Re: [PATCH] AF_UNIX: Implement SO_TIMESTAMP and SO_TIMETAMPNS on Unix sockets

[Alban Crequy <alban.crequy@collabora.co.uk>]

Le Mon, 04 Oct 2010 18:41:44 +0200,
Eric Dumazet <eric.dumazet@gmail.com> a écrit :

> Le lundi 04 octobre 2010 à 16:38 +0100, Alban Crequy a écrit :
> > Userspace applications can already request to receive timestamps
> > with: setsockopt(sockfd, SOL_SOCKET, SO_TIMESTAMP, ...)
> > 
> > Although setsockopt() returns zero (success), timestamps are not
> > added to the ancillary data. This patch fixes that.
> > 
> > Signed-off-by: Alban Crequy <alban.crequy@collabora.co.uk>
> > ---
> >  net/unix/af_unix.c |    4 ++++
> >  1 files changed, 4 insertions(+), 0 deletions(-)
> > 
> > diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
> > index 617bea4..142ccea 100644
> > --- a/net/unix/af_unix.c
> > +++ b/net/unix/af_unix.c
> > @@ -1697,6 +1697,8 @@ static int unix_dgram_recvmsg(struct kiocb
> > *iocb, struct socket *sock, if (err)
> >  		goto out_free;
> >  
> > +	sock_recv_timestamp(msg, sk, skb);
> > +
> >  	if (!siocb->scm) {
> >  		siocb->scm = &tmp_scm;
> >  		memset(&tmp_scm, 0, sizeof(tmp_scm));
> > @@ -1877,6 +1879,8 @@ static int unix_stream_recvmsg(struct kiocb
> > *iocb, struct socket *sock, copied += chunk;
> >  		size -= chunk;
> >  
> > +		sock_recv_timestamp(msg, sk, skb);
> > +
> >  		/* Mark read part of skb as used */
> >  		if (!(flags & MSG_PEEK)) {
> >  			skb_pull(skb, chunk);
> 
> Are you sure its needed for unix_stream case ?

I was about to say it does not cost anything to add it...

> We dont do this for TCP for example, only for datagrams.
> 
> As shown in the past, sock_recv_timestamp() is a bit expensive because
> it takes care of many possible options.

Is it really expensive? It looks like a few flag checks in an inline
function to me.

> It would be better to use in AF_UNIX case (only software timestamps) :
> 
> Solution 1) 
> if (sock_flag(sk, SOCK_RCVTSTAMP))
> 	__sock_recv_timestamp(msg, sk, skb);
> 
> Solution 2) 
> Or something already used elsewhere since 2.6.35 and commit
> 767dd03369ac1 (net: speedup sock_recv_ts_and_drops()) :
> 
> sock_recv_ts_and_drops(msg, sk, skb);
> 
> 
> I would vote for the 1) solution

The patch in the next email implements the 1) solution on SOCK_DGRAM and
SOCK_SEQPACKET (without SOCK_STREAM).

-- 
Alban