netdev - Re: [PATCH net-next 0/5] net: Improve netns handling in RTNL and ip

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: <CABAhCOR5G+7TprR1NnFpjM61kCBpjAZ+SFhQDxKKr6EsB7Y5JQ@mail.gmail.com>
Date: Thu, 31 Oct 2024 11:08:37 +0800
From: Xiao Liang <shaw.leon@...il.com>
To: Jakub Kicinski <kuba@...nel.org>
Cc: netdev@...r.kernel.org, "David S. Miller" <davem@...emloft.net>, 
	David Ahern <dsahern@...nel.org>, Eric Dumazet <edumazet@...gle.com>, 
	Paolo Abeni <pabeni@...hat.com>, Kuniyuki Iwashima <kuniyu@...zon.com>, Ido Schimmel <idosch@...dia.com>
Subject: Re: [PATCH net-next 0/5] net: Improve netns handling in RTNL and ip_tunnel

On Thu, Oct 31, 2024 at 7:35 AM Jakub Kicinski <kuba@...nel.org> wrote:
>
> On Wed, 30 Oct 2024 10:10:32 +0800 Xiao Liang wrote:
> > > Do you think the netns_atomic module param is really necessary?
> > > I doubt anyone cares about the event popping up in the wrong
> > > name space first.
> >
> > We used FRRouting in our solution which listens to link notifications to
> > set up corresponding objects in userspace. Since the events are sent
> > in different namespaces (thus via different RTNL sockets), we can't
> > guarantee that the events are received in the correct order, and have
> > trouble processing them. The way to solve this problem I can think of is
> > to have a multi-netns RTNL socket where all events are synchronized,
> > or to eliminate the redundant events in the first place. The latter seems
> > easier to implement.
>
> I think we're on the same page. I'm saying that any potential user
> will run into a similar problem, and I don't see a clear use case
> for notifications in both namespaces. So we can try to make the
> behavior of netns_atomic=1 the default one and get rid of the module
> param.

Ah.. I misunderstood your question... Besides notifications there's another
behavior change.
In this patchset, RTNL core passes link-netns as src_net to drivers. But
ip tunnel driver doesn't support source netns currently, and that's why
we have Patch 4. As a side effect, source netns will be used if link-netns
is not specified.For example:

  ip -n ns1 link add netns ns2 gre1 type gre ...

In current implementation, the link-netns is ns2. While with netns_atomic=1,
link-netns will be ns1 (source netns). The module param serves as a way to
keep the current behavior.
I think we can make it default for drivers that already have source
netns support.