Re: [discovery] Adding CORS to NSD API - proposal and issues

On Thu, Oct 3, 2013 at 11:11 PM, Jean-Claude Dufourd
<jean-claude.dufourd@telecom-paristech.fr> wrote:
>
> Le 3/10/13 14:51 , Rich Tibbett a écrit :
>>
>> On Thu, Oct 3, 2013 at 10:04 PM, Jean-Claude Dufourd <jean-claude.dufourd@telecom-paristech.fr> wrote:
>>>
>>> Le 3/10/13 13:22 , Rich Tibbett a écrit :
>>>>
>>>> - For SSDP, a <service> node contained within a UPnP Device Descriptor
>>>> File must provide a <corsEnabled> sub-element whose value must be set
>>>> to '1', 'yes', 'y' or 'true'. Otherwise, the <service> is said not to
>>>> support CORS and is therefore not accessible to web pages (except if
>>>> the service type is whitelisted by the user or user agent).
>>>>
>>> JCD: Does this proposal amount to mandating an NSD-specific extension of SSDP ?
>>> It sounds like it and if so, no go.
>>> NSD must fully support legacy SSDP devices/services.
>>
>>
>> Please read the full thread and provided references in http://lists.w3.org/Archives/Public/public-device-apis/2013Sep/0029.html
>
> JCD: Actually, I did the full thread. It is not because I did not chime in that I did not read every single message.
> I just did not have anything to say until I realized the consequences when reading your proposal.
>
>> In case you have done that, would you care to explain your position? Are you speaking on behalf of all device implementers here? (honest question).
>
> JCD: If NSD requires modifications to SSDP, it will work with 0 devices at the beginning. This is a sure way to have a standard that will not be used.
> Note: I had the exact same criticism of Claes' work on making WebIntents work with SSDP by extending SSDP. I am coherent on this subject.
>
> NSD has to have a mode which works with existing devices with reasonable usability and reasonable security and most features.
> Otherwise, I cannot sell it to my prospective customers. It will be impossible to "bootstrap" the NSD ecosystem.

The proposal includes a contingency for a 'network services whitelist'
for interacting with existing devices and services. That would be a
reasonable compromise given the security concerns raised.

This was all discussed on the previous thread.

>>>>
>>>> - For DNS-SD, a developer must add a 'corsEnabled' key to their DNS-SD
>>>> SRV record's corresponding DNS TXT record. This key would be
>>>> case-insensitive in line with the DNS-SD spec itself [2]. Otherwise,
>>>> the DNS-SD service is said not to support CORS and is therefore not
>>>> accessible to web pages (except if the service type is whitelisted by
>>>> the user or user agent).
>>>
>>> JCD: same question for Bonjour...
>>>
>>>>
>>>> - For DIAL, the discovery message response must contain a
>>>> 'Access-Control-Allow-Origin' HTTP header and the value of this header
>>>> must be '*'. Otherwise, the DIAL service is said not to support CORS
>>>> and is therefore not accessible to web pages (except if the DIAL
>>>> service type is whitelisted by the user or user agent).
>>>>
>>>> Each of the above mechanisms is currently a tentative proposal and
>>>> each warrants further discussion both on this list and in their
>>>> respective standardization groups.
>>>>
>>>> We should also offer an override mechanism for users and/or user
>>>> agents to create a network services whitelist - enabling access to
>>>> non-CORS-enabled networked services also. Implementation of such a
>>>> network services whitelist remains at an implementer's discretion.
>>>>
>>>> In case any of the conditions above occur, a network service could be
>>>> accessed from web pages via the the Network Service Discovery API
>>>> (subject to all the conditions therein). Web page themselves are not
>>>> privy to the CORS-detection procedures outlined above since the CORS
>>>> opt-in process is abstracted away from the API itself to the
>>>> respective discovery processes.
>>>>
>>>> ---
>>>>
>>>> Current Proposal Issues:
>>>>
>>>> The main problem with this approach is that dissonance has now been
>>>> introduced between a.) the indicating of support for CORS during the
>>>> discovery process and b.) _actual_ support for CORS in subsequent
>>>> service interactions. i.e. If a networked service indicates it
>>>> supports CORS during the discovery process and then subsequently fails
>>>> to provide 'Access-Control-Allow-Origin: *' in all subsequent HTTP
>>>> responses or the networked service doesn't implement the ability to
>>>> respond to CORS preflight requests correctly (among other potential
>>>> CORS-related pitfalls) then the process of communicating with a
>>>> networked service fails and the service is broken for all meaningful
>>>> purposes (I can't communicate with a discovered process from a web
>>>> page).
>>>>
>>>> One solution to this issue may be to require networked services to
>>>> opt-in to cross-site requests during their discovery processes (as
>>>> proposed above) but then for the user agent to 'simulate' CORS support
>>>> for that networked service's URL endpoint. This is similar to the way
>>>> the API is currently drafted, by adding service URLs to a URL
>>>> whitelist that requires the user agent to treat service URLs as if
>>>> they supported CORS without the service itself needing to support CORS
>>>> directly.
>>>
>>> JCD: I would apply this method of simulating CORS support on a device whitelist, with the CORS whitelisting being done by the client as part of the security check UI.
>>> If the user accepts that this web app has the right to discuss with e.g. this UPnP ContentDirectory, then CORS support could be simulated on this legacy ContentDirectory service.
>>> However, the CORS simulation would need to happen during a later exchange with the service, after discovery, so outside of the scope of NSD, no ?
>>>
>>
>> So you are suggesting no change to the specification?
>
> JCD: It may seem that my suggestion is to make no change, but actually my suggestion is to go with CORS whitelisting as part of the security check UI, to include the messaging into NSD, and thus, bring the CORS simulation back into scope.

