netdev - Re: RMNET QMAP data aggregation with size greater than 16384

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: <87bl6aqrat.fsf@miraculix.mork.no>
Date:   Fri, 06 Aug 2021 21:58:02 +0200
From:   Bjørn Mork <bjorn@...k.no>
To:     subashab@...eaurora.org
Cc:     Aleksander Morgado <aleksander@...ksander.es>,
        Daniele Palmas <dnlplm@...il.com>,
        Network Development <netdev@...r.kernel.org>,
        Sean Tranchetti <stranche@...eaurora.org>
Subject: Re: RMNET QMAP data aggregation with size greater than 16384

subashab@...eaurora.org writes:

> Unfortunately, this seems to be a limitation of qmi_wwan (usbnet)
> where its tying the RX to the TX size through usbnet_change_mtu.
>
> Ideally, we should break this dependency and have a sysfs or some other
> configuration scheme to set the rx_urb_size.
>
> Looks like this discussion has happened a while back and the option to
> use
> a configurable scheme for rx_urb_size was rejected by Bjorn and Greg KH.

Ouch, I never meant to shoot down the proposal.  I had to go back and
read my comments. I see that it might have come out that way...

My main point was that I'd prefer this to work without any userspace
invervention, if possible.  And I'm still not sure we've explored that
alternative to the end?

> The summary of the thread was to set a large rx_urb_size during probe
> itself for qmi_wwan.

Yes, I think it would be good to make the driver DTRT automatically.
Coding driver specific quirks into ModemManager might work, but it feels
wrong to work around a Linux driver bug. We don't have to do that.  We
can fix the driver.

> https://patchwork.kernel.org/project/linux-usb/patch/20200803065105.8997-1-yzc666@netease.com/
>
> We could try setting a large value as suggested there and it should
> hopefully
> solve the issue you are seeing.

Why can't we break the rx_urb_size dependency on MTU automatically when
pass_through or qmi_wwan internal muxing is enabled? Preferably with
some fixed default size which would Just Work for everyone.

I'm not rejecting a rx_urb_size knob if it is absolutely necessary. But
I hope someone will write a good explanation on how to tune that value
then. Knobs are only useful if the user can make an intelligent choice.
This should go into the ABI docs.

Bjørn