I'm not sure allowing plugins to modify an apps security policy is a good
idea because CSP only really works when the dev understands it and puts
thought into it.
A build step for CSP might be tricky because we don't actually know which
.html files might be navigated to (as opposed to XHR'ed for templates). It
could also be that some pages need different CSP than others.
So, with Ian's whitelist changes
- We disallow apps from navigating, openExteral, and XHR'ing by default
- If they want the <access> behaviour back, they can install the
legacy-whitelist plugin.
Question is, what do we want them to actually do?
Right now there's two new whitelist plugins:
- navigation-whitelist & intent-whitelist
- They look for <allow-navigation> and <allow-intent> respectively
- Neither of these actually open up all network access.
I'd like to propose that for simplicity, we have only one "new" whitelist
plugin that:
- Does what navigation-whitelist & intent-whitelist do
- Opens up all network requests on the native side
- Has JS that runs on start-up that alert()s if no CSP meta tag is
present.
- It should recommend adding in the CSP that is used in the default
app template as a start
This should cover 99% of use-cases (people shouldn't need to write their
own whitelist plugins), and (I hope) will be simple enough to figure out
without reading too much documentation.
On Fri, Feb 20, 2015 at 5:12 PM, Jason Chase <cha...@chromium.org> wrote:
> Chuck,
>
> Thanks for the feedback, it's good to know others are interested in CSP.
>
> I've created a doc to capture the proposal in a little more detail, and
> allow for more robust comments:
>
> https://docs.google.com/document/d/1sfFs6LB1_giodyR4QwBMQssLKP_UxACZifk-VYVX2T8/edit?usp=sharing
>
> In that doc, I've attempted to address the questions/comments both from
> your email, as well as Michal's earlier response. I'll let all interested
> parties continue the conversation in the doc.
>
> Thanks,
> Jason
>
> On 20 February 2015 at 10:54, Chuck Lantz <cla...@microsoft.com> wrote:
>
> > Hey Jason - Glad to see this proposal! A number of us at Microsoft have
> > been talking along these same lines actually. Windows 10 apps will
> include
> > CSP support as the latest version of IE has support so I'd say we're
> > completely in support of moving Cordova apps down this path. In fact I'd
> > want to make sure that any CSP related metadata tag injection also
> applied
> > to the Windows platform as well.
> >
> > A few of thoughts:
> >
> > 1. I definitely know there is quite a bit of interest still in being able
> > to enable hosted (https accessed and controlled by the developer) app
> > content access Cordova device APIs (which is currently a shortcoming of
> > Windows 8.0/8.1 apps so we hear about it quite a bit). As a result,
> we'll
> > want to be sure Cordova doesn't inhibit this use case at a base level.
> > That said, having a default CSP policy that restricts hosted in the
> > template is fine and would promote secure practices since you need to
> > exercise caution when mixing in any remote content even when you control
> it
> > completely. Also agree with inline being high risk.
> >
> > 2. Re: Long term, one thing that CSP doesn't cover well is which URIs
> > should be granted elevated device access. Given hosted content with
> plugin
> > device API access is still a scenario we'll need to consider, perhaps we
> > should consider using the config.xml <access> element to represent URIs
> > that have device API access (beyond standard browser access). Otherwise
> we
> > get into a bit of an "all or nothing" situation as it pertains to hosted
> > app content which poses a larger security risk if you opt to extend
> device
> > API access beyond local content. (It also strikes me this is a general
> gap
> > in the web standard as a whole.)
> >
> > 3. Eval is actually a bit tougher - I know when we've look at this in the
> > past it impacted JS frameworks far more than inline did. (Ex: With
> Angular
> > you can stop using eval but you take a perf hit which is a bigger deal on
> > mobile than desktop.) Definitely the most secure practice - but it also
> > could cause the default template to appear to "not work." If we omit the
> > "unsafe-eval" directive in the CSP policy in the template we'll want to
> be
> > crystal clear on how to alter it. That could be solved with proper
> > documentation and blog posts though.