This proposal gives no reason to implement CORS other than to display
that CORS is enabled to the user - where the CORS functionality is
defunct anyway since we can just use URL whitelisting (or 'CORS
simulation') at all times.

This makes no sense unless I've interpreted your proposal above
incorrectly. It could, of course, be brilliant and I missed it.

>>
>> Please fully read the background to this thread. I believe the proposed changes are inevitable in light of implementer feedback.
>
> JCD: Changes are inevitable.

>
> Your proposal would endanger NSD by making its bootstrapping very difficult.

See 'network services whitelist' discussion above.

> I hope we can find other changes that will not require starting an ecosystem from scratch.

One proposal you have mentioned elsewhere on this list is to implement
a structured UPnP Messaging API as a way to mitigate stated security
risks (http://lists.w3.org/Archives/Public/public-device-apis/2013Sep/0030.html).

Let's talk about that briefly since I cannot follow what else is being
requested in this thread.

I don't think a high-level messaging API would solve any security
issues as cleanly as it would seem. It simply defers problems to a
different level of the stack. We would need to create a node-based API
for messaging and implement multiple variations of service messaging
types - discarding innovation in service types or messaging formats
for the future. I think there's little appetite to enable or repeat
more DOM-manipulation-like APIs here. Most vulnerabilities present in
unstructured HTTP messaging could easily also occur in a structured
HTTP messaging API. Any API with a large surface area would also
increase the number of attack vectors available.

I've written about why a low-level unstructured approach, building on
top of existing web platform details such as XHR, WebSockets, etc, is
a good idea for the NSD API in
http://lists.w3.org/Archives/Public/public-device-apis/2013Aug/0004.html.
Also, http://extensiblewebmanifesto.org is a good primer for why we
want to implement low-level APIs for new features initially. Nothing
here rules out higher-level APIs down the line but that is not the
initial objective.

>
>>>> The new part in this process is that networked services need to opt-in
>>>> to cross-site requests during discovery (via the mechanisms above)
>>>> rather than the user agent providing that access for all networked
>>>> services by default (as per the previously specced approach). That
>>>> seems to meet the objectives of this exercise without introducing any
>>>> potential dissonance between the service discovery process and the
>>>> service communication process.
>>>>
>>>> Minor point: If we do agree to go with this idea instead then it may
>>>> make more sense to rename 'corsEnabled' to 'xsEnabled' in the
>>>> discovery processes above.
>>>>
>>>> So the process essentially becomes CORS-like (where services need to
>>>> opt-in to cross-site sharing) but the result is 'simulated CORS'
>>>> (where the user agent acts as if CORS is enabled for URLS belonging to
>>>> shared networked services).
>>>>
>>>> -------
>>>>
>>>> Alternative Proposal:
>>>>
>>>> During the current drafting an alternative proposal came up and it may
>>>> be worth mentioning here.
>>>>
>>>> Instead of introducing the dissonance discussed above we could instead
>>>> rely on issuing tentative preflight requests [3] to networked service
>>>> endpoint URLs once the network service discovery process has been
>>>> completed as normal (without services needing to opt-in at the
>>>> discovery stage). This approach has the benefit of actually ensuring
>>>> CORS is properly supported since we are feature-detecting CORS
>>>> directly on actual network service URLs rather than trusting a
>>>> corsEnabled directive during discovery that assues us that CORS is
>>>> supported on any corresponding network service URLs.
>>>>
>>>> ---
>>>>
>>>> Alternative Proposal Issues:
>>>>
>>>> The major problem with adopting this alternative approach is that the
>>>> root of a networked service endpoint URL is not always configured to
>>>> return 200 OK responses (some networked services may provide access
>>>> only in sub-directories or non-standard root locations which can
>>>> differ per network service type). CORS preflight requests abort if the
>>>> HTTP response is not 200 OK, in which case this approach would fail to
>>>> capture legitimate CORS-enabled networked services during this
>>>> process.
>>>>
>>>> ---
>>>>
>>>> So it seems there are a few things we need to think about further here
>>>> and all feedback is welcome on the best way for us to proceed.
>>>>
>>>> br/ Rich
>>>>
>>>> ---
>>>>
>>>> [1] http://lists.w3.org/Archives/Public/public-device-apis/2013Sep/0040.html
>>>>
>>>> [2] http://tools.ietf.org/html/rfc6763#section-6
>>>>
>>>> [3] http://www.w3.org/TR/cors/#preflight-request
>>>>

****

Please use plain-text and proper indentation for W3C mailing list
discussions. Your current email settings make it difficult/impossible
to discuss in any reasonably structured way.

see: http://wiki.whatwg.org/wiki/FAQ#Mailing_List and
http://www.w3.org/2008/webapps/wiki/WorkMode#Mail_List_Policy.2C_Usage.2C_Etiquette.2C_etc.

Received on Thursday, 3 October 2013 14:59:15 UTC