Partial workaround for TCP Fast Open issue. Bug 2907

4 files changed, 17 insertions, 1 deletions

diff --git a/src/src/smtp_out.c b/src/src/smtp_out.c
index 7b7bdf752..7f364d942 100644
--- a/src/src/smtp_out.c
+++ b/src/src/smtp_out.c
@@ -372,7 +372,7 @@ if (!save_errno)
 # ifdef TCP_FASTOPEN_CONNECT
     else
       {						/* expecting client data */
-      debug_printf(" set up lazy-connect\n");
+      DEBUG(D_transport|D_acl|D_v) debug_printf(" set up lazy-connect\n");
       setsockopt(sock, IPPROTO_TCP, TCP_FASTOPEN_CONNECT, US &on, sizeof(on));
       /* fastopen_blob = NULL;		 lazy TFO, triggered by data write */
       }
diff --git a/src/src/transports/smtp.c b/src/src/transports/smtp.c
index d7f251b0b..bbff1cad8 100644
--- a/src/src/transports/smtp.c
+++ b/src/src/transports/smtp.c
@@ -5312,6 +5312,17 @@ retry_non_continued:
     uschar *retry_message_key = NULL;
     uschar *serialize_key = NULL;
 
+    /* Deal slightly better with a possible Linux kernel bug that results
+    in intermittent TFO-conn fails deep into the TCP flow.  Bug 2907 tracks.
+    Hack: Clear TFO option for any further hosts on this tpt run. */
+
+    if (total_hosts_tried > 0)
+      {
+      DEBUG(D_transport|D_acl|D_v)
+	debug_printf("Clearing TFO as not first host for message\n");
+      ob->hosts_try_fastopen = US"";
+      }
+
     /* Default next host is next host. :-) But this can vary if the
     hosts_max_try limit is hit (see below). It may also be reset if a host
     address is looked up here (in case the host was multihomed). */
diff --git a/test/stderr/0388 b/test/stderr/0388
index c3c11ca75..ac9dc436a 100644
--- a/test/stderr/0388
+++ b/test/stderr/0388
@@ -110,6 +110,7 @@ set_process_info: pppp delivering 10HmaX-0005vi-00: just tried 127.0.0.1 [127.0.
 address match test: subject=*@127.0.0.1 pattern=*
 127.0.0.1 in "*"? yes (matched "*")
 *@127.0.0.1 in "*"? yes (matched "*")
+Clearing TFO as not first host for message
 checking status of V4NET.0.0.0
  locking TESTSUITE/spool/db/retry.lockfile
  locked  TESTSUITE/spool/db/retry.lockfile
diff --git a/test/stderr/0512 b/test/stderr/0512
index e54ebb17e..630242d13 100644
--- a/test/stderr/0512
+++ b/test/stderr/0512
@@ -20,6 +20,7 @@ LOG: MAIN
   H=127.0.0.1 [127.0.0.1] Connection refused
 added retry item for T:127.0.0.1:127.0.0.1:PORT_S: errno=dd more_errno=dd,A flags=2
 temporary delivery error(s) override hosts_max_try (message older than host's retry time)
+Clearing TFO as not first host for message
 getting address for 127.0.0.1
 checking status of 127.0.0.1
 no message retry record
@@ -31,6 +32,7 @@ LOG: MAIN
   H=127.0.0.1 [127.0.0.1] Connection refused
 added retry item for T:127.0.0.1:127.0.0.1:PORT_S: errno=dd more_errno=dd,A flags=2
 temporary delivery error(s) override hosts_max_try (message older than host's retry time)
+Clearing TFO as not first host for message
 getting address for 127.0.0.1
 checking status of 127.0.0.1
 no message retry record
@@ -42,6 +44,7 @@ LOG: MAIN
   H=127.0.0.1 [127.0.0.1] Connection refused
 added retry item for T:127.0.0.1:127.0.0.1:PORT_S: errno=dd more_errno=dd,A flags=2
 temporary delivery error(s) override hosts_max_try (message older than host's retry time)
+Clearing TFO as not first host for message
 getting address for 127.0.0.1
 checking status of 127.0.0.1
 no message retry record
@@ -108,6 +111,7 @@ LOG: MAIN
   H=127.0.0.1 [127.0.0.1] Connection refused
 added retry item for T:127.0.0.1:127.0.0.1:PORT_S: errno=dd more_errno=dd,A flags=2
 temporary delivery error(s) override hosts_max_try (message older than host's retry time)
+Clearing TFO as not first host for message
 getting address for 127.0.0.1
 checking status of 127.0.0.1
 no message retry record