[7464] in Kerberos
don't expect Beta6 to work well on AIX4
daemon@ATHENA.MIT.EDU (Sam Hartman)
Tue Jun 11 22:10:57 1996
To: kerberos@MIT.EDU
From: Sam Hartman <hartmans@MIT.EDU>
Date: 11 Jun 1996 21:55:06 -0400
Looking over old messages, I realize that most of this
happened on krb5-bugs and not here. So, most users are probably not
aware that telnetd and rlogin (as compiled in Beta 6) do *not* work on
AIX 4.1.4, and probably other AIX4 varients. Note that they work fine
on AIX3.2.5, but ksu will probably fail on versions previous to
3.2.3e. We have no intension to support ksu on AIX previous to 3.2.3e
at this time, as I believe it would require using undocumented system
calls.
I am interested in tracking down the AIX4 problem, and will
present a partial solution at the end of this message.
Basically, under AIX4.1.4, the kernel panics generally while
running login.krb5 or telnetd, crashing the system and producing a
dump.
At least in the one case where I was able to run crash over
the dump it failed deep within the streams revoke handling code for
the slave pty that had been opened. I suspect that there is a null
pointer dereference somewhere in the kernel when a pty is revoked,
especially if the pty has never been opened or has no outstanding
processes.
irregular access to AIX4 systems, I cannot easily isolate the
problem. Also, I don't maintain any AIX4 RISC/6000s that have support
contracts, so if ends up being a kernel bug, I cannot report it.
Doug Engert worked on this problem and came up with an
interum solution. If in the appl/bsd Makefile or Makefile.in, you
arrange for the symbol DO_NOT_USE_K_LOGIN and USE_LOGIN_F to be
defined, krlogind will work with the vendor-supplied /bin/login.
Then, with a line in /etc/inetd.conf like the following, you might get
it to work:
eklogin stream tcp nowwait root /krb5/sbin/klogind
eklogin -5ec -L /usr/bin/login
Note that the above two lines should be combined into one line
in /etc/inetd.conf.
Doug says that this works fine for him, but I recently managed
to crash an AIX box even with this patch.
If you have any info on this problem, particularly
confirmation that your systems also dump in the kernel revoke routine,
please let me know. If it ends up being a revoke problem, I can
disable use of revoke on AIX4.1.x (or until whatever version IBM fixes
the problem.) Because of the way AIX ptys are allocated, this will
not create a security issue.
--Sam
