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] [day] [month] [year] [list]

Date:   Thu, 5 Aug 2021 23:12:46 +0200
From:   Daniele Palmas <dnlplm@...il.com>
To:     Aleksander Morgado <aleksander@...ksander.es>
Cc:     Subash Abhinov Kasiviswanathan <subashab@...eaurora.org>,
        Bjørn Mork <bjorn@...k.no>,
        Network Development <netdev@...r.kernel.org>,
        Sean Tranchetti <stranche@...eaurora.org>
Subject: Re: RMNET QMAP data aggregation with size greater than 16384

Il giorno gio 5 ago 2021 alle ore 23:01 Aleksander Morgado
<aleksander@...ksander.es> ha scritto:
>
> >> > > I'm playing with the whole QMAP data aggregation setup with a USB
> >> > > connected Fibocom FM150-AE module (SDX55).
> >> > > See https://gitlab.freedesktop.org/mobile-broadband/libqmi/-/issues/71
> >> > > for some details on how I tested all this.
> >> > >
> >> > > This module reports a "Downlink Data Aggregation Max Size" of 32768
> >> > > via the "QMI WDA Get Data Format" request/response, and therefore I
> >> > > configured the MTU of the master wwan0 interface with that same value
> >> > > (while in 802.3 mode, before switching to raw-ip and enabling
> >> > > qmap-pass-through in qmi_wwan).
> >> > >
> >> > > When attempting to create a new link using netlink, the operation
> >> > > fails with -EINVAL, and following the code path in the kernel driver,
> >> > > it looks like there is a check in rmnet_vnd_change_mtu() where the
> >> > > master interface MTU is checked against the RMNET_MAX_PACKET_SIZE
> >> > > value, defined as 16384.
> >> > >
> >> > > If I setup the master interface with MTU 16384 before creating the
> >> > > links with netlink, there's no error reported anywhere. The FM150
> >> > > module crashes as soon as I connect it with data aggregation enabled,
> >> > > but that's a different story...
> >> > >
> >> > > Is this limitation imposed by the RMNET_MAX_PACKET_SIZE value still a
> >> > > valid one in this case? Should changing the max packet size to 32768
> >> > > be a reasonable approach? Am I doing something wrong? :)
> >> > >
> >> > > This previous discussion for the qmi_wwan add_mux/del_mux case is
> >> > > relevant:
> >> > > https://patchwork.ozlabs.org/project/netdev/patch/20200909091302.20992-1-dnlplm@gmail.com/..
> >> > > The suggested patch was not included yet in the qmi_wwan driver and
> >> > > therefore the user still needs to manually configure the MTU of the
> >> > > master interface before setting up all the links, but at least there
> >> > > seems to be no maximum hardcoded limit.
> >> > >
> >> > > Cheers!
> >> >
> >> > Hi Aleksander
> >> >
> >> > The downlink data aggregation size shouldn't affect the MTU.
> >> > MTU applies for uplink only and there is no correlation with the
> >> > downlink path.
> >> > Ideally, you should be able to use standard 1500 bytes (+ additional
> >> > size for MAP header)
> >> > for the master device. Is there some specific network which is using
> >> > greater than 1500 for the IP packet itself in uplink.
> >> >
> >>
> >> I may be mistaken then in how this should be setup when using rmnet.
> >> For the qmi_wwan case using add_mux/del_mux (Daniele correct me if
> >> wrong!), we do need to configure the MTU of the master interface to be
> >> equal to the aggregation data size reported via QMI WDA before
> >> creating any mux link; see
> >> http://paldan.altervista.org/linux-qmap-qmi_wwan-multiple-pdn-setup/
> >>
> >
> > Right: it's not for the MTU itself, but for changing the rx_urb_size, since usbnet_change_mtu has that side effect.
> >
>
> I knew there was a reason even if not obvious. Should we fix that rx
> urb size value to 16384 to avoid needing that extra step? Was that
> what you were suggesting in that patch that was never merged?
>

Yes, that was my purpose, but thinking about it twice, I thought it
was not a very good idea.

There are too many modems/hosts/firmware versions combinations, each
one of them with its own bunch of bugs/quirks (daily experienced):
after all I think it's better to have the possibility to have the urb
size configurable (even if through an odd way), than to have it fixed.

Regards,
Daniele

>
> --
> Aleksander
> https://aleksander.es