[PATCH] for Re: OSF_SIA bug in 2.3.0p1

[John P Speno]

On Wed, Mar 28, 2001 at 10:29:46AM -0600, Chris Adams wrote:
> Once upon a time, John P Speno <speno at isc.upenn.edu> said:
> > Could you test these patches on your Tru64 UNIX 4.x and 5.x systems. They
> > implement the above ideas. In short, do_login is skipped when HAVE_OSF_SIA
> > is enabled since the things do_login does are also done better in the
> > Tru64 SIA routines.
> > 
> > Also, session_setup_sia will now show /etc/motd if appropriate. I needed a place to
> > stick this, and session_setup_sia in auth-sia.c seemed ok at the time. I'm not sure of
> > that now. Consider this a first draft for changes:
> 
> It looks good, except you don't check for .hushlogin.  I pulled
> .hushlogin checking and MOTD printing into separate functions in
> session.c (to avoid code duplication).

Great. Nice work. The patches look good to me.

> There is still a problem (maybe someone else can see it): there is a
> race condition in displaying the error message back to the user when a
> session is not started.  Sometimes you get (when connecting to a locked
> account):
> 
> $ ssh -l burdell fly
> Account is disabled -- see Account Administrator.
> 
> Connection to fly closed by remote host.
> Connection to fly closed.
> $ 
> 
> and sometimes you get:
> 
> $ ssh -l burdell fly
> Connection to fly closed by remote host.
> Connection to fly closed.
> $ 

I wasn't able to duplicate that in my testing, though I was just ssh'ing to my
localhost, and maybe it's related to network delay.

Perhaps you can flush the output stream in cases where sia_ses_estab fails.

Regardless, I think these changes should be incorporated into the CVS!

Thanks.