netdev - Re: [PATCH 5/5] mac80211: set NETIF_F_LLTX when using intermediate tx queues

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]

Date:   Tue, 16 Apr 2019 11:02:50 +0100
From:   Toke Høiland-Jørgensen <toke@...hat.com>
To:     Johannes Berg <johannes@...solutions.net>,
        Herbert Xu <herbert@...dor.apana.org.au>
Cc:     Arend Van Spriel <arend.vanspriel@...adcom.com>,
        Felix Fietkau <nbd@....name>, linux-wireless@...r.kernel.org,
        Eric Dumazet <eric.dumazet@...il.com>, netdev@...r.kernel.org
Subject: Re: [PATCH 5/5] mac80211: set NETIF_F_LLTX when using intermediate tx queues

Johannes Berg <johannes@...solutions.net> writes:

> On Tue, 2019-04-16 at 17:37 +0800, Herbert Xu wrote:
>> On Tue, Apr 16, 2019 at 11:33:50AM +0200, Johannes Berg wrote:
>> > On Tue, 2019-04-16 at 10:33 +0100, Toke Høiland-Jørgensen wrote:
>> > > 
>> > > > It is true because we have an entire buffering layer in mac80211 (in
>> > > > this case at least) and never push back to the stack.
>> > > 
>> > > I'm wondering if we should be?
>> > 
>> > I don't think so? We'd just buffer packets in yet another place.
>> 
>> But you do realise that you're giving up on the rich queueing
>> functionality that Linux provides (net/sched),
>
> Yes, that was a trade-off we always knew about. The model that Linux
> provides is just not suited for wifi.

As explained at great length here:
https://www.usenix.org/conference/atc17/technical-sessions/presentation/hoilan-jorgesen
(you already know that of course, Johannes)

>> not to mention
>> breaking certain applications that rely on congestion feedback?
>
> This I don't understand. The congestion feedback happens through socket
> buffer space etc. which is still there (as long as nobody sneaks in an
> skb_orphan() call)

Sure, for TCP, the TSQ mechanism should keep the upper-level queue low
as long as the SKBs are alive. But is this also the case for UDP?

-Toke