netdev - Re: [RFC net-next] net: don't dump stack on queue timeout

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: Fri, 10 Nov 2023 09:01:29 +0100
From: Daniele Palmas <dnlplm@...il.com>
To: Jakub Kicinski <kuba@...nel.org>
Cc: davem@...emloft.net, netdev@...r.kernel.org, edumazet@...gle.com, 
	pabeni@...hat.com, syzbot+d55372214aff0faa1f1f@...kaller.appspotmail.com, 
	jhs@...atatu.com, xiyou.wangcong@...il.com, jiri@...nulli.us
Subject: Re: [RFC net-next] net: don't dump stack on queue timeout

Il giorno gio 9 nov 2023 alle ore 16:18 Jakub Kicinski
<kuba@...nel.org> ha scritto:
>
> On Thu, 9 Nov 2023 08:40:00 +0100 Daniele Palmas wrote:
> > For example, I can see the splat with MBIM modems when radio link
> > failure happens, something for which the host can't really do
> > anything. So, the main result of using WARN is to scare the users who
> > are not aware of the reasons behind it and create unneeded support
> > requests...
>
> Is it not possible to clear the carrier on downstream devices?
> Radio link failure sounds like carrier loss.

The problem is that the MBIM standard does not define the
CDC_NOTIFY_NETWORK_CONNECTION, so carrier loss detection is managed
through the indications on the control channel.

But the kernel is not aware of what's passing through the control
channel, so it's the userspace tool that should detect carrier loss,
disconnect the bearers and set the network interface down.

For example, ModemManager is capable of doing that, but the problem is
that usually the standard modem notifications on the control channel
arrive later than the splat: increasing watchdog_timeo does not seem
to me a good option, since the notification could arrive much later.

One possible solution is to have some proprietary notifications on the
control channel that detect RLF early and trigger the above described
process before the warn happens: by coincidence, I wrote a custom
ModemManager patch for this a few days ago
https://gitlab.freedesktop.org/dnlplm/ModemManager/-/commit/89ba8ab65d4bfbd4cf1ff11ed58c08b112aca80f

Regards,
Daniele