addslashes isn't sufficient to protect literal strings in embedded JavaScript code

[Ned Batchelder <ned@…>]

When creating literal strings in embedded JavaScript code, the addslashes filter is used to escape characters significant to JavaScript:

<script>
var x = "{{x|addslashes}}";
blah(x);
</script>

But if the variable x includes the string "</script>", then this script block is ended too early, and the page is broken.

Attached is a patch that also escapes the </ sequence to ensure that this can't happen.

[ned@…]

The patch!

[Collin Grady <cgrady@…>]

But should this really be in addslashes? It's a pretty specific edge-case, and may not be desirable anywhere else - e.g. non-Javascript strings.

Seems like a custom filter made to escape </script> would be better, IMO :)

[Simon G. <dev@…>]

It does sound like a corner case, but isn't escaping things for javascript the primary usage for addslashes?

[anonymous]

It is a corner case, but there's no harm in escaping </ even in non-javascript settings. Given the current difficulty of making Django apps XSS-safe, I'd think we'd want to take an advantage we could.

The ticket has been marked needs_better_patch. Is there some action I should be taking?

[ned@…]

(edited, with attribution) It is a corner case, but there's no harm in escaping </ even in non-javascript settings. Given the current difficulty of making Django apps XSS-safe, I'd think we'd want to take any advantage we could.

The ticket has been marked needs_better_patch. Is there some action I should be taking?

[Chris Beaven]

There potentially could be harm in escaping </ in a non-javascript setting where your escaped \/ doesn't resolve to /.

However, I agree that XSS-safety should be a priority. Perhaps a specific filter for javascript "escaping"?

[Jeremy Dunck <jdunck@…>]

I'm attaching another patch which escapes all things not allowed in a JS string literal per ECMA 262 7.8.4 and also incorporates Ned's change for </script>.

Please do consider changing addslashes to include this or consider adding another filter to handle all embedded strings in JS.

[max.rabkin@…]

If this change is not made to addslashes, a warning should be added to its documentation to the effect that it is insufficient to protect strings in javascript.

[Jack Moffitt <metajack@…>]

I just ran into this bug today (specifically not escaping \n) and this patch works great. Would love to see this in the main tree.

[Jeremy Dunck]

Unflagging "Patch needs improvement" since I attached a new patch quite a while ago, and the ticket has had no action on it since then.

[durdinator]

addslashes as it stands is useful for things other than javascript, which may not accept \/, \n; csv for example (see ​http://www.djangoproject.com/documentation/outputting_csv/ for a prominent example).

I'm attaching a patch which adds this code as a new filter "escapejs", along with docs and tests.

Also, I question the removal in Jeremy's patch of carriage returns, so they're being converted to \r.

[Chris Beaven]

Looks good, Andy.

Small change required to documentation: *form feed* ``\f`` to ``\f`` should read *form feed* to ``\f``

Perhaps you could bring this up in django-dev to get a design decision.

[Jeremy Dunck]

Sweet, Andy.

FWIW, ECMA 262 7.8.4, to paraphrase, says you can't have a LineTerminator inside a string:

SourceCharacter but not double-quote " or backslash \ or LineTerminator \ EscapeSequence

I'm not sure what to make of that last "\ EscapeSequence".

LineTerminator is defined in 7.3 as LF, CR, LS, PS:

\u000A Line Feed <LF> 
\u000D Carriage Return <CR> 
\u2028 Line separator <LS> 
\u2029 Paragraph separator <PS> 

In any case, I agree it's a bit ambiguous, and probably isn't a common issue. I'd much rather see escapejs go in as-is than debate whether it's completely correct according to a mis-worded spec. At the end of the day, this is about not breaking web pages and avoiding security problems, which I'm sure the patch does better than current trunk. :)

Will you bring it up on django-dev, or should I?

[Collin Grady <cgrady@…>]

Isn't that talking about the literal characters though, not the escapes?

So the literal character represented by \u000D or \n wouldn't be allowed, yet '\n' is?

[Jeremy Dunck]

Oh. Hah. Yeah, Andy's right, that's just a bug in my patch. Andy's is correct. I just read his comment and mistook his meaning.

[durdinator]

escapejs with tests and docs (2nd attempt)

[durdinator]

@SmileyChris: Patch updated, docs fixed.

@jdunck: Can you bring it up on dj-d? I was just trying my hand at a bit of triaging on a handful of tickets, and saw that this one could do with a quick (and easy) tidy.

[Jeremy Dunck]

Thread here:
​http://groups.google.com/group/django-developers/browse_thread/thread/f0fb763a94c32e22

[Jacob]

(In [6892]) Fixed #4131: added an "escapejs" filter for use in JavaScript strings, and updated the documentation on addslashes to point to the new ticket. Featuring contributions from Ned Batchelder, Jeremy Dunck, and Andy Durdin.