bug#28350: enriched.el code execution

[Charles A. Roelli]

> Date: Sat, 09 Sep 2017 19:55:37 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> CC: 28350@debbugs.gnu.org
>
> > > > +See Info node `(elisp)Display Property' for the use of these
> > > > +display specifications."
> > > > +  (ignore-errors
> > > > +    (or (stringp prop)
> > >             ^^^^^^^^^^^^
> > > What about an image spec (including a slice spec)?
> > 
> > Okay, I see that image specs can be safe.  But are they all safe?
> 
> I think they are.  Does anyone know different?

I read over the documentation some more and they do look alright.

> > And I don't understand how a slice spec is used together with an image
> > spec.  Is the slice spec used inside of IMAGE-PROPS, i.e. as you might
> > gather from the manual:
> > 
> > ‘(image . IMAGE-PROPS)’
> >      This kind of display specification is an image descriptor (*note
> >      Images).  When used as a display specification, it means to
> >      display the image instead of the text that has the display
> >      specification.
> > 
> > ‘(slice X Y WIDTH HEIGHT)’
> >      This specification together with ‘image’ specifies a “slice” (a
> >      partial area) of the image to display. 
> > 
> > ?
> 
> AFAIU, like this:
> 
>   ((slice X Y WIDTH HEIGHT) (image . IMAGE-PROPS))
> 
> You can see examples of this in image.el and image-mode.el.

Thanks.  I forgot that the display property can be set to a list or
vector of display specifications.  I've updated the patch to reflect
this:

+        (and (seqp prop) (seq-every-p 'enriched-display-prop-safe-p prop)))))

and I've added slice/image specifications.

> > At this point it seems that unsafe display specs are more the
> > exception than the rule, so it might make sense to define the
> > `enriched-display-prop-safe-p' function by excluding the unsafe
> > specifications instead of including the safe ones.  What do you
> > think?
> 
> I'm not sure.  The display spec can be complex, so to make sure none
> of these exceptions sneak through, you will have to recursively unpack
> the spec data structure and examine each of the elements, which smells
> too similar to emulating 'eval'.  No?

Thank you.  I've kept the current approach.  Please see again the
attached patch.

Also, should the left-fringe/right-fringe display specifications be
considered safe?  They seem innocuous.

0001-Prevent-code-execution-by-text-enriched-files-Bug-28.patch
Description: Text Data