Re: [Captive-portals] Use Case: "Carrier Grade Captive Portal"

[David Bird <[e-mail]>]

Thanks for the explanation...

Concerning the variations,
1. Agreed
2. Indeed, this is the situation we are in today when networks take
measures to avoid captive portal detection. PvD just made that easier for
them to do. Also, it has been argued that in Android the "captive portal
check" is really a surrogate for "connectivity check".... So, would client
really *not* to any probing when told there is no portal?
3. It is a big assumption to think portals will not load when redirected to
by unknown networks... assuming it does load, and seemingly "works", the
PvD device will be captive when non-PvD devices notice *nothing* (no portal
or problems accessing).

I will add another,
4. The network has a captive portal, but it is broken (for whatever reason,
RADIUS server is down, for example). In this case the PvD device discovered
the portal, the user interacted with the portal, the user should be
on-line, but isn't. What happens here? The 'problem' could have been
temporary, so ideally we want the user to return to the portal... my guess
is that the client is still probing like today in *all* cases...

I will reiterate a concern I have generally with PvD controlling the device
from a web service which may or may not be within the NAS itself (probably
not, in fact, since that would require many TLS certificates). Which is,
there will be many 'broken' networks. PvD is saying one thing, but the NAS
thinks differently (maybe because it received a RADIUS CoA, or a local NAS
timeout or issue, or any number of implementation 'glitches').


On Mon, Jul 10, 2017 at 8:26 PM, Tommy Pauly <[e-mail]> wrote:

