On 2013-09-05 12:21 PM, Kumar McMillan wrote:
On Sep 4, 2013, at 6:04 PM, Ehsan Akhgari <ehsan.akhg...@gmail.com> wrote:
On 2013-09-04 5:03 PM, Kumar McMillan wrote:
Hi all,
After gaining much experience using navigator.mozPay()
(https://wiki.mozilla.org/WebAPI/WebPayment) to implement Firefox OS payments
it feels like the API may not be useful as a generic API for the broader web.
I'd like to propose deprecating navigator.mozPay() and replacing it with
lower-level primitives that payment providers can use directly. Thoughts and
feedback on this?
The problem: mozPay() adds some features for carrier billing but does so by
imposing a rigid end to end transaction flow. Not all payment providers need to
adopt this end to end flow and many providers already have their own end to end
flow. Awkwardly, mozPay also has a whitelist of who can use it which means
Mozilla (or whoever builds B2G) has to grant access to each provider.
mozPay comes in two parts:
* Developers use navigator.mozPay() to conduct a transaction and disperse
digital goods within their app. This is roughly the equivalent of doing
window.open() and using postMessage to communicate with the new window.
* Whitelisted payment providers (such as Firefox Marketplace) host their custom
payment flow in an iframe controlled by the chrome layer of mozPay. That iframe
has a special global called window.mozPaymentProvider which is defined here:
https://wiki.mozilla.org/WebAPI/WebPaymentProvider
I'd like to propose deprecating mozPay and instead figuring out a way to
securely expose mozPaymentProvider since it exposes primitives that websites
can use to implement mobile payments themselves. This would be a big win for
Firefox OS: anyone on the web could process payments just the same as Mozilla
does.
What are the exact primitives we need to expose?
1. To optimize carrier billing authentication mozPaymentProvider provides an
observeSilentSms method. This allows the payment provider to exchange a
challenge/response via SMS to get an unspoofable link to the user's phone bill.
We could expose this to all web content but we couldn't allow arbitrary numbers
if it were publicly accessible. We would have to whitelist the short codes that
can be used in this function (a short code incurs no cost to the user; it is
paid for by the payment authority). We might also need to prevent DOS'ing a
short code? There won't be too many of these to whitelist. Mozilla could also
host a web service for the whitelist which would make it easy to maintain.
This covers the use case we have for Mozilla's carrier billing and it is a
somewhat typical pattern. FaceTime registration works via silent SMS:
http://theiphonewiki.com/wiki/FaceTime#FaceTime_Activation_.2F_Registration
(Note that some operators support HTTP header injection to link to a user's
phone bill. In that case, the web doesn't need a new API; you can use HTTP
redirects.)
Would that not tie us down to the payment providers who use one of the
white-listed numbers?
I suppose it would, yes. However, the current mozPay already ties us down to
white-listed payment providers, only more brutally. If we expose
observeSilentSms to all web content, we need to protect it from abuse somehow.
It could be abused by web content entering a number that charges the user money
repeatedly in the background without anyone knowing it. Maybe someone has
another idea for protecting it?
Fair enough. Perhaps a white-list of this kind is not very bad after all.
2. Payment providers need to know the user's network. Use cases: the provider
may need to enable/disable certain regions (e.g. when in pre-production) and
the provider needs to know which whitelisted short code to use for silent SMS.
This is currently available as MCC/MNC on mozPaymentProvider. What is the
danger of exposing this to all web content? Do we need to prompt the user like
geolocation?
Can't they just use the client's IP address to obtain that information?
I've been told that IP address is not reliable enough to detect region and
network. There seems to be proxying scenarios that create a lot of edge cases.
Anyway, the fact that IP addresses are already exposed on the web makes me
think that exposing MCC/MNC is not such a big deal.
Yeah, in addition to that the IP address is not useful if you're
roaming, or are connected to wifi, etc.
3. Since carrier billing is tied to SIMs the payment provider must know when a
SIM is/isn't present, when a new SIM is inserted, and when a SIM is removed.
Multiple SIMs should be supported too. Mozilla's payment provider persists
carrier billing identity in a cookie (to avoid repeat authentication) so we
need to know when to delete / recreate that cookie. Currently
mozPaymentProvider exposes an array of ICCs that can be used for this so we'd
want to expose this to all web content. A provider doesn't need to know the
*exact* ICCs, it could use obfuscated values. Can we safely expose these
data/events separately to web content?
