On Thu, Feb 19, 2015 at 9:23 AM, David Bruant <bruant.d at gmail.com> wrote:

> Hi,
>
> Half a million times the following meta-exchange happened on es-discuss:
> - if an attacker modifies Object.prototype, then you're doomed in all
> sorts of ways
> - Don't let anyone modify it. Just do Object.freeze(Object.prototype)!
>
> I've done it on client-side projects with reasonable success. I've just
> tried on a Node project and lots of dependencies started throwing errors.
> (I imagine the difference is that in Node, it's easy to create projects
> with a big tree of dependencies which I haven't done too much on the client
> side).
>
> I tracked down a few of these errors and they all seem to relate to the
> override mistake [1].
> * In jsdom [2], trying to add a "constructor" property to an object fails
> because Object.prototype.constructor is configurable: false, writable: false
> * in tough-cookie [3] (which is a dependency of the popular 'request'
> module), trying to set Cookie.prototype.toString fails because
> Object.prototype.toString is configurable: false, writable: false
>
> Arguably, they could use Object.defineProperty, but they won't because
> it's less natural and it'd be absurd to try to fix npm. The
> Cookie.prototype.toString case is interesting. Of all the methods being
> added, only toString causes a problem. Using Object.defineProperty for this
> one would be an awkward inconsistency.
>
>
> So, we're in a state where no module needs to modify Object.prototype, but
> I cannot freeze it because the override mistake makes throw any script that
> tries to set a toString property to an object.
> Because of the override mistake, either I have to let Object.prototype
> mutable (depite no module needing it to be mutable) or freeze it first hand
> and not use popular modules like jsdom or request.
>
> It's obviously possible to replace all built-in props by accessors [4], of
> course, but this is a bit ridiculous.
>

It is indeed ridiculous. Not fixing this in the ES5 timeframe was my single
biggest failure and disappointment as a member of TC39.

For reference, Caja's implementation of the technique described in [4] is at

https://code.google.com/p/google-caja/source/browse/trunk/src/com/google/caja/ses/repairES5.js#278

As it states, our term for freezing an object so that it does not provoke
this problem is "tamper proofing".



> Can the override mistake be fixed? I imagine no web compat issues would
> occur since this change is about throwing less errors.
>

There was a time when some of the browsers did not suffer from the override
mistake, and in so doing, were technically out of conformance with the ES5
spec. During this window, it was clearly still web compatible to fix the
override mistake. It was during this window that I raised the issue and
argued that it be fixed. I brought this up in meetings several times and
never made any progress. Once all browsers suffered equally from the
override mistake, it was no longer *clearly* web compatible to fix it, so I
gave up and focussed on other things instead.

However, I agree with your suspicion that it actually would still be web
compatible to fix this. Unfortunately, the only way to find out at this
point is for a browser to try deploying without the override mistake. I
don't have much hope.

Instead, I suggest you promote tamper proofing to those audiences to which
you currently promote freezing. Though the need for it is indeed
ridiculous, it actually works rather well. We've been using it successfully
for many years now.




>
> David
>
> [1] http://wiki.ecmascript.org/doku.php?id=strawman:fixing_
> override_mistake
> [2] https://github.com/tmpvar/jsdom/blob/6c5fe5be8cd01e0b4e91fa96d02534
> 1aff1db291/lib/jsdom/utils.js#L65-L95
> [3] https://github.com/goinstant/tough-cookie/blob/
> c66bebadd634f4ff5d8a06519f9e0e4744986ab8/lib/cookie.js#L694
> [4] https://github.com/rwaldron/tc39-notes/blob/
> c61f48cea5f2339a1ec65ca89827c8cff170779b/es6/2012-07/july-
> 25.md#fix-override-mistake-aka-the-can-put-check
> _______________________________________________
> es-discuss mailing list
> es-discuss at mozilla.org
> https://mail.mozilla.org/listinfo/es-discuss
>



-- 
    Cheers,
    --MarkM
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.mozilla.org/pipermail/es-discuss/attachments/20150219/2bf650be/attachment.html>

On Thu, Feb 19, 2015 at 9:23 AM, David Bruant <bruant.d at gmail.com> wrote:

> It's obviously possible to replace all built-in props by accessors [4], of
> course, but this is a bit ridiculous.

It is indeed ridiculous. Not fixing this in the ES5 timeframe was my single
biggest failure and disappointment as a member of TC39.

For reference, Caja's implementation of the technique described in [4] is at

https://code.google.com/p/google-caja/source/browse/trunk/src/com/google/caja/ses/repairES5.js#278

As it states, our term for freezing an object so that it does not provoke
this problem is "tamper proofing".



> Can the override mistake be fixed? I imagine no web compat issues would
> occur since this change is about throwing less errors.

There was a time when some of the browsers did not suffer from the override
mistake, and in so doing, were technically out of conformance with the ES5
spec. During this window, it was clearly still web compatible to fix the
override mistake. It was during this window that I raised the issue and
argued that it be fixed. I brought this up in meetings several times and
never made any progress. Once all browsers suffered equally from the
override mistake, it was no longer *clearly* web compatible to fix it, so I
gave up and focused on other things instead.

However, I agree with your suspicion that it actually would still be web
compatible to fix this. Unfortunately, the only way to find out at this
point is for a browser to try deploying without the override mistake. I
don't have much hope.

Instead, I suggest you promote tamper proofing to those audiences to which
you currently promote freezing. Though the need for it is indeed
ridiculous, it actually works rather well. We've been using it successfully
for many years now.