On Sat, Aug 28, 2010 at 4:13 PM, David Miller [off-list ref] wrote:

From: "H.K. Jerry Chu" <redacted>
Date: Fri, 27 Aug 2010 22:13:28 -0700

quoted

@@ -556,7 +559,14 @@ static void tcp_keepalive_timer (unsigned long data)

      elapsed = keepalive_time_elapsed(tp);

      if (elapsed >= keepalive_time_when(tp)) {
-             if (icsk->icsk_probes_out >= keepalive_probes(tp)) {
+             /* If the TCP_USER_TIMEOUT option is enabled, use that
+              * to determine when to timeout instead.
+              */
+             if ((icsk->icsk_user_timeout != 0 &&
+                 elapsed >= icsk->icsk_user_timeout &&
+                 icsk->icsk_probes_out > 0) ||
+                 (icsk->icsk_user_timeout == 0 &&
+                 icsk->icsk_probes_out >= keepalive_probes(tp))) {
                      tcp_send_active_reset(sk, GFP_ATOMIC);
                      tcp_write_err(sk);
                      goto out;

I think if we want to add a socket option which overrides, it makes
more sense to provide overrides in the same units.  This
transformation here is transforming a check against apples into a
check against oranges.

But if that's how this thing is specified, so be it... I guess. :-/

Correct.  It seems that there has been a bit of inconsistency regarding the
unit of "timeouts". RFC1122 says "R1 and R2 might be measured in time
units or as a count of retransmissions." Most of the OSes including Linux
seem to measure timeout in # of retries. But RFC5482 defines its "User
Timeout Option" in time units.

Personally I think as an API, it's easier for an application to grasp
the concept
of a time quantity than # of retransmissions. (E.g., how will an app
determine it
needs 10 retries vs 20 retries?)

Jerry