netdev - Re: Hardcoded multicast queue length in macvlan.c driver causes poor multicast receive performance

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

Message-ID: <7f18cc47-f1a1-4fa6-0f31-63deb12d9d1b@paneda.se>
Date:   Sat, 28 Nov 2020 00:13:25 +0100
From:   Thomas Karlsson <thomas.karlsson@...eda.se>
To:     Jakub Kicinski <kuba@...nel.org>
CC:     "davem@...emloft.net" <davem@...emloft.net>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        <jiri@...nulli.us>, <kaber@...sh.net>, <edumazet@...gle.com>,
        <vyasevic@...hat.com>, <alexander.duyck@...il.com>
Subject: Re: Hardcoded multicast queue length in macvlan.c driver causes poor
 multicast receive performance

On 2020-11-27 18:27, Jakub Kicinski wrote:
> On Thu, 26 Nov 2020 21:00:46 +0100 Thomas Karlsson wrote:
>> On 2020-11-26 00:01, Jakub Kicinski wrote:
>>> On Wed, 25 Nov 2020 23:15:39 +0100 Thomas Karlsson wrote:  
>>>> Or is there a way to set the parameters in a more "raw" form that
>>>> does not require a patch to iproute2 with parameter parsing, error
>>>> handing, man pages updates, etc. I feel that I'm getting in over my
>>>> head here.  
>>>
>>> We're here to assist! Netlink takes a little bit of effort 
>>> to comprehend but it's very simple once you get the mechanics!
>>>   
>>
>> Thanks for the encouragement, I have been able to build iproute2 today and
>> I am successfully communicating with the driver now being able to set and retrieve my queue len!
>>
>> As I'm working on this I do got a question. I placed the bc_queue_len into the struct macvlan_port *port
>> since that is where the bc_queue is located today. But when I change and retrieve the queue from userspace I realize
>> that all macvlan interfaces that share the same physical lowerdev uses the same port structure and thus
>> the same bc_queue_len.
> 
> Indeed looks like its an ingress attribute.
> 
>> It confused me at first and I'm not sure if that is how it should be. I expected the driver to have different
>> bc_queues for all macvlan interfaces no matter which lowerdev they were using but obviously that is not the case.
>>
>> It may be a bit confusing to change bc_queue_len on one macvlan and see that the change was applied to more than one.
>>
>> But I'm not sure if I should just move bc_queue_len to the struct macvlan_dev either. because then different macvlans will use different queue lengths while they still use the same queue. Which may also be considered a bit illogical
>>
>> Let me know what you prefer here!
> 
> I'd record the queue len requested by each interface in their struct
> macvlan_dev and then calculate a max over the members to set the actual
> value in struct macvlan_port.

That sounds like a good approach to me. I can see that for example mtu is
handled in a similar aggregated way. I'll give it a try.
Thanks!

> 
> Let me CC some extra people, looks like macvlan does not have a
> maintainer..
>