linux-kernel - Re: [PATCH net-next] net: veth: Disable netpoll support

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]

Message-Id: <1764919679.7wwurfe3mz.astroid@yuna.none>
Date: Fri, 05 Dec 2025 08:35:12 +0100
From: Fabian Grünbichler <f.gruenbichler@...xmox.com>
To: Jakub Kicinski <kuba@...nel.org>
Cc: "David S. Miller" <davem@...emloft.net>, Eric Dumazet
	<edumazet@...gle.com>, Breno Leitao <leitao@...ian.org>, leit@...a.com,
	open list <linux-kernel@...r.kernel.org>, "open list:NETWORKING DRIVERS"
	<netdev@...r.kernel.org>, Paolo Abeni <pabeni@...hat.com>
Subject: Re: [PATCH net-next] net: veth: Disable netpoll support

On December 5, 2025 2:34 am, Jakub Kicinski wrote:
> On Thu, 04 Dec 2025 10:20:06 +0100 Fabian Grünbichler wrote:
>> On August 5, 2024 11:40 am, Breno Leitao wrote:
>> > The current implementation of netpoll in veth devices leads to
>> > suboptimal behavior, as it triggers warnings due to the invocation of
>> > __netif_rx() within a softirq context. This is not compliant with
>> > expected practices, as __netif_rx() has the following statement:
>> > 
>> > 	lockdep_assert_once(hardirq_count() | softirq_count());
>> > 
>> > Given that veth devices typically do not benefit from the
>> > functionalities provided by netpoll, Disable netpoll for veth
>> > interfaces.  
>> 
>> this patch seems to have broken combining netconsole and bridges with
>> veth ports:
>> 
>> https://bugzilla.proxmox.com/show_bug.cgi?id=6873
>> 
>> any chance this is solvable?
> 
> What's the reason to set up netcons over veth?

I don't think there is a particular reason to do so, the veth devices
just get "caught in the crossfire", so to speak - if the netconsole
setup includes the bridge that the veth device is plugged into.

> Note that unlike normal IP traffic netcons just blindly pipes out fully
> baked skbs, it doesn't use the IP stack. So unlike normal IP traffic
> I think you can still point it at the physical netdev, even if that
> physical netdev is under a bridge.

yes, pointing it at a physical bridge port works!

I mainly wanted to make you aware of this regression, since it seems it
was not on the radar when the original patch was written and applied. if
fixing it is too much of a hassle/has too many other unwanted side
effects, I do think (hope? ;)) people can live with this restriction. I
definitely agree that restricting netconsole to the actual links where
the traffic is supposed to go out is the sensible choice in any case.