> To chime in on the problem of misconfigured networks, I can picture three
> variations of issues:
>
> 1. If the PvD captivity information is unavailable due to
> misconfiguration, I would argue that an implementation MUST fall back to
> any legacy mechanisms like today’s HTTP probes. This means that on
> misconfigured networks that advertised a PvD server, we would have a delay
> after failing to fetch the information.
> 2. If the PvD information is wrong and lies saying that there is no
> captivity, when in fact there is, that could be detected by clients by the
> redirects that happen with future connections. This is essentially the
> situation we’re in today in which a captive network whitelists hosts such
> as captive.apple.com, and the user is forced to manually browse to a site
> and get redirected. This case is unfortunate, and exists today (and likely
> with any solution).
> 3. If the PvD information is wrong and lies that that there is captivity
> when there isn’t any, I would assume that the portal site would fail to
> connect or load, and would be ignored or dismissed by the system. The
> system could also run explicit probes in this case.
>
> Between all three of these, there shouldn’t be any fundamental reason a
> device that is PvD-aware would fail to join a network that a legacy device
> was able to join. These cases may still involve probing, or waiting for
> connections to fail, but if we can hope that misconfiguration is not the
> norm (which must always be our hope), then we’re still benefiting most
> cases.
>
> As for cases in which you join a network and then captivity starts partway
> through after expiration, I think the PvD solution is very elegant: we
> would still do explicit PvD discovery, and be altered that there is an
> expiration time on the access from the moment the network is joined.
>
> Also, the configuration that’s accessed for the PvD captivity doesn’t need
> to be public—the information can be specific to a local network, and only
> needs to tell as much information as the network is willing to share for
> the device’s benefit.
>
> Thanks,
> Tommy
>
>
> On Jul 10, 2017, at 8:54 AM, Dave Dolson <[e-mail]> wrote:
>
> David,
> Is it fair to say your concerns are mainly about misconfigured networks?
> And this is the reason that devices will always be incented to probe
> regardless of any method of provisioning?
>
> -Dave
>
>
> *From:* David Bird [mailto:[e-mail] <[e-mail]>]
> *Sent:* Monday, July 10, 2017 9:39 AM
> *To:* Tommy Pauly
> *Cc:* Dave Dolson; Eric Vyncke (evyncke); [e-mail]
> *Subject:* Re: [Captive-portals] Use Case: "Carrier Grade Captive Portal"
>
> On Sat, Jul 8, 2017 at 6:14 PM, Tommy Pauly <[e-mail]> wrote:
> [snip]
>
> The idea with explicit PvD discovery is that it would, as a step, replace
> a separate captive portal detection strategy.
>
> My overall concern with discovery mechanisms that are specific to only
> captive portals is that this is an extra step that is performed potentially
> on every network association, that may have limited extensibility for
> non-captive use cases. Since the explicit PvD design promises a way to
> discover many properties beyond captivity, and is bootstrapped very early
> on in the network association, it should hopefully allow clients to avoid
> the extra probe.
>
>
>
> I have concerns with the PvD approach, as described.
>
> If a network was misconfigured to advertise a PvD that does have a
> (Internet based) HTTPS server with a JSON file on it describing a captive
> portal network, then devices utilizing the PvD information will *never* get
> on this network while devices not using the PvD information do. That could
> be very confusing to users and network administrators alike.
>
> If you have seen walled garden configurations for large networks, you will
> notice a lot about the network operator's marketing partners. Indeed, many
> walled gardens are much larger than the network really wants... sometimes
> they just need to make things work in the garden. My point here is that
> operators may not *want* to list out their walled garden configuration on a
> public JSON file...
>
> At the end of the day, I'd argue that the client *will always probe* --
> wether it means to or not... A networking using PvD could just advertise
> all networks routes are available so that the device connects only to get
> caught up in a captive portal redirect anyway... back to step 1 and captive
> portal detection..
>
> I'm also unclear how PvD would deal with scenarios where you might start
> out with internet connectivity (e.g. "MAC Authentication") then to have a
> captive portal return after a session timeout has occurred...
>
>
>
>
>
>
> Note: the same “captivePortal” key is also defined in section 5.3 as a
> Boolean. Should I consider this to be a defect in the draft, or am I
> missing something?
>
>
> The updated version of the draft (https://tools.ietf.org/html/
> draft-bruneau-intarea-provisioning-domains-01) leaves out the specific
> keys for captive portals, and discusses it more abstractly. That would be a
> good thing to nail down at the Prague meeting. If PvD detection is done
> generically on network association, then a boolean or some way to indicate
> that this is *not* a captive portal will allow the device to not perform
> extra probing. If there is a captive network, we should be able to get the
> page or instructions on how to get beyond captivity.
>
> Thanks,
> Tommy
>
>
>
>
> -Dave
>
>
>
> *From:* Eric Vyncke (evyncke) [mailto:[e-mail]
> <[e-mail]>]
> *Sent:* Sunday, June 25, 2017 8:27 PM
> *To:* Dave Dolson; [e-mail]
> *Cc:* David Bird
> *Subject:* Re: [Captive-portals] Use Case: "Carrier Grade Captive Portal"
>
> At least Erik Kline and myself are following the captive-portal list :-)
>
> And the more we think about it, PvD could really be useful and we, the PvD
> I-D authors, would be pleased to present at your WG
>
> -éric
>
>
> *From: *Captive-portals <[e-mail]> on behalf of
> Dave Dolson <[e-mail]>
> *Date: *Friday 23 June 2017 at 11:57
> *To: *"[e-mail]" <[e-mail]>
> *Cc: *David Bird <[e-mail]>
> *Subject: *Re: [Captive-portals] Use Case: "Carrier Grade Captive Portal"
>
> [resend with fewer recipients to avoid mailing list problems]
>
> To echo David’s request,
> > If the authors of the PvD concept (re-)present their I-D to the mailing
> list, and stick around for discussion, that would be helpful.
>
>
> *From:* David Bird [mailto:[e-mail] <[e-mail]>]
> *Sent:* Wednesday, June 14, 2017 9:36 AM
> *To:* Erik Kline
> *Cc:* Gunther Nitzsche; Mark Townsley; Heiko Folkerts; Martin Thomson;
> [e-mail]; Livingood, Jason; Herzig, Willi; Warren Kumari;
> Dave Dolson
> *Subject:* Re: [Captive-portals] Use Case: "Carrier Grade Captive Portal"
>
> On Sun, Jun 11, 2017 at 11:17 PM, Erik Kline <[e-mail]> wrote:
> I'm not sure we have enough input on whether 511 is useful or not.  There
> seemed to be some suggestion it would help, and some that it wouldn't.
> Perhaps one question we could ask is whether it's harmful?  And if we agree
> it's not harmful, is it worth developing some recommendations for its use?
>
>
> In of itself, I don't believe it is harmful. However, if vendors use it as
> a reason to continue to terminate TLS connection in order to deliver the
> 511, then perhaps it is a bit harmful - or at least misleading. As the
> world moves to TLS (and QUIC), I think the time for the 511 code has
> already passed, to some degree. That, combined with the fact you may still
> have browsers not handling that return code properly, I don't see the value
> for any vendor or venue to implement this.
>
>
>
> As for the ICMP unreachable option, I certainly don't think it would be
> harmful (with the extra URL bits removed for now).  Is that something we
> wish to progress?
>
>
>
> I will work on a new draft that is only the basics. The additional fields
> could always be add in their own draft as extensions.
>
>
>
> Given that we're probably looking at a portal detection method based on
> entirely new work, it seems to me we're free to look at new things like
> utilizing the PVD detection scheme (DNS queries for "provisioning domain
> names", followed by other interaction still TBD).  Have the portal
> implementors reviewed this and given consideration as to whether its
> useful?  (I think of the discovery of the portal and subsequent interaction
> with it as 2 separate processes conducted, obviously, in serial.)
>
>
>
> I believe there are several talking points here, as the PvD method seems
> to have several possible implementations.
>
> I think requiring Ipv6 to configure Ipv4 is weird (I believe that was one
> proposed method to convey configuration)
>
> Several points I made in the thread "Arguments against any Capport API"
> regarding a web service - detached from the NAS - controlling the
> UE/station I think are relevant.
>
> If the authors of the PvD concept (re-)present their I-D to the mailing
> list, and stick around for discussion, that would be helpful.
>
>
>
> Thoughts?
>
>
> _______________________________________________
> Captive-portals mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/captive-portals
>
>
>
> _______________________________________________
> Captive-portals mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/captive-portals
>
>
>
> _______________________________________________
> Captive-portals mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/captive-portals
>
>
>