From: Christoph Hellwig <hch@xxxxxx>
Date: Mon, 26 May 2025 07:32:27 +0200
> On Fri, May 23, 2025 at 11:21:09AM -0700, Kuniyuki Iwashima wrote:
> > Let's restore sock_create_kern() that holds a netns reference.
> >
> > Now, it's the same as the version before commit 26abe14379f8 ("net:
> > Modify sk_alloc to not reference count the netns of kernel sockets.").
> >
> > Back then, after creating a socket in init_net, we used sk_change_net()
> > to drop the netns ref and switch to another netns, but now we can
> > simply use __sock_create_kern() instead.
> >
> > $ git blame -L:sk_change_net include/net/sock.h 26abe14379f8~
> >
> > DEBUG_NET_WARN_ON_ONCE() is to catch a path calling sock_create_kern()
> > from __net_init functions, since doing so would leak the netns as
> > __net_exit functions cannot run until the socket is removed.
>
> Is reusing the name as the old sock_create_kern a good idea? It can
> lead to bugs by people used to the old semantics.
In the old days, sock_create_kern() did take a ref to netns,
but an implicit change that avoids taking the ref has caused
a lot of problems for people who used to the old semantics.
This series rather rolls back the change, so I think using
the same name here is better than leaving the catchy
sock_create_kern() error-prone.
> It's also
> not really an all that descriptive name for either variant. I'm
> not really a net stack or namespace expert, but maybe we can come
> up with more descriptive version for both this new sock_create_kern
> and the old sock_create_kern/__sock_create_kern?
