Re: [PATCH 4.19 03/89] cgroup, netclassid: periodically release file_lock on classid updating

[Konstantin Khlebnikov]

On 18/03/2020 12.02, Pavel Machek wrote:
> Hi!
>
> > From: Dmitry Yakunin <zeil@xxxxxxxxxxxxxx>
> >
> > [ Upstream commit 018d26fcd12a75fb9b5fe233762aa3f2f0854b88 ]
> ...
> > Now update is non atomic and socket may be skipped using calls:
> >
> > dup2(oldfd, newfd);
> > close(oldfd);
> >
> > But this case is not typical. Moreover before this patch skip is possible
> > too by hiding socket fd in unix socket buffer.
>
> Dunno. This makes interface even more interesting.

This is part of brilliant cgroup-v1 design. =)
Cgroup-v2 doesn't recolor sockets when task is moved.

> > +
> >   static int update_classid_sock(const void *v, struct file *file, unsigned n)
> >   {
> >   	int err;
> > +	struct update_classid_context *ctx = (void *)v;
> >   	struct socket *sock = sock_from_file(file, &err);
> ...
> > +	if (--ctx->batch == 0) {
> > +		ctx->batch = UPDATE_CLASSID_BATCH;
> > +		return n + 1;
> > +	}
> >   	return 0;
> >   }
>
> We take "const void *" and then write to it. That's asking for
> trouble... right? Should the const annotation be removed, at least for
> sake of humans trying to understand the code?

Indeed, there is no much sense in opaque const void *.
This is how iterate_fd() is declared.

> Best regards,
> 									Pavel