Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
WPA3 Wifi Security Protocol Announced (wi-fi.org)
146 points by melzarei on Jan 9, 2018 | hide | past | favorite | 51 comments


A critical problem I see that needs solving is the Starbucks style open wifi login pages that fail to redirect when trying to visit HTTPS sites.

A new secure authentication system is needed for these set ups.

I’m blanking on the name of them but clearly they are widely deployed and likely causing lots of regular people to have trouble connecting to pages because they tried to connect to Facebook instead of an HTTP site.


Captive portals? Browsers and some mobile devices have a workaround where on new wifi connections they try to connect to a particular url and if that gives unexpected content they know it's a captive portal and prompt you to log in. But ideally this would be in the network descriptor somehow.


I wish gogo worked with this. I've ended up bookmarking http://captive.apple.com just to trigger their login page.



I use perdu.com to test my connection and for captive portals.

The website is fast, light (to say the least), always up, the URL is short and there is no valid HTTPS version so no need to type http://.

It also makes people around smile (and ask what this website is while I'm testing stuff and not thinking about it). English translation:

    Lost on the Internet?
    Don't panic, we are going to help you

        * <----- you are here
It is amazing how this joke has been there since 1998, without a change.

In 2016, perdu.com was displayed 1-2 million times per day and his author was spending €65 per year for this (source: https://www.nouvelobs.com/rue89/rue89-internet/20160603.RUE3...).


That's a cute page, I personally use cbc.ca here in Canada because it's only 5 letters and always HTTP only.


You can use http://neverssl.com for that purpose


I have had this problem as well, and it became increasingly difficult to find non-https domains so I started using one of my own sub-domain for it (I needed it on the dutch railway system). But recently, Ubuntu added a captive portals catcher and gives a small browser window to login (I think it tries to contact an Ubuntu sub-domain first to test the connection, you can turn it off if you want). This is very handy!


Why is this a problem? Just mash some keys and add a .com to the end. The portal doesn't check if the domain is real or not, it just intercepts all DNS requests until you've logged in.


and for the rare cases of those that don't, you can tunnel data over dns


Heh, I've been doing the same thing for wifi in Arriva and NS trains. I haven't gotten around to making my own website HTTPS yet, so I just use that.


RFC 7710 solves this via DHCP. What's missing is support both on operating systems and captive portal devices.


>A critical problem I see that needs solving is the >Starbucks style open wifi login pages that fail to redirect when trying to visit HTTPS sites.

The "fix" is to have automatic mechanisms that query a known-http page and detect a redirect. Which is what browsers already do. If you want to build your own system or don't want the browser to do the magic use neverssl.com.

> A new secure authentication system is needed for these set ups.

What would that mean to have "secure authentication" here? What are the security expectations of a "click here to get internet" system?

IMHO the best solution for captive portals would be to just not do them.


I would keep a wary eye open until actual specifications are announced and publicly available.

If history is any guide, some parts of it might become security mis-features, like Wi-Fi Protected Setup (WPS) which became a security issue because it allows WPS PIN recovery.

Also note that the IEEE controls the actual wireless specification as IEEE 802.11 and also specifies the actual encryption standards used during wireless transmission and reception (this used to be called IEEE 802.11i-2004 [1]).

My educated guess is that WPA3 won't change the block cyphers used to encrypt data (AES-CCMP) and the initial handshaking protocol as that is part of the IEEE Spec and cannot be changed by the Wi-Fi Alliance. Instead, it may specify additional requirements that clients have to fulfill before connecting (like public key certificates and proof of identities).

[1] https://en.wikipedia.org/wiki/IEEE_802.11i


Where can I actually read the spec? I'm not even much of a cryptographer, but I'm curious --- and the closed attitude is a contributing factor to things like this:

https://blog.cryptographyengineering.com/2017/10/16/falling-...


This is just the announcement that they're _going_ to build the spec XD


Is WPA3 going to require new hardware (wifi client cards/chips, wifi routers) or will it be possible to add WPA3 support via a software update in common operating systems?


One feature I wish to see is some form of AP authentication, maybe with TOFU and some PK pinning. That is currently, name an AP the same as another and a machine that had the AP name saved will try to connect to it. If you name an AP "Apple Store" or "Starbucks" you can watch devices connect (with their real MAC even, defeating MAC randomization used when scanning for APs), and if it's an open AP or you got hold of the passphrase, monitor and possibly MITM connections.


We definitely need some kind of mechanism to authenticate APs, right now we only have the SSID and MAC, both of which are trivial to spoof.

There isn't even any collision detection, nothing stopping you from giving your AP the same SSID as an already present AP, and clients will helpfully connect to whichever one has the strongest signal, which is probably going to be the rogue AP right next to you, rather than the actual Starbucks AP at the other end of the store. According to 802.11, APs with identical names are considered to be part of the same network, no questions asked.

I consider it a fundamental flaw of wifi, alongside the boneheadedness of WEP ("wired equivalent protection"... Yeah right, pull the other one), and the idiotic layout of channels in the 2.4GHz spectrum.

The only way to use wifi as a consumer is with WPA2 (AES only, no TKIP) a randomly generated strong passkey, and all smