[REBOL] A small security hole REBOL, and a huge one!
From: brian:hawley:bigfoot at: 18-Aug-2000 23:46
I hate to be the bearer of bad tidings...
First, the small security hole:
I just found out that second returns the original
code block when applied to a function value. This
code block can then be changed, which changes the
function without recreating it and reassigning it.
This kind of change is invisible to the query and
protect functions. This means that a script can
add code to any of the mezzanine functions without
you being able to tell. I don't know any easy way
to protect the REBOL interpreter session from this
kind of attack.
I know that you should read untrusted code before
you execute it on your system, but the WWReb is
based on executing distributed code. The first and
second functions were changed before to return a
copy of the blocks when applied to object values.
Is there some reason that second and third applied
to function values returns the original? Memory
efficiency, perhaps? Avoidance of deep-copying
possibly cyclic structures?
Can we count on this behavior in the future, or is
it subject to change? It enables self-modifying
code, for example. If we could count on this trick
it would make things more interesting for advanced
REBOL programmers.
I don't see how adding modules will change this in
any way. I don't see how you could prohibit changes
in the inner blocks and parens referenced by the
code without too much overhead.
I guess the best way to secure REBOL is to use the
launch function and run the script in a separate,
secure interpreter environment. This brings me to
my next discovery though...
The big security hole: The new launch native!
Right now, the launch native launches a new REBOL
process with the command line arguments you pass
as a parameter. Command line arguments just like
those accepted on the command line of REBOL. This
command line can include REBOL options, including
--nowindow --quiet --secure none
, with no kind
of security warning!
Launch should be rewritten so that the command
line options normally accepted by REBOL are passed
as refinements to the launch native. The /secure
refinement then needs to have the same restrictions
that the secure native would have if called at the
same point in the code. Any command line options in
the argument any-string! should be ignored by the
new REBOL process.
This hole is such a major issue that I have Bcc'd
this message to Feedback. Fortunately the launch
native is only implemented in the experimental
releases. Until this is fixed, don't ever run any
untrusted code with an experimental REBOL if it
has a working launch native (only /View so far)!
Don't kill the messenger, please! :(
Brian Hawley