[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANn89iL3E6TtmoNfHziZJbg-8nTZRg8-D3LyFg6g5GLvTmDfaA@mail.gmail.com>
Date: Tue, 25 Mar 2025 12:09:42 +0100
From: Eric Dumazet <edumazet@...gle.com>
To: Jason Xing <kerneljasonxing@...il.com>
Cc: davem@...emloft.net, kuba@...nel.org, pabeni@...hat.com, horms@...nel.org,
ncardwell@...gle.com, kuniyu@...zon.com, dsahern@...nel.org,
netdev@...r.kernel.org
Subject: Re: [PATCH net-next v4 1/2] tcp: support TCP_RTO_MIN_US for
set/getsockopt use
On Mon, Mar 17, 2025 at 1:03 PM Jason Xing <kerneljasonxing@...il.com> wrote:
>
> Support adjusting/reading RTO MIN for socket level by using set/getsockopt().
>
> This new option has the same effect as TCP_BPF_RTO_MIN, which means it
> doesn't affect RTAX_RTO_MIN usage (by using ip route...). Considering that
> bpf option was implemented before this patch, so we need to use a standalone
> new option for pure tcp set/getsockopt() use.
>
> When the socket is created, its icsk_rto_min is set to the default
> value that is controlled by sysctl_tcp_rto_min_us. Then if application
> calls setsockopt() with TCP_RTO_MIN_US flag to pass a valid value, then
> icsk_rto_min will be overridden in jiffies unit.
>
> This patch adds WRITE_ONCE/READ_ONCE to avoid data-race around
> icsk_rto_min.
>
> Signed-off-by: Jason Xing <kerneljasonxing@...il.com>
> ---
> Documentation/networking/ip-sysctl.rst | 4 ++--
> include/net/tcp.h | 2 +-
> include/uapi/linux/tcp.h | 1 +
> net/ipv4/tcp.c | 13 ++++++++++++-
> 4 files changed, 16 insertions(+), 4 deletions(-)
>
> diff --git a/Documentation/networking/ip-sysctl.rst b/Documentation/networking/ip-sysctl.rst
> index 054561f8dcae..5c63ab928b97 100644
> --- a/Documentation/networking/ip-sysctl.rst
> +++ b/Documentation/networking/ip-sysctl.rst
> @@ -1229,8 +1229,8 @@ tcp_pingpong_thresh - INTEGER
> tcp_rto_min_us - INTEGER
> Minimal TCP retransmission timeout (in microseconds). Note that the
> rto_min route option has the highest precedence for configuring this
> - setting, followed by the TCP_BPF_RTO_MIN socket option, followed by
> - this tcp_rto_min_us sysctl.
> + setting, followed by the TCP_BPF_RTO_MIN and TCP_RTO_MIN_US socket
> + options, followed by this tcp_rto_min_us sysctl.
>
> The recommended practice is to use a value less or equal to 200000
> microseconds.
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index 7207c52b1fc9..6a7aab854b86 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -806,7 +806,7 @@ u32 tcp_delack_max(const struct sock *sk);
> static inline u32 tcp_rto_min(const struct sock *sk)
> {
> const struct dst_entry *dst = __sk_dst_get(sk);
> - u32 rto_min = inet_csk(sk)->icsk_rto_min;
> + u32 rto_min = READ_ONCE(inet_csk(sk)->icsk_rto_min);
>
> if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
> rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
> diff --git a/include/uapi/linux/tcp.h b/include/uapi/linux/tcp.h
> index 32a27b4a5020..b2476cf7058e 100644
> --- a/include/uapi/linux/tcp.h
> +++ b/include/uapi/linux/tcp.h
> @@ -137,6 +137,7 @@ enum {
>
> #define TCP_IS_MPTCP 43 /* Is MPTCP being used? */
> #define TCP_RTO_MAX_MS 44 /* max rto time in ms */
> +#define TCP_RTO_MIN_US 45 /* min rto time in us */
>
> #define TCP_REPAIR_ON 1
> #define TCP_REPAIR_OFF 0
> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> index 46951e749308..b89c1b676b8e 100644
> --- a/net/ipv4/tcp.c
> +++ b/net/ipv4/tcp.c
> @@ -3352,7 +3352,7 @@ int tcp_disconnect(struct sock *sk, int flags)
> icsk->icsk_probes_out = 0;
> icsk->icsk_probes_tstamp = 0;
> icsk->icsk_rto = TCP_TIMEOUT_INIT;
> - icsk->icsk_rto_min = TCP_RTO_MIN;
> + WRITE_ONCE(icsk->icsk_rto_min, TCP_RTO_MIN);
Semi-orthogonal to your patch, apparently at disconnect() we throw
away user/eBPF choice.
Most socket options, once set via setsockopt() should stay even if the
socket is re-used.
Reviewed-by: Eric Dumazet <edumazet@...gle.com>
Powered by blists - more mailing lists