Wouldn't that raise fingerprinting concerns? I guess that depends on what kind
of obfuscation you had in mind though.
The main use case here is detecting when one SIM is removed and another one
inserted. That's all we need to solve for payments -- there are probably a
couple different ways to do it safely without enabling fingerprinting. I also
thought of window.addEventListener('moz-sim-changed', ...) or something.
Can we just expose this information to content loaded in the trusted UI?
That's it! These are the only primitives in mozPaymentProvider that we need to
expose to web content if we want to deprecate mozPay().
Also, we've heard from partners that they want to access the Secure Element
present on UICC SIMs. This primitive will enable them to set up carrier billing
easier -- an API is in the works for NFC related needs.
I mentioned that mozPay is roughly the equivalent of window.open() and
postMessage. The way I envision apps doing payments is by including a
JavaScript file in their app built by the payment provider. For example, an app
would include a JS file to do Stripe payments (like they do already). If Stripe
were to support carrier billing in addition to credit cards, it would benefit
from access to the mozPaymentProvider data above.
Exposing low level payment functionality to the app (or any web content) is a
paradigm shift from mozPay. The *app* would be managing a new payment window
instead of how mozPay's innards currently manage the new window. However, this
is pretty standard practice on today's web (PayPal, Google Wallet, Stripe, etc,
all work by injecting javascript into web pages).
As a consequence of deprecating mozPay() Firefox Marketplace will need
something like window.open(url, {mozTrusted: true}). This flag could be
restricted to whitelisted domains; it would open the current trusted UI which
is used by Firefox Marketplace and Persona. This would *not* be meant as a fix
or evolution to the current trusted UI but as a way for Mozilla to transition
out of mozPay(). Making the trusted UI more flexible is a separate ongoing
discussion.
Why is mozTrusted needed?
When a developer calls navigator.mozPay(), the chrome layer opens a Trusted UI
which is a special window that makes it more secure to enter sensitive payment
details. You can see Gaia's interface for it here:
https://github.com/mozilla-b2g/gaia/blob/master/apps/system/js/trusted_ui.js If
we were to deprecate mozPay, web content would need a way to retain this
functionality. The Trusted UI is not very well understood (yet) so the easiest
way to retain this functionality would be to continue whitelisting Marketplace
and Persona as the only domains who are allowed to open a Trusted UI. My idea
was to let web content do window.open(url, {mozTrusted: true}) to open a
Trusted UI which decouples it from mozPay. Only whitelisted domains would be
allowed to do that; any other URLs would trigger an error. In the future, the
Trusted UI might evolve into something that all web content can use if we can
figure out how to trust content :)
I see, thanks! FWIW window.open already has a features parameter which
we can extend for this purpose.
On a separate but related note, Request Autocomplete
(http://www.chromium.org/developers/using-requestautocomplete) would add
additional benefit to users who make payments on mobile. For example, typing a
credit card on mobile is painful.
I never understood what the point of requestAutocomplete was.
The point of it is to enable users to autocomplete forms that they are asked to
fill out regularly. Imagine that you are using your mobile phone for the first
time and you want to pay for something on a web app. The app might ask you to
enter your credit card number, expiration, and CVV. This is cumbersome to type
on a mobile device. Now, imagine that another unrelated app asks you to pay for
something the next day. You have to enter you credit card number again! This
would repeat infinitely for all unique payment providers on the web that you
encounter. With request autocomplete, you would only need to enter your credit
card number once -- you could auto-complete it the next time. Everyone on the
web wins :)
Actually, the auto-complete part can still be done by the UA using the
autocomplete content attribute. When talking to Jonas about this last
week, he mentioned that the use case for requestAutocomplete is for the
browser to do smart things such as generate one-time CC numbers, etc.
It also allows the UA to prompt for a PIN code for instance which is
used to encrypt the stored CC number. These use cases are not evident
from the proposed spec at all, but I guess it makes sense.
It seems to me like Google is pushing this so that they can integrate Google
Wallet in Chrome's UI, but if that is not a requirement for us, then why
wouldn't we just let the UA provide the autocomplete service as it does today?
In the context of Firefox OS, request autocomplete has nothing to do with
Google Wallet. It would improve the user experience around payments and other
cumbersome form entry.
I think if we decide to adopt the one-time CC number generation, that
part is done through Google Wallet (or perhaps other similar services).
But we don't necessarily need to figure out that part for now.
Cheers,
Ehsan
_______________________________________________
dev-webapps mailing list
dev-webapps@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-webapps
