Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!PURDUE.EDU!narten
From: narten@PURDUE.EDU (Thomas Narten)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: Odd FTP Problem (4.3 BSD kernel patches)
Message-ID: <8902251856.AA07509@percival.cs.purdue.edu>
Date: 25 Feb 89 18:55:57 GMT
References: <8902241624.AA26178@monk.proteon.com>
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 44

[ Stuff about using adb to xero out errors in inetctlerrmap ]

The suggested fix causes 4.3 to ignore ICMP unreachable errors in all
cases, something that one probably does not want to do.  For instance,
I *much* prefer to have telnet attempts abort quickly with a "network
unreachable" than with a "connection timed out" some 60 seconds later.
On the other hand, once a connection has been established, I'd prefer
stray ICMP errors not break a connection.  Moreover, nuking ICMP
unreachable errors weakens utilities like ping that understand such
messages.  One of ping's useful features is the printing of ICMP
errors it receives.

The following patch (perhaps not pretty, but precise) treats ICMP
unreachable errors as before, except that they won't break established
connections.

Thomas

*** /tmp/,RCSt1025442	Sat Feb 25 13:46:58 1989
--- /tmp/,RCSt2025442	Sat Feb 25 13:46:59 1989
***************
*** 258,264 ****
  tcp_notify(inp)
  	register struct inpcb *inp;
  {
! 
  	wakeup((caddr_t) &inp->inp_socket->so_timeo);
  	sorwakeup(inp->inp_socket);
  	sowwakeup(inp->inp_socket);
--- 258,271 ----
  tcp_notify(inp)
  	register struct inpcb *inp;
  {
! 	if (inp->inp_socket->so_state != SS_ISCONNECTING) {
! 	        register int error = inp->inp_socket->so_error;
! 	        if ((error == EHOSTUNREACH) || (error == ENETUNREACH)
! 		     || (error == EHOSTDOWN)) {
! 		        inp->inp_socket->so_error = 0; /* clear error */
! 			return;
! 		}
! 	} 
  	wakeup((caddr_t) &inp->inp_socket->so_timeo);
  	sorwakeup(inp->inp_socket);
  	sowwakeup(inp->inp_socket);