Re: [PATCH] sock: Make sock->sk_tstamp thread-safe

From: Deepa Dinamani <deepa.kernel@gmail.com>
To: Arnd Bergmann <arnd@arndb.de>
Cc: "David S. Miller" <davem@davemloft.net>,
	Linux Network Devel Mailing List <netdev@vger.kernel.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	y2038 Mailman List <y2038@lists.linaro.org>
Subject: Re: [PATCH] sock: Make sock->sk_tstamp thread-safe
Date: Sat, 22 Dec 2018 08:34:55 -0800	[thread overview]
Message-ID: <CABeXuvrdz2UwGBj1+gNhnwcy-uG_dKezM_fPqU99rKuQFsyHJw@mail.gmail.com> (raw)
In-Reply-To: <CAK8P3a3LivERgLTTmS0A35CoebS3n+cNWNXkoYKGam+fVR-Dxg@mail.gmail.com>

> Are we actually worried about concurrent writers here? I thought the
> only problem was a race between writer and reader, which would mean
> that we could solve it using only a seqcount_t which is cheaper to
> update than a seqlock_t.

I considered using just the seqcount_t. But, I think we do care about
concurrent writers here.
A couple of scenarios I can think of:

1. When you have 2 concurrent recvmsg() calls on a socket, and they
both try to update sk_tstamp.
2. If a socket has don't have one of the SO_TIMESTAMP/NS options set
and you have a first recvmsg and a concurrent ioctl call on the
socket.

These are corner cases and if we don't care aout these then we can use
just the sequence counters.

I have missed out tstamp update in the sunrcpc code. If everyone is ok
with this approach, I will add it in when I post an update

-Deepa