> >
> > 4. I'd suggest we also consider the new "browser" platform here since
> > Chrome/Firefox/IE (as of Win 10) have support. Should be "free", but I'm
> > guessing the metadata tag injection you mention is something we could
> > probably just do all-up rather than only for specific platforms.
> >
> > -Chuck
> >
> > -----Original Message-----
> > From: mmo...@google.com [mailto:mmo...@google.com] On Behalf Of Michal
> > Mocny
> > Sent: Thursday, February 19, 2015 2:25 PM
> > To: dev
> > Subject: Re: Proposal for CSP support
> >
> > Thanks for this clear outline.
> >
> > Jason, I know you've been working on the short-term items for a while as
> > part of your investigation, fixing things as you went -- what is the
> > current state of CSP support in platforms / plugins? What portion
> already
> > has fixes (or PR for them), what work is known but undone, and what
> hasn't
> > been investigated much at all?
> >
> > -Michal
> >
> > On Thu, Feb 19, 2015 at 4:55 PM, Jason Chase <cha...@chromium.org>
> wrote:
> >
> > > I'm interested in full-blown support for CSP (Content Security Policy)
> > > in Cordova. While we're close to having new and improved whitelist
> > > functionality, there are gaps in what the whitelist is able to protect
> > > against. In particular, inline script and eval() are higher risks that
> > > are not addressed by whitelists.
> > >
> > > Many Cordova apps may use only static content, or not include any
> > > third-party content. However, there are certainly examples of apps
> > > that need to include user input/third-party content, mixed with the
> > > app's own HTML content. In some cases, platforms may even restrict
> > > functionality (see [1]). I think CSP is a compelling answer for these
> > > scenarios, and for security in general for apps.
> > >
> > > Assuming CSP support is valuable, the question is how to implement?
> > > Support for CSP is not universal across platforms. It is known to be
> > > supported on Android (KitKat and later), iOS (since 7.1), and Firefox.
> > > Where supported, it is typically via a HTTP response header, or a META
> > > tag in the document.
> > >
> > > I've done some investigation into feasible approaches. As a result,
> > > I'm proposing as below.
> > >
> > > Long term goal:
> > > Cordova supports CSP in apps *and* plugins, and is enabled/secure by
> > > default. Ideally, CSP rules can be configurable and automatically
> > > applied to all content (i.e. so developers can fall into the pit of
> > > success)
> > >
> > > Achieving this goal will likely require incremental progress over a
> > > number of releases. At a high level, first make changes so developers
> > > can manually apply CSP to their apps. Longer term, add support for
> > > configurability and automatic application of CSP.
> > >
> > > Short term plan:
> > > - Change new app template to contain CSP meta tag with a default,
> > > secure policy (i.e. no inline script, eval(), only local app content)
> > > - Remove any blockers to default policy from framework and core
> plugins.
> > > This would be a continuation of the work in CB-8210, applied to other
> > > platforms. For example, this would fix any framework code that relies
> > > on sending javascript to be executed inline, from the native side
> > > - Deprecate any framework APIs that allow less secure practices. Many
> > > already are marked as deprecated (at least on Android)
> > > - Update docs/samples to include CSP, and clearly state that use of
> > > inline javascript is deprecated
> > >
> > > Long term plan:
> > > - In a future major release, remove the previously deprecated
> > > framework APIs
> > > - Define/implement a configuration model for CSP rules
> > > - Implement a build/package step to apply configured CSP rules to all
> > > content as meta tags. Run-time support involves re-writing content,
> > > and/or intercepting resource requests. The feasibility of
> > > intercepting requests is highly variable across platforms, at greater
> > > cost/complexity than build-time.
> > >
> > > I'm very interested in any comments on this proposal. This includes
> > > questions around use cases (missing or otherwise), different
> > > requirements, technical concerns, .etc.
> > >
> > > Thanks,
> > > Jason
> > > Google Cordova Team
> > >
> > > [1] http://callback.markmail.org/thread/yxmmya2o2lc26tpi
> > >
> >
>
