2014-04-15 12:35:36

by Kenjiro Nakayama

Subject: [PATCH v2] ipv4: Add option to get TCP_FASTOPEN to getsockopt()

Hi,

TCP_FASTOPEN option can be set via setsockopt(), but the value cannot be
gotten via getsockopt(). This patch adds the option to getsockopt().

Sighned-off-by: Kenjiro Nakayama <[email protected]>

Add option to get TCP_FASTOPEN to getsockopt()
---
net/ipv4/tcp.c | 7 +++++++
1 file changed, 7 insertions(+)

diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
index 4bd6d52..86186f4 100644
--- a/net/ipv4/tcp.c
+++ b/net/ipv4/tcp.c
@@ -2916,6 +2916,13 @@ static int do_tcp_getsockopt(struct sock *sk, int level,
case TCP_USER_TIMEOUT:
val = jiffies_to_msecs(icsk->icsk_user_timeout);
break;
+
+ case TCP_FASTOPEN:
+ val = icsk->icsk_accept_queue.fastopenq ?
+ icsk->icsk_accept_queue.fastopenq->max_qlen
+ : sysctl_tcp_fastopen;
+ break;
+
case TCP_TIMESTAMP:
val = tcp_time_stamp + tp->tsoffset;
break;
--
1.9.0

2014-04-15 13:52:39

by Eric Dumazet

Subject: Re: [PATCH v2] ipv4: Add option to get TCP_FASTOPEN to getsockopt()

On Tue, 2014-04-15 at 21:35 +0900, Kenjiro Nakayama wrote:
> Hi,
>
> TCP_FASTOPEN option can be set via setsockopt(), but the value cannot be
> gotten via getsockopt(). This patch adds the option to getsockopt().
>
> Sighned-off-by: Kenjiro Nakayama <[email protected]>
>
> Add option to get TCP_FASTOPEN to getsockopt()
> ---
> net/ipv4/tcp.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> index 4bd6d52..86186f4 100644
> --- a/net/ipv4/tcp.c
> +++ b/net/ipv4/tcp.c
> @@ -2916,6 +2916,13 @@ static int do_tcp_getsockopt(struct sock *sk, int level,
> case TCP_USER_TIMEOUT:
> val = jiffies_to_msecs(icsk->icsk_user_timeout);
> break;
> +
> + case TCP_FASTOPEN:
> + val = icsk->icsk_accept_queue.fastopenq ?
> + icsk->icsk_accept_queue.fastopenq->max_qlen
> + : sysctl_tcp_fastopen;
> + break;
> +
> case TCP_TIMESTAMP:
> val = tcp_time_stamp + tp->tsoffset;
> break;
> --
> 1.9.0

I have no idea why you keep using sysctl_tcp_fastopen in your patches.

It seems you only want your application to
read /proc/sys/net/ipv4/tcp_fastopen ?

And you don't have CONFIG_PROC_FS maybe, and you add a workaround ?

As already mentioned, even if sysctl_tcp_fastopen is set to one, it does
not mean this socket is fastopen enabled.

setsockopt()/getsockopt() for a given key should be symmetrical.

ie :

getsockopt(fd, SOL_TCP, TCP_FASTOPEN, &val, &len);
setsockopt(fd, SOL_TCP, TCP_FASTOPEN, &val, sizeof(val));

should be a no operation. Its obviously not the case with your proposal.

Please address Neal concerns, and add in the changelog what is this
option supposed to do _exactly_.

hint : max_qlen and sysctl_tcp_fastopen do not have the same units,
its like comparing apples and oranges.

Thanks

2014-04-15 13:57:10

by Neal Cardwell

Subject: Re: [PATCH v2] ipv4: Add option to get TCP_FASTOPEN to getsockopt()

On Tue, Apr 15, 2014 at 8:35 AM, Kenjiro Nakayama
<[email protected]> wrote:
> Hi,

Please omit the "Hi," at the top of the patch description, so the top
of the email can be used directly as a git commit description.

> TCP_FASTOPEN option can be set via setsockopt(), but the value cannot be
> gotten via getsockopt(). This patch adds the option to getsockopt().
>
> Sighned-off-by: Kenjiro Nakayama <[email protected]>

Typo: this should be "Signed-off-by:"

> Add option to get TCP_FASTOPEN to getsockopt()

Likewise, please omit the extra descriptive text below the
"Signed-off-by:" footer.

(And, of course, please note Eric's remarks as well...)

thanks,
neal