Re: [Tiger-devel] Merge update and problems with check_logfiles merge.

[Javier Fernandez-Sanguino]

Ryan Bradetich wrote:
> Hello all,
>
>
> I have managed to sync up my local tree against the current CVS HEAD. 
> The only file that requires a non-trivial merge was the
> scripts/check_logfiles.
(...)
> Not all of the patch was applied, so please ignore the rest of this
> email if the patch is still in the process of being applied.

It is now, you are right to pointing at some pieces that I missed.

> My first concern is with this piece of code:
(...)
> I agree that the LOGINLOG does not appear to be present under Linux, 
> but BTMP definately is.  By default the BTMP file is not present, but
> if the file is created, bad login data will be stored in the file.

You are right. I will add it to the check.
> The next section I need has the HP-UX specific logfiles.  I see the way 
> to modify which files are present has changes slightly, no problem I can
> regenerate a patch in that format once agree on the patch contents :)
What hast changed? There has not been that much of a change... I 
probably have to add code to say something if the variables are not defined.

> The next section is the permission check for the BTMP files was not
> added :)  Generally the permissions on the BTMP file need to be tighter
(...)

You are right.

> The WTMP, LOGINLOG, UTMP, LASTLOG, and MESSAGES do not check to see if
> the variables are defined to a file before throwing an error, claiming
> the file does not exist.  Arguably the WTMP and UTMP will be present on
> all systems, but according to the lastb man page WTMP is a local
> configuration option as well.  HP-UX only has WTMP, UTMP, and BTMP
> defined, and Tru64 only has WTMP and UTMP defined.  I think it is a good
> idea to verify the log file is defined before throwing a message it does
> not exist.

Ok. I will add some message in the else clause of the [ -n "LOGFILE" ]. 
Maybe a configuration error.

> Thoughts/feedback?
>
>
> I will be happy to generate a new patch against the CVS HEAD with this 
> if everyone is agreeable, or I can make any changes once I understand
> what is wrong with my logic :)

The only issue is that I don't want to do an
+if [ -n "$UTMP" ] ; then
(...)
+fi

Without an else clause, some variables should be defined since they are 
configured in the script:

WTMP=/var/log/wtmp
BTMP=/var/log/btmp
LASTLOG=/var/log/lastlog
UTMP=/var/run/utmp
LOGINLOG=/var/log/loginlog
MESSAGES=/var/log/messages

And then modified per OS. It might make more sense to define this 
variables in the config script of each OS with a failback here if not 
defined or add them to the haveallvars check (after the 'haveallcmds'). 
A proper flow could be:

1.- each OS config script defines (and exports) proper 
WTMP/BTMP/LASTLOG... variables

2.- The script checks if a common set of variables are defined (WTMP, 
UTMP and MESSAGES?), this is done by haveallvars. If they are not it 
will abort.

3.- Other variables (BTMP, LASTLOG, LOGINLOG, BTMP..) must be checked 
before using them (with the [ -n "XXX" ] code) but no warning will be 
given if they are not available.

This avoids adding the "$OS" clauses and puts that configuration into 
where it should be (systems/$OS/config).

I will have to patch some more pieces than just this script for that to 
work. Give me some time and I will implement it using some of the 
information you provided in the patch and in your mail.

Thanks!

Javi