Unused 'cv'

[perl5.git] / pod / perlsec.pod

=head1 NAME

perlsec - Perl security

=head1 DESCRIPTION

Perl is designed to make it easy to program securely even when running
with extra privileges, like setuid or setgid programs.  Unlike most
command line shells, which are based on multiple substitution passes on
each line of the script, Perl uses a more conventional evaluation scheme
with fewer hidden snags.  Additionally, because the language has more
builtin functionality, it can rely less upon external (and possibly
untrustworthy) programs to accomplish its purposes.

Perl automatically enables a set of special security checks, called I<taint
mode>, when it detects its program running with differing real and effective
user or group IDs.  The setuid bit in Unix permissions is mode 04000, the
setgid bit mode 02000; either or both may be set.  You can also enable taint
mode explicitly by using the B<-T> command line flag. This flag is
I<strongly> suggested for server programs and any program run on behalf of
someone else, such as a CGI script. Once taint mode is on, it's on for
the remainder of your script.

While in this mode, Perl takes special precautions called I<taint
checks> to prevent both obvious and subtle traps.  Some of these checks
are reasonably simple, such as verifying that path directories aren't
writable by others; careful programmers have always used checks like
these.  Other checks, however, are best supported by the language itself,
and it is these checks especially that contribute to making a set-id Perl
program more secure than the corresponding C program.

You may not use data derived from outside your program to affect
something else outside your program--at least, not by accident.  All
command line arguments, environment variables, locale information (see
L<perllocale>), results of certain system calls (C<readdir()>,
C<readlink()>, the variable of C<shmread()>, the messages returned by
C<msgrcv()>, the password, gcos and shell fields returned by the
C<getpwxxx()> calls), and all file input are marked as "tainted".
Tainted data may not be used directly or indirectly in any command
that invokes a sub-shell, nor in any command that modifies files,
directories, or processes, B<with the following exceptions>:

=over 4

=item *

Arguments to C<print> and C<syswrite> are B<not> checked for taintedness.

=item *

Symbolic methods

    $obj->$method(@args);

and symbolic sub references

    &{$foo}(@args);
    $foo->(@args);

are not checked for taintedness.  This requires extra carefulness
unless you want external data to affect your control flow.  Unless
you carefully limit what these symbolic values are, people are able
to call functions B<outside> your Perl code, such as POSIX::system,
in which case they are able to run arbitrary external code.

=item *

Hash keys are B<never> tainted.

=back

For efficiency reasons, Perl takes a conservative view of
whether data is tainted.  If an expression contains tainted data,
any subexpression may be considered tainted, even if the value
of the subexpression is not itself affected by the tainted data.

Because taintedness is associated with each scalar value, some
elements of an array or hash can be tainted and others not.
The keys of a hash are B<never> tainted.

For example:

    $arg = shift;		# $arg is tainted
    $hid = $arg, 'bar';		# $hid is also tainted
    $line = <>;			# Tainted
    $line = <STDIN>;		# Also tainted
    open FOO, "/home/me/bar" or die $!;
    $line = <FOO>;		# Still tainted
    $path = $ENV{'PATH'};	# Tainted, but see below
    $data = 'abc';		# Not tainted

    system "echo $arg";		# Insecure
    system "/bin/echo", $arg;	# Considered insecure
				# (Perl doesn't know about /bin/echo)
    system "echo $hid";		# Insecure
    system "echo $data";	# Insecure until PATH set

    $path = $ENV{'PATH'};	# $path now tainted

    $ENV{'PATH'} = '/bin:/usr/bin';
    delete @ENV{'IFS', 'CDPATH', 'ENV', 'BASH_ENV'};

    $path = $ENV{'PATH'};	# $path now NOT tainted
    system "echo $data";	# Is secure now!

    open(FOO, "< $arg");	# OK - read-only file
    open(FOO, "> $arg"); 	# Not OK - trying to write

    open(FOO,"echo $arg|");	# Not OK
    open(FOO,"-|")
	or exec 'echo', $arg;	# Also not OK

    $shout = `echo $arg`;	# Insecure, $shout now tainted

    unlink $data, $arg;		# Insecure
    umask $arg;			# Insecure

    exec "echo $arg";		# Insecure
    exec "echo", $arg;		# Insecure
    exec "sh", '-c', $arg;	# Very insecure!

