Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site ccivax.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!mit-eddie!genrad!panda!talcott!harvard!seismo!rochester!ritcv!ccivax!crp
From: crp@ccivax.UUCP (Chuck Privitera)
Newsgroups: net.bugs.uucp,net.bugs.4bsd
Subject: Uucp incorrectly chooses 't' protocol (1 of 2)
Message-ID: <242@ccivax.UUCP>
Date: Wed, 1-May-85 14:09:08 EDT
Article-I.D.: ccivax.242
Posted: Wed May  1 14:09:08 1985
Date-Received: Fri, 3-May-85 04:47:25 EDT
Distribution: net
Organization: CCI Telephony Systems Group,  Rochester NY
Lines: 90
Xref: watmath net.bugs.uucp:521 net.bugs.4bsd:1488


Index:	usr.bin/uucp/cico.c 4.2BSD FIX

Description:
	The (self proclaimed kludgy) IsTcpIp flag is initialized 
	in uucpdefs.[ch] (depending on what version of uucp you
	have) to 0 and in cico.c to 1 if the line is determined
	to be to a TCP/IP host. However, once it is set to 1, it
	never gets cleared. Thus, running the uucp queue with jobs
	queued to TCP/IP machines as well as to non-TCP/IP machines
	AND the TCP/IP machine(s) get called first, when uucico gets
	to the non-TCP/IP machine it will incorrectly choose the
	't' protocol (if the other machine supports it).
Repeat-By:
	Queue a job to a tcp/ip host and to a non-tcp/ip host such
	that the non-tcp/ip site will be called last. If you don't
	talk to any TCP/IP machines, you can add a line like:

	Localhost Any TCP uucp Localhost login: uucp

	To your L.sys file and a line like:

	TCP ttyXX unused 9600 TCP

	To your L-devices file. Then queue something to Localhost
	(or whatever) and to a non-tcp/ip machine (who is gauranteed
	to be called second, and supports the 't' protocol) with
	uucp -r so the job is only queued. Then run uucico by hand
	with debugging turned on so you can see the protocol selection
	and everything. Notice that the 't' protocol will also
	be selected for the non-tcp/ip site.

	If the non-tcp/ip site has a version of the 't' protocol
	compatible with yours (there was a substantial change
	to the 't' protocol sometime around last July), the
	conversation will probably succeed, if not you will probably
	get a core dump in trddata() (because previously the total 
	transmission byte count was sent at the beginning of the
	conversation, and now a byte count is sent at the front of
	each buffer). Even if the conversation goes OK, the integrity
	of the data cannot be guaranteed because nobody was checking
	for integrity (normally, the 'g' protocol or TCP ensures 
	data integrity, but TCP wasn't really the transport
	mechanism, the tty driver was.)
Fix:
	The fix is simple, just explicitly reset the IsTcpIp flag
	if it is determined that the new connection is to a non-tcp/ip
	site. A diff listing of the old and new code follows (line
	numbers are probably off). See the next article to find out
	what you can do to protect yourself from remote sites with
	this problem.

RCS file: RCS/cico.c,v
retrieving revision 1.30
diff -c3 -r1.30 cico.c
*** /tmp/,RCSt1003841	Wed May  1 14:07:27 1985
--- cico.c	Wed May  1 10:47:01 1985
***************
*** 1,5
  #ifndef lint
! static char	*RcsId = "@(#) $Header: cico.c,v 1.30 85/03/17 17:42:51 rick Exp $";
  /* from: @(#)cico.c	5.3 (Berkeley) 10/3/83 */
  #endif !lint
  

--- 1,5 -----
  #ifndef lint
! static char	*RcsId = "@(#) $Header: cico.c,v 1.31 85/05/01 10:42:33 root Exp $";
  /* from: @(#)cico.c	5.3 (Berkeley) 10/3/83 */
  #endif !lint
  
***************
*** 499,505
  		if (isatty(Ifn) ==  0) {
  			IsTcpIp = 1;
  			DEBUG(4, "TCPIP connection -- ioctl-s disabled\n", CNULL);
! 		}
  #endif
  
  		if (setjmp(Sjbuf))

--- 499,506 -----
  		if (isatty(Ifn) ==  0) {
  			IsTcpIp = 1;
  			DEBUG(4, "TCPIP connection -- ioctl-s disabled\n", CNULL);
! 		} else
! 			IsTcpIp = 0;
  #endif
  
  		if (setjmp(Sjbuf))