openssh unix dev - Apr 2004 - patch for "gone - no logout" output from last(1)

Greetings,

After upgrading one of my systems to the linux 2.6.4 kernel, it started 
having problems recording the logout entries in the wtmp file.  I traced 
this to a bug in loginrec.c, where a variable (line) is declared as an 
array of 8 characters.  This length is substantially shorter than the 16 
character array size of ut_line, so openssh effectively truncates the 
pseudo-tty name when the pseudo-tty number exceeds 999.  With the new 
linux kernels, the pseudo-ttys are not reused as in the past, so a busy 
system can reach the magic number of 1000 pretty quickly...

Here's an example of what I see on my system via the last command, so you 
can get an idea.  The top 2 entries were after I applied the patch below.
Prior to its application, every login with a four-digit pts/ number that
had completed was marked with "gone - no logout".

  $ last -20 | grep pts/
  jpeisen  pts/2057     some.host.com    Wed Apr  7 20:00 - 20:00  (00:00)
  jpeisen  pts/2052     localhost        Wed Apr  7 19:31 - 19:31  (00:00)
  jpeisen  pts/2051     localhost        Wed Apr  7 19:26    gone - no logout
  jpeisen  pts/2050     some.host.com    Wed Apr  7 19:26   still logged in
  jpeisen  pts/2046     some.host.com    Wed Apr  7 19:07   still logged in
  jpeisen  pts/2035     some.host.com    Wed Apr  7 17:27    gone - no logout
  jpeisen  pts/2030     some.host.com    Wed Apr  7 16:46   still logged in
  mgosk    pts/2021     other.host.com   Wed Apr  7 15:28   still logged in
  mgosk    pts/2019     other.host.com   Wed Apr  7 15:14    gone - no logout
  jmoras   pts/2013     other.host.com   Wed Apr  7 14:33   still logged in

Below is my version of the patch.  Since I'm not on this list, there may
be other, better versions that have already found their way here.  If so,
then I apologize for wasting your time.  I did try to be good and wrap the
change in an appropriate #ifdef, and the change is very simple - just using
UT_LINESIZE for the length of the array if it is defined.

-John

-- 
John P. Eisenmenger
jpe at eisenmenger.org

*** openssh-3.7.1p2/loginrec.c.orig	Sun Jul  6 01:20:46 2003
--- openssh-3.7.1p2/loginrec.c	Wed Apr  7 19:35:43 2004
***************
*** 1349,1355 ****
--- 1349,1359 ----
  syslogin_perform_logout(struct logininfo *li)
  {
  # ifdef HAVE_LOGOUT
+ # ifdef UT_LINESIZE
+ 	char line[UT_LINESIZE];
+ # else
  	char line[8];
+ # endif
  
  	(void)line_stripname(line, li->line, sizeof(line));

John P. Eisenmenger wrote:
> After upgrading one of my systems to the linux 2.6.4 kernel, it started 
> having problems recording the logout entries in the wtmp file.  I traced 
> this to a bug in loginrec.c, where a variable (line) is declared as an 
> array of 8 characters.  This length is substantially shorter than the 16 
> character array size of ut_line, so openssh effectively truncates the 
> pseudo-tty name when the pseudo-tty number exceeds 999.  With the new
> linux kernels, the pseudo-ttys are not reused as in the past
Is that deliberate?  Sounds like a kernel bug...

[snip]
> Below is my version of the patch.
Seems reasonable, applied.  Thanks.

-- 
Darren Tucker (dtucker at zip.com.au)
GPG key 8FF4FA69 / D9A3 86E9 7EEE AF4B B2D4  37C9 C982 80C7 8FF4 FA69
     Good judgement comes with experience. Unfortunately, the experience
usually comes from bad judgement.

Darren Tucker wrote:
> John P. Eisenmenger wrote:
>> Below is my version of the patch.
> 
> Seems reasonable, applied.  Thanks.
Hmm, I wonder if it would be cleaner to have this in defines.h:
#ifndef UT_LINESIZE
# define UT_LINESIZE 8
#endif

Then just have this in loginrec.c:
char line[UT_LINESIZE];

-- 
Darren Tucker (dtucker at zip.com.au)
GPG key 8FF4FA69 / D9A3 86E9 7EEE AF4B B2D4  37C9 C982 80C7 8FF4 FA69
     Good judgement comes with experience. Unfortunately, the experience
usually comes from bad judgement.

On Thu, 8 Apr 2004, Darren Tucker wrote:
> Darren Tucker wrote:
> 
> > John P. Eisenmenger wrote:
> > > Below is my version of the patch.
> > 
> > Seems reasonable, applied.  Thanks.
> 
> Hmm, I wonder if it would be cleaner to have this in defines.h:
> #ifndef UT_LINESIZE
> # define UT_LINESIZE 8
> #endif
> 
> Then just have this in loginrec.c:
> char line[UT_LINESIZE];
 
I like this better.
 

-- 
Tim Rice				Multitalents	(707) 887-1469
tim at multitalents.net