I think how Cordova works right now was the best way. Have access blocked by default, but have a <access origin="*"/> in the default template. It makes the setting visible, while still working out-of-the-box.
If we turned on requests when no whitelist plugin is installed, then existing apps that have <access> tags will have their whitelist removed with 4.0.0 and not know it. If someone updates and their app can't hit the network anymore, then I think Stack Overflow will tell them why pretty quickly. We should also be very clear in the release notes and upgrade guide. On Wed, Mar 4, 2015 at 7:54 PM, Nikhil Khandelwal <nikhi...@microsoft.com> wrote: > I like Ian's proposal of blocking network access only when a whitelist > plugin is added to do so and is choosing to override the default behavior. > > Scanning config.xml on upgrade might be a good way to warn devs to refer > them to use this plugin. These changes should also be documented in the > migration guide from Android 3.x to 4.0. > > Thanks, > Nikhil > > > -----Original Message----- > From: Jesse [mailto:purplecabb...@gmail.com] > Sent: Wednesday, March 4, 2015 11:05 AM > To: dev@cordova.apache.org > Subject: Re: Android's new Whitelist Plugins > > I like the defaults as discussed, regardless of how they are achieved. > ie. network yes, intents no > This is similar to how a plain webview works if you add it to a native app > on ios or android, at least the network part, not sure what the default > intent handling is. > > Are there portions of this functionality that make more sense as part of > the platform native code? To me a plugin that is installed by default is > just modular platform code. Is there ever a reason to NOT want this plugin, > versus just opening up access? > > > > > > > > > > @purplecabbage > risingj.com > > On Wed, Mar 4, 2015 at 9:37 AM, Michal Mocny <mmo...@chromium.org> wrote: > > > I've been working on adding support to just install the whitelist > > plugin by default, and to add the <access origin="*"> to the default app. > > > > Is that sufficient? I think we may still need to do what Ian suggests > > and prompt on upgrade (or prepare)? > > > > For downstreams, especially IDE based ones, they will need to make > > sure the plugin is added by default however they do that. > > > > On Wed, Mar 4, 2015 at 12:06 PM, Ian Clelland <iclell...@chromium.org> > > wrote: > > > > > On Tue, Mar 3, 2015 at 8:20 PM, Nikhil Khandelwal < > > nikhi...@microsoft.com> > > > wrote: > > > > > > > Here are my thoughts on the default behavior: > > > > - navigation should be disabled. > > > > - XHR & network request should be enabled. > > > > > > > > > > And application launch through intent URLs should also be disabled. > > > (IMO) > > > > > > That's not a bad default -- it enables CSP usage by default, which I > > think > > > is good. It also (I think) means we're giving up on suggesting that > > network > > > requests can be completely blocked by default, because that's > > > definitely not the case on Android. > > > > > > We can implement this within the new framework: there is the idea of > > > a 'default policy' that only comes into effect when no plugins take > > > responsibility for the whitelist. As soon as any plugin, though, > > > handles the shouldAllowRequest() call, for instance, the default > > > policy is no longer in effect, and it is a true whitelist > > > (block-by-default) > > > > > > My biggest concern with this is that developers are going to blindly > > update > > > to Cordova 4.0.0, and when their app *just works*, they are not > > > going to realize that they are actually less secure than before. > > > (Without a > > plugin, > > > we've opened up all network access) > > > > > > Idea -- maybe we can scan config.xml -- at run time, or on prepare, > > > or on upgrade -- and if we see any access tag other than <access > > > origin="*"> we can display a loud message, suggesting strongly that > > > they install an appropriate plugin. > > > > > > Ian > > > > > > > > > > > > > > The plugin name is fine. > > > > > > > > I'm not convinced about a user having to add this plugin to enable > > > network > > > > requests for Android/iOS. This default behavior should work with > > > > the platform and should not require a plugin. This inhibits users > > > > from > > > getting > > > > the ground running on a Cordova app. It breaks existing templates > > > > in > > IDEs > > > > and other downstream CLIs as well - as all of them need to include > > > > this plugin to have any network access work. > > > > > > > > Thanks, > > > > Nikhil > > > > > > > > > > > > -----Original Message----- > > > > From: mmo...@google.com [mailto:mmo...@google.com] On Behalf Of > > > > Michal Mocny > > > > Sent: Tuesday, March 3, 2015 11:22 AM > > > > To: dev > > > > Subject: Re: Android's new Whitelist Plugins > > > > > > > > I've filed a JIRA issue with my thoughts on how to approach this: > > > > https://issues.apache.org/jira/browse/CB-8597 > > > > > > > > On Tue, Mar 3, 2015 at 1:02 PM, Andrew Grieve > > > > <agri...@chromium.org> > > > > wrote: > > > > > > > > > Like your ideas a lot. Updating the project template makes a lot > > > > > of > > > > sense. > > > > > > > > > > Tried to make it clear in the README, so if any part was not > > > > > clear please fix it. But, the CSP tag is the more important bit, > > > > > since <access> can't actually block all requests. The only > > > > > reason to even leave <access> in there is to support pre-kitkat > > > > > webviews, where no CSP support exists. CSP is also used to set a > > > > > navigation whitelist > > for > > > > > subframes, which the native side is not able to do. > > > > > > > > > > On Tue, Mar 3, 2015 at 11:22 AM, Michal Mocny > > > > > <mmo...@chromium.org> > > > > wrote: > > > > > > > > > > > My thoughts: > > > > > > > > > > > > - The split between <allow-navigation>, <allow-intent>, and > > <access>: > > > > > Like > > > > > > it a lot. > > > > > > - I think the defaults *for the plugin* are very reasonable. > > > > > > However, we may want to provide a default set of tags for the > > > > > > hello world app. A > > > > > year > > > > > > or so ago we added a default access * whitelist and I think > > > > > > maybe > > we > > > > > should > > > > > > continue that. (on the other hand, I've gotten used to > > > > > > explicitly whitelisting every url as part of chrome packaged > > > > > > app development and its not so bad). > > > > > > - Additionally, that means this plugin should be installed > > > > > > by > > > > default. > > > > > > As we discussed this morning, with the new plugin --save > > > > > > functionality we could just add this to the helloworld > > > > > > config.xml, > > I > > > > think! > > > > > > - Do you really need a CSP meta tag *and* <access> declarations? > > > > Thats > > > > > > what the README.md implies, but I would assume CSP trumps? > > > > > > > > > > > > -Michal > > > > > > > > > > > > On Mon, Mar 2, 2015 at 9:38 PM, Andrew Grieve < > > agri...@chromium.org> > > > > > > wrote: > > > > > > > > > > > > > I've tried to explain it in the plugin's readme: > > > > > > > > > > > > > > https://github.com/apache/cordova-plugins/tree/master/url-po > > > > > > > licy > > > > > > > > > > > > > > Some points for discussion: > > > > > > > - What should the default behaviour be for the three > > > > > > > whitelists (what should happen if not whitelist plugin is > installed). > > > > > > > - right now it can't open external URLs > > > > > > > - and can't do XHRs to http(s) > > > > > > > - Is the plugin name decent ("url-policy"). We should make a > > > > > > > dedicated > > > > > > git > > > > > > > repo for it (as well as for legacy-whitelist plugin) > > > > > > > > > > > > > > > > > > > > > > > > > > > > > --------------------------------------------------------------------- > To unsubscribe, e-mail: dev-unsubscr...@cordova.apache.org > For additional commands, e-mail: dev-h...@cordova.apache.org >
