diff options
| author | Paolo Abeni <pabeni@redhat.com> | 2025-02-11 13:08:02 +0100 |
|---|---|---|
| committer | Paolo Abeni <pabeni@redhat.com> | 2025-02-11 13:08:02 +0100 |
| commit | ae9b3c0e79bcc154f80f6e862d3085de31bcb3ce (patch) | |
| tree | a30101f515f68abb751a449a79570c0458059f11 /net/ipv4/tcp.c | |
| parent | 812122783ae8b347633b7359a71cfba28ffdb66b (diff) | |
| parent | 1280c26228bd7eb14bdecd67dedbdd871f8fdda5 (diff) | |
Merge branch 'tcp-allow-to-reduce-max-rto'
Eric Dumazet says:
====================
tcp: allow to reduce max RTO
This is a followup of a discussion started 6 months ago
by Jason Xing.
Some applications want to lower the time between each
retransmit attempts.
TCP_KEEPINTVL and TCP_KEEPCNT socket options don't
work around the issue.
This series adds:
- a new TCP level socket option (TCP_RTO_MAX_MS)
- a new sysctl (/proc/sys/net/ipv4/tcp_rto_max_ms)
Admins and/or applications can now change the max rto value
at their own risk.
Link: https://lore.kernel.org/netdev/20240715033118.32322-1-kerneljasonxing@gmail.com/T/
====================
Link: https://patch.msgid.link/20250207152830.2527578-1-edumazet@google.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Diffstat (limited to 'net/ipv4/tcp.c')
| -rw-r--r-- | net/ipv4/tcp.c | 14 |
1 files changed, 13 insertions, 1 deletions
diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c index 2021f2709ec3..992d5c9b2487 100644 --- a/net/ipv4/tcp.c +++ b/net/ipv4/tcp.c @@ -423,7 +423,7 @@ void tcp_init_sock(struct sock *sk) { struct inet_connection_sock *icsk = inet_csk(sk); struct tcp_sock *tp = tcp_sk(sk); - int rto_min_us; + int rto_min_us, rto_max_ms; tp->out_of_order_queue = RB_ROOT; sk->tcp_rtx_queue = RB_ROOT; @@ -432,6 +432,10 @@ void tcp_init_sock(struct sock *sk) INIT_LIST_HEAD(&tp->tsorted_sent_queue); icsk->icsk_rto = TCP_TIMEOUT_INIT; + + rto_max_ms = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rto_max_ms); + icsk->icsk_rto_max = msecs_to_jiffies(rto_max_ms); + rto_min_us = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_rto_min_us); icsk->icsk_rto_min = usecs_to_jiffies(rto_min_us); icsk->icsk_delack_max = TCP_DELACK_MAX; @@ -3807,6 +3811,11 @@ int do_tcp_setsockopt(struct sock *sk, int level, int optname, secs_to_retrans(val, TCP_TIMEOUT_INIT / HZ, TCP_RTO_MAX / HZ)); return 0; + case TCP_RTO_MAX_MS: + if (val < MSEC_PER_SEC || val > TCP_RTO_MAX_SEC * MSEC_PER_SEC) + return -EINVAL; + WRITE_ONCE(inet_csk(sk)->icsk_rto_max, msecs_to_jiffies(val)); + return 0; } sockopt_lock_sock(sk); @@ -4643,6 +4652,9 @@ zerocopy_rcv_out: case TCP_IS_MPTCP: val = 0; break; + case TCP_RTO_MAX_MS: + val = jiffies_to_msecs(tcp_rto_max(sk)); + break; default: return -ENOPROTOOPT; } |