[1685] in Kerberos-V5-bugs
libpty wtmp handling not quite right
daemon@ATHENA.MIT.EDU (Sam Hartman)
Sun Oct 22 15:48:07 1995
Date: Sun, 22 Oct 1995 15:47:55 -0400
From: Sam Hartman <hartmans@MIT.EDU>
To: krb5-bugs@MIT.EDU
OK, apparently at least Solaris wants the username in a logout
record in wtmp and wtmpx. This is really broken because it means that
telnetd has to grovel through, find the utmp record for the line and
then pull the username. It's kind of scary to think that init has to
do this as well. If you don't put the username in the logout record,
last gets somewhat confused. If you do a last username, it pulls only
the records with that username, then finds logins and logouts. I.E.
It'll miss the logout records. However, if you do a last |grep
username, it will notice the logout records and report statistics
correctly.
This is a minor bug, so I'll leave libpty the way it is until
I decide how to fix this.In particular, I would appreciate any help
determining what feature set implies this behavior. It is decidely
not present on systems like Ultrix or NetBSD--where you don't have
ut_type, so a null username implies logout. It is also not present on
AIX, and I don't think on OSF (although pao's wtmp is somewhat
questionable as an example). If I could get telnetd working on an
SGI, I would test that. I hope it is related to presence of utmpx or
something mildly sane like that.
When I fix this problem, I'll fix the ttyslot problem for
Sunos. I'm still thinking about how to change the interfaces to
convey what I'm trying to do at an abstract level.
--Sam
