From:         Tom Christiansen <tchrist@mox.perl.com>
Organization: Perl Consulting and Training
Subject:      Re: Setuid script problems
Newsgroups:   comp.lang.perl.misc,comp.infosystems.www.authoring.cgi

In comp.lang.perl.misc, mheins@bluestem.prairienet.org (Mike Heins) writes:
:Don Reid (donr@cv.hp.com) wrote:
:: : Taint checking is described in the perlsec(1) man page, and is also
:: : covered in the Camel book.
:
:: :         /(.*)/;
:
:: And the perlsec page says not to use (.*) (which matches anything), but
:: something like (\w*) which removes unwanted characters.  You probably
:: want to create the most restrictive pattern that will fit the data you
:: expect.
:
:Good clarification.  I prefer to check each user-entered item with
:a more rigorous subroutine, but perhaps I should have emphasized it
:a bit more.

Agreed.

:The problem with \w+ for most users is that it won't pass an email
:address, which is the most common item that needs to be "scrubbed".

OK, I've seen this whole thing too often.  It's making me nervous:
I think everyone is going at the scrubbing bits the wrong way.

Why do you need to "scrub" an email address?  I believe that scrubbing for
shells is ab initio a fatally flawed concept. If you EVER EVER EVER call a
program with a chance of a shell getting its hands on stuff, you're
doomed.  You *MUST* use system() with a real list, not a string, or else
the real exec() the same way (which is hard).  See the CGI security FAQ.

For example, here's the wrong way to call a program if you're security
minded:

    system "program $arg1 @list";

and here are right ways:

    system "program", $arg1, @list;
    system "program", split " " => $some_maybe_dangerous_string;

    if ($?) { warn "program didn't exit nicely: $?" }

Avoiding the shell entirely the only safe way.  For pipes and backticks, it's
harder.  You'll can't just do this

    @lines = `program $arg1 @list`;

Because there is no way with backticks or popens to tell not to call
the stupid shell and open up your front door to pillaging.

Instead you need this:

    if (open(PROGRAM, "-|")) {
	@lines = <PROGRAM>;
	close PROGRAM;
	if ($?) { warn "PROGRAM didn't exit nicely: $?" }
    } else {
	exec("program", $arg1, @list) || die "can't exec program: $!";
    } 

For simple mailing, you should call sendmail in a way that will
let it deal with the headers.  User mail agents will be a real
pain, because you're likely to forget things like tilde escapes
or dtiry shell characters.

    open (SENDMAIL, "|/usr/lib/sendmail -oi -t") 
	|| die "cannot fork: $!";

    print SENDMAIL <<QQ_EOF_QQ;
To: $him
From: $me
Subject: Whatever

More text goes here.  Use your 
$vars and @lists if you'd like
to get your \$0.02 in, 
and remember that of course
that `backticks` DO NOT happen.

--me

    login\@host.domain

QQ_EOF_QQ

    close(SENDMAIL);
    if ($?) { warn "sendmail didn't exit nicely: $?" }

Notice how in all these mechanisms, the shell never pokes its problematic
proboscis into your business.  It just can't.  This is critical.  Anything
else is merely potentially flawed insurance, not a provable guarantee.

If you're ever writing to a pipe, you may well want to install a pipe
handler to catch things if the program on the other end bails -- 
or isn't found.  Remember that popens return true even when the 
program isn't found.

    $SIG{PIPE} = sub {
	my($signame) = $_[0];
	my($package,$filename,$line) = caller;
	my($fh);
	for $fh ( \*STDOUT, \*STDERR ) { 
	    print $fh "Received an unexpected SIG$signame at line ",
			    "$line in $filename\n";
	}
	exit(1);
    };

Finally, any security conscious program, like setuid or setgid programs,
network servers, or even simple CGI progams, should always with start
something like this:

    #!/usr/bin/perl -wT
    require 5.001;
    use strict;
    $ENV{PATH} = join ':' => split(" ", <<'Q_EOPATH_Q');
		    /usr/bin
		    /usr/ucb
		    /bin
Q_EOPATH_Q

My goodness, this is almost turning into a FMTEYEWTK on safely calling
other programs while in a state of dubious security.

--tom