    @files = <*.c>;		# insecure (uses readdir() or similar)
    @files = glob('*.c');	# insecure (uses readdir() or similar)

    # In Perl releases older than 5.6.0 the <*.c> and glob('*.c') would
    # have used an external program to do the filename expansion; but in
    # either case the result is tainted since the list of filenames comes
    # from outside of the program.

    $bad = ($arg, 23);		# $bad will be tainted
    $arg, `true`;		# Insecure (although it isn't really)

If you try to do something insecure, you will get a fatal error saying
something like "Insecure dependency" or "Insecure $ENV{PATH}".

The exception to the principle of "one tainted value taints the whole
expression" is with the ternary conditional operator C<?:>.  Since code
with a ternary conditional

    $result = $tainted_value ? "Untainted" : "Also untainted";

is effectively

    if ( $tainted_value ) {
        $result = "Untainted";
    } else {
        $result = "Also untainted";
    }

it doesn't make sense for C<$result> to be tainted.

=head2 Laundering and Detecting Tainted Data

To test whether a variable contains tainted data, and whose use would
thus trigger an "Insecure dependency" message, you can use the
C<tainted()> function of the Scalar::Util module, available in your
nearby CPAN mirror, and included in Perl starting from the release 5.8.0.
Or you may be able to use the following C<is_tainted()> function.

    sub is_tainted {
        return ! eval { eval("#" . substr(join("", @_), 0, 0)); 1 };
    }

This function makes use of the fact that the presence of tainted data
anywhere within an expression renders the entire expression tainted.  It
would be inefficient for every operator to test every argument for
taintedness.  Instead, the slightly more efficient and conservative
approach is used that if any tainted value has been accessed within the
same expression, the whole expression is considered tainted.

But testing for taintedness gets you only so far.  Sometimes you have just
to clear your data's taintedness.  Values may be untainted by using them
as keys in a hash; otherwise the only way to bypass the tainting
mechanism is by referencing subpatterns from a regular expression match.
Perl presumes that if you reference a substring using $1, $2, etc., that
you knew what you were doing when you wrote the pattern.  That means using
a bit of thought--don't just blindly untaint anything, or you defeat the
entire mechanism.  It's better to verify that the variable has only good
characters (for certain values of "good") rather than checking whether it
has any bad characters.  That's because it's far too easy to miss bad
characters that you never thought of.

Here's a test to make sure that the data contains nothing but "word"
characters (alphabetics, numerics, and underscores), a hyphen, an at sign,
or a dot.

    if ($data =~ /^([-\@\w.]+)$/) {
	$data = $1; 			# $data now untainted
    } else {
	die "Bad data in '$data'"; 	# log this somewhere
    }

This is fairly secure because C</\w+/> doesn't normally match shell
metacharacters, nor are dot, dash, or at going to mean something special
to the shell.  Use of C</.+/> would have been insecure in theory because
it lets everything through, but Perl doesn't check for that.  The lesson
is that when untainting, you must be exceedingly careful with your patterns.
Laundering data using regular expression is the I<only> mechanism for
untainting dirty data, unless you use the strategy detailed below to fork
a child of lesser privilege.

The example does not untaint C<$data> if C<use locale> is in effect,
because the characters matched by C<\w> are determined by the locale.
Perl considers that locale definitions are untrustworthy because they
contain data from outside the program.  If you are writing a
locale-aware program, and want to launder data with a regular expression
containing C<\w>, put C<no locale> ahead of the expression in the same
block.  See L<perllocale/SECURITY> for further discussion and examples.

=head2 Switches On the "#!" Line

When you make a script executable, in order to make it usable as a
command, the system will pass switches to perl from the script's #!
line.  Perl checks that any command line switches given to a setuid
(or setgid) script actually match the ones set on the #! line.  Some
Unix and Unix-like environments impose a one-switch limit on the #!
line, so you may need to use something like C<-wU> instead of C<-w -U>
under such systems.  (This issue should arise only in Unix or
Unix-like environments that support #! and setuid or setgid scripts.)

=head2 Taint mode and @INC

When the taint mode (C<-T>) is in effect, the "." directory is removed
from C<@INC>, and the environment variables C<PERL5LIB> and C<PERLLIB>
are ignored by Perl. You can still adjust C<@INC> from outside the
program by using the C<-I> command line option as explained in
L<perlrun>. The two environment variables are ignored because
they are obscured, and a user running a program could be unaware that
they are set, whereas the C<-I> option is clearly visible and
therefore permitted.

Another way to modify C<@INC> without modifying the program, is to use
the C<lib> pragma, e.g.:

  perl -Mlib=/foo program

The benefit of using C<-Mlib=/foo> over C<-I/foo>, is that the former
will automagically remove any duplicated directories, while the later
will not.

Note that if a tainted string is added to C<@INC>, the following
problem will be reported:

  Insecure dependency in require while running with -T switch

=head2 Cleaning Up Your Path

For "Insecure C<$ENV{PATH}>" messages, you need to set C<$ENV{'PATH'}> to
a known value, and each directory in the path must be absolute and
non-writable by others than its owner and group.  You may be surprised to
get this message even if the pathname to your executable is fully
qualified.  This is I<not> generated because you didn't supply a full path
to the program; instead, it's generated because you never set your PATH
environment variable, or you didn't set it to something that was safe.
Because Perl can't guarantee that the executable in question isn't itself
going to turn around and execute some other program that is dependent on
your PATH, it makes sure you set the PATH.

The PATH isn't the only environment variable which can cause problems.
Because some shells may use the variables IFS, CDPATH, ENV, and
BASH_ENV, Perl checks that those are either empty or untainted when
starting subprocesses. You may wish to add something like this to your
setid and taint-checking scripts.

    delete @ENV{qw(IFS CDPATH ENV BASH_ENV)};   # Make %ENV safer

It's also possible to get into trouble with other operations that don't
care whether they use tainted values.  Make judicious use of the file
tests in dealing with any user-supplied filenames.  When possible, do
opens and such B<after> properly dropping any special user (or group!)
privileges. Perl doesn't prevent you from opening tainted filenames for reading,
so be careful what you print out.  The tainting mechanism is intended to
prevent stupid mistakes, not to remove the need for thought.

Perl does not call the shell to expand wild cards when you pass C<system>
and C<exec> explicit parameter lists instead of strings with possible shell
wildcards in them.  Unfortunately, the C<open>, C<glob>, and
backtick functions provide no such alternate calling convention, so more
subterfuge will be required.

Perl provides a reasonably safe way to open a file or pipe from a setuid
or setgid program: just create a child process with reduced privilege who
does the dirty work for you.  First, fork a child using the special
C<open> syntax that connects the parent and child by a pipe.  Now the
child resets its ID set and any other per-process attributes, like
environment variables, umasks, current working directories, back to the
originals or known safe values.  Then the child process, which no longer
has any special permissions, does the C<open> or other system call.
Finally, the child passes the data it managed to access back to the
parent.  Because the file or pipe was opened in the child while running
under less privilege than the parent, it's not apt to be tricked into
doing something it shouldn't.

Here's a way to do backticks reasonably safely.  Notice how the C<exec> is
not called with a string that the shell could expand.  This is by far the
best way to call something that might be subjected to shell escapes: just
never call the shell at all.  

        use English '-no_match_vars';
        die "Can't fork: $!" unless defined($pid = open(KID, "-|"));
        if ($pid) {           # parent
            while (<KID>) {
                # do something
            }
            close KID;
        } else {
            my @temp     = ($EUID, $EGID);
            my $orig_uid = $UID;
            my $orig_gid = $GID;
            $EUID = $UID;
            $EGID = $GID;
            # Drop privileges
            $UID  = $orig_uid;
            $GID  = $orig_gid;
            # Make sure privs are really gone
            ($EUID, $EGID) = @temp;
            die "Can't drop privileges"
                unless $UID == $EUID  && $GID eq $EGID;
            $ENV{PATH} = "/bin:/usr/bin"; # Minimal PATH.
	    # Consider sanitizing the environment even more.
            exec 'myprog', 'arg1', 'arg2'
                or die "can't exec myprog: $!";
        }

A similar strategy would work for wildcard expansion via C<glob>, although
you can use C<readdir> instead.

Taint checking is most useful when although you trust yourself not to have
written a program to give away the farm, you don't necessarily trust those
who end up using it not to try to trick it into doing something bad.  This
is the kind of security checking that's useful for set-id programs and
programs launched on someone else's behalf, like CGI programs.

This is quite different, however, from not even trusting the writer of the
code not to try to do something evil.  That's the kind of trust needed
when someone hands you a program you've never seen before and says, "Here,
run this."  For that kind of safety, check out the Safe module,
included standard in the Perl distribution.  This module allows the
programmer to set up special compartments in which all system operations
are trapped and namespace access is carefully controlled.

=head2 Security Bugs

Beyond the obvious problems that stem from giving special privileges to
systems as flexible as scripts, on many versions of Unix, set-id scripts
are inherently insecure right from the start.  The problem is a race
condition in the kernel.  Between the time the kernel opens the file to
see which interpreter to run and when the (now-set-id) interpreter turns
around and reopens the file to interpret it, the file in question may have
changed, especially if you have symbolic links on your system.

Fortunately, sometimes this kernel "feature" can be disabled.
Unfortunately, there are two ways to disable it.  The system can simply
outlaw scripts with any set-id bit set, which doesn't help much.
Alternately, it can simply ignore the set-id bits on scripts.

However, if the kernel set-id script feature isn't disabled, Perl will
complain loudly that your set-id script is insecure.  You'll need to
either disable the kernel set-id script feature, or put a C wrapper around
the script.  A C wrapper is just a compiled program that does nothing
except call your Perl program.   Compiled programs are not subject to the
kernel bug that plagues set-id scripts.  Here's a simple wrapper, written
in C:

    #define REAL_PATH "/path/to/script"
    main(ac, av)
	char **av;
    {
	execv(REAL_PATH, av);
    }

Compile this wrapper into a binary executable and then make I<it> rather
than your script setuid or setgid.

In recent years, vendors have begun to supply systems free of this
inherent security bug.  On such systems, when the kernel passes the name
of the set-id script to open to the interpreter, rather than using a
pathname subject to meddling, it instead passes I</dev/fd/3>.  This is a
special file already opened on the script, so that there can be no race
condition for evil scripts to exploit.  On these systems, Perl should be
compiled with C<-DSETUID_SCRIPTS_ARE_SECURE_NOW>.  The F<Configure>
program that builds Perl tries to figure this out for itself, so you
should never have to specify this yourself.  Most modern releases of
SysVr4 and BSD 4.4 use this approach to avoid the kernel race condition.

=head2 Protecting Your Programs

There are a number of ways to hide the source to your Perl programs,
with varying levels of "security".

First of all, however, you I<can't> take away read permission, because
the source code has to be readable in order to be compiled and
interpreted.  (That doesn't mean that a CGI script's source is
readable by people on the web, though.)  So you have to leave the
permissions at the socially friendly 0755 level.  This lets 
people on your local system only see your source.

Some people mistakenly regard this as a security problem.  If your program does
insecure things, and relies on people not knowing how to exploit those
insecurities, it is not secure.  It is often possible for someone to
determine the insecure things and exploit them without viewing the
source.  Security through obscurity, the name for hiding your bugs
instead of fixing them, is little security indeed.

You can try using encryption via source filters (Filter::* from CPAN,
or Filter::Util::Call and Filter::Simple since Perl 5.8).
But crackers might be able to decrypt it.  You can try using the byte
code compiler and interpreter described below, but crackers might be
able to de-compile it.  You can try using the native-code compiler
described below, but crackers might be able to disassemble it.  These
pose varying degrees of difficulty to people wanting to get at your
code, but none can definitively conceal it (this is true of every
language, not just Perl).

If you're concerned about people profiting from your code, then the
bottom line is that nothing but a restrictive license will give you
legal security.  License your software and pepper it with threatening
statements like "This is unpublished proprietary software of XYZ Corp.
Your access to it does not give you permission to use it blah blah
blah."  You should see a lawyer to be sure your license's wording will
stand up in court.

=head2 Unicode

Unicode is a new and complex technology and one may easily overlook
certain security pitfalls.  See L<perluniintro> for an overview and
L<perlunicode> for details, and L<perlunicode/"Security Implications
of Unicode"> for security implications in particular.

=head2 Algorithmic Complexity Attacks

Certain internal algorithms used in the implementation of Perl can
be attacked by choosing the input carefully to consume large amounts
of either time or space or both.  This can lead into the so-called
I<Denial of Service> (DoS) attacks.

=over 4

=item *

Hash Function - the algorithm used to "order" hash elements has been
changed several times during the development of Perl, mainly to be
reasonably fast.  In Perl 5.8.1 also the security aspect was taken
into account.

In Perls before 5.8.1 one could rather easily generate data that as
hash keys would cause Perl to consume large amounts of time because
internal structure of hashes would badly degenerate.  In Perl 5.8.1
the hash function is randomly perturbed by a pseudorandom seed which
makes generating such naughty hash keys harder.
See L<perlrun/PERL_HASH_SEED> for more information.

In Perl 5.8.1 the random perturbation was done by default, but as of
5.8.2 it is only used on individual hashes if the internals detect the
insertion of pathological data. If one wants for some reason emulate the
old behaviour (and expose oneself to DoS attacks) one can set the
environment variable PERL_HASH_SEED to zero to disable the protection
(or any other integer to force a known perturbation, rather than random). 
One possible reason for wanting to emulate the old behaviour is that in the
new behaviour consecutive runs of Perl will order hash keys differently,
which may confuse some applications (like Data::Dumper: the outputs of two
different runs are no longer identical).

B<Perl has never guaranteed any ordering of the hash keys>, and the
ordering has already changed several times during the lifetime of
Perl 5.  Also, the ordering of hash keys has always been, and
continues to be, affected by the insertion order.

Also note that while the order of the hash elements might be
randomised, this "pseudoordering" should B<not> be used for
applications like shuffling a list randomly (use List::Util::shuffle()
for that, see L<List::Util>, a standard core module since Perl 5.8.0;
or the CPAN module Algorithm::Numerical::Shuffle), or for generating
permutations (use e.g. the CPAN modules Algorithm::Permute or
Algorithm::FastPermute), or for any cryptographic applications.

=item *

Regular expressions - Perl's regular expression engine is so called NFA
(Non-deterministic Finite Automaton), which among other things means that
it can rather easily consume large amounts of both time and space if the
regular expression may match in several ways.  Careful crafting of the
regular expressions can help but quite often there really isn't much
one can do (the book "Mastering Regular Expressions" is required
reading, see L<perlfaq2>).  Running out of space manifests itself by
Perl running out of memory.

=item *

Sorting - the quicksort algorithm used in Perls before 5.8.0 to
implement the sort() function is very easy to trick into misbehaving
so that it consumes a lot of time.  Starting from Perl 5.8.0 a different
sorting algorithm, mergesort, is used by default.  Mergesort cannot
misbehave on any input.

=back

See L<http://www.cs.rice.edu/~scrosby/hash/> for more information,
and any computer science textbook on algorithmic complexity.

=head1 SEE ALSO

L<perlrun> for its description of cleaning up environment variables.