do it, but it has to be on a different interface than the
incoming traffic.
Say incoming traffic is on eth0 (WAN) and outgoing LAN traffic is on
eth1. Then you can DSCP mark with iptables on incoming eth0, and since
the traffic is forwarded to eth1 the DSCP marking is going to be
preserved. 

On Thu, 2017-12-07 at 22:48 +0000, John Sager wrote:
> I've wondered that, and at the moment I don't think there is. The
> ingress
> qdisc stuff happens before the packets hit any iptables hooks so you
> can't
> use iptables rules to do it. None of the actions that can be attached
> to a
> filter on the ingress qdisc seem to offer what you want.
> 
> I use packet marks and connmarks to classify outgoing traffic using
> iptables
> rules. The connmarks are then transferred back to incoming packets
> using the
> connmark action, followed by the mirred action to send them to ifb0
> where
> the real downstream qdiscs - htb and fq_codel - are applied. This
> works well
> to partition my upstream & downstream bandwidth between different
> traffic
> classes, but I can't yet see how to adapt this to work with cake.
> 
> John
> 
> 
> On 07/12/17 13:34, Mark Captur wrote:
> > I am using cake on latest lede nightly. I'm using diffserv 4 which
> > creates 4
> > tins bulk, best effort, video and voice.
> > 
> > Is there a way to change dscp markings on in comming traffic to
> > place it in
> > te video tin. More specifically i would like all incoming traffic
> > with
> > source port 23000 to be set to dscp class CS4.
> > 
> > Thanks,
> > Mark
> > 
> 
> _______________________________________________
> Cake mailing list
> Cake@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cake

Re: [Cake] dscp marking ingress traffic (was (no subject))

[John Sager]

That would work for a single LAN interface but I run several internal vlans
off my router and I need to manage the ingress traffic at a single point. I
haven't yet found a way of doing that other than with the ifb stuff. I could
run two virtual routers in tandem but I don't think the distro I'm using
(Leaf-Bering) supports that easily.

John

On 07/12/17 23:05, Georgios Amanakis wrote:
> You can do it, but it has to be on a different interface than the
> incoming traffic.
> Say incoming traffic is on eth0 (WAN) and outgoing LAN traffic is on
> eth1. Then you can DSCP mark with iptables on incoming eth0, and since
> the traffic is forwarded to eth1 the DSCP marking is going to be
> preserved. 
> 
> On Thu, 2017-12-07 at 22:48 +0000, John Sager wrote:
>> I've wondered that, and at the moment I don't think there is. The
>> ingress
>> qdisc stuff happens before the packets hit any iptables hooks so you
>> can't
>> use iptables rules to do it. None of the actions that can be attached
>> to a
>> filter on the ingress qdisc seem to offer what you want.
>>
>> I use packet marks and connmarks to classify outgoing traffic using
>> iptables
>> rules. The connmarks are then transferred back to incoming packets
>> using the
>> connmark action, followed by the mirred action to send them to ifb0
>> where
>> the real downstream qdiscs - htb and fq_codel - are applied. This
>> works well
>> to partition my upstream & downstream bandwidth between different
>> traffic
>> classes, but I can't yet see how to adapt this to work with cake.
>>
>> John
>>
>>
>> On 07/12/17 13:34, Mark Captur wrote:
>>> I am using cake on latest lede nightly. I'm using diffserv 4 which
>>> creates 4
>>> tins bulk, best effort, video and voice.
>>>
>>> Is there a way to change dscp markings on in comming traffic to
>>> place it in
>>> te video tin. More specifically i would like all incoming traffic
>>> with
>>> source port 23000 to be set to dscp class CS4.
>>>
>>> Thanks,
>>> Mark
>>>
>>
>> _______________________________________________
>> Cake mailing list
>> Cake@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/cake

Re: [Cake] (no subject)

[Andy Furniss]

Mark Captur wrote:
> I am using cake on latest lede nightly. I'm using diffserv 4 which creates
> 4 tins bulk, best effort, video and voice.
> 
> Is there a way to change dscp markings on in comming traffic to place it in
> te video tin. More specifically i would like all incoming traffic with
> source port 23000 to be set to dscp class CS4.

Maybe, I got it working using conntrack, but it was fragile in the sense
that kernel/iptables/iproute versions needed some sort of lucky match.

IIRC I took down my desktop PC messing around.

 From my router box it seems to work.

You need cake on ifb with traffic redirected from ingress interface.

https://lists.bufferbloat.net/pipermail/cake/2017-February/002491.html

Is just a test to mark ipv4 icmp which worked.
I did get conntrack to work, though you can just match the port anyway.

The exact commands needed would depend on your precise setup.
If your router is just routing then you could probably just have cake
on egress and save the hassle.