Re: [PATCH mptcp 02/10] mptcp: tag sequence_seq with socket state

From: Florian Westphal <fw@strlen.de>
To: Mat Martineau <mathew.j.martineau@linux.intel.com>
Cc: Florian Westphal <fw@strlen.de>, Paolo Abeni <pabeni@redhat.com>,
	mptcp@lists.linux.dev
Subject: Re: [PATCH mptcp 02/10] mptcp: tag sequence_seq with socket state
Date: Mon, 12 Apr 2021 10:05:31 +0200	[thread overview]
Message-ID: <20210412080531.GA14932@breakpoint.cc> (raw)
In-Reply-To: <9f60e0e9-a44f-b659-14f6-f91161cf6569@linux.intel.com>

Mat Martineau <mathew.j.martineau@linux.intel.com> wrote:
> On Thu, 8 Apr 2021, Florian Westphal wrote:
> 
> > +	return sk->sk_state << 24u;
> 
> I think this works because sk->sk_state gets automatically promoted to an
> int, but maybe "(u32)sk->sk_state << 24" is the intent?

Yes, I can add an explicit cast.

> What I didn't realize at that time was that mptcp_sockopt_sync_all() might
> not be called for a very long time (or at all?) if there are no joins to
> trigger it. To really protect against overflow, mptcp_setsockopt() can call
> (or schedule) mptcp_sockopt_sync_all() if the non-sk_state bits of
> msk->setsockopt_seq exceed a threshold.

Next patch adds sockopt_seq_inc() which is supposed to guarantee that it
never wraps to 0.  I can change it so it will wrap in the 24bit sequence
space only, so it would cycle back to sk->sk_state << 24u.