[4830] in Kerberos

home help back first fref pref prev next nref lref last post

Kerb V5 portability

daemon@ATHENA.MIT.EDU (Tomas Gustavsson)
Fri Mar 17 04:48:40 1995

To: kerberos@MIT.EDU
Date: 15 Mar 1995 13:47:27 GMT
From: tomasg@mars.dsv.su.se (Tomas Gustavsson)

I have just been compiling Kerberos V5 Beta4.2 on a HP machine
model 9000/7xx running HP-UX 9.x.
These are my opinions from the porting (anyone has some other?)
Maybe all of this has been up before, but I haven't seen it.

The kerberos libraries ports nicely, but the bsd applications require
some changes.

in krcp.c -line 62 i added #define HAVE_SETREUID, since this is the
define that is used in the code to get the right one. Without this
define setuid() is tried but HP uses setresuid().
This I believe is a small bug!

in krshd.c -line 178 I added #define setpgrp... just like below in
#ifdef __svr4__. __svr4__ is not defined on my HP, and the define is
needed.
Also small bug i believe!

in krlogind.c -line 136 I added #include <sys/bsdtty.h> inside the
#ifdef hpux. This is because krlogind uses TIOCGPGRP which is nowhere
except in bsdtty.h on HP.
I'm not REALLY shure this is the right way?

Last in bsd appls, login.c does not seem to be very prtable at all.
Has anyone ported this to HP?
Lots of terminal stuff (TIOCxxxx), which is not in HP. I added my HP
into the AIX and IBMR2 defines just to pass it with the
compiler. Probably does not work for real. 
Hints anyone?

In appl/telnet/telnetd/sys_term.c ther are also problems with
setpgrp(). In the beginning there is a macro under #ifdef
SETPGRP_TWOARGS, that should make it right. But in the code there are
faulty calls to setpgrp (without args). I changed these calls to
setpgrp(0,0) instead of setpgrp(0). Off course all calls should be
according to #ifdefs and macros in the beginning.

My last opinion is about the gss-api. This library uses the deep rsa
functions from mit (MD5Init etc), instead of the kerberos api.
This is bad since it forces you to make changes in 2 places, if you
want to switch from mit's des-rsa library to something of your own.
I think the gss-api should use kerberos functions like md5_sum_func
instead.

Thanks!
===================================
Tomas Gustavsson 
Dept. of Computer and System Sciences, Stockholm University/KTH
Email: tomasg@dsv.su.se
tel: +46 8 16 17 01

home help back first fref pref prev next nref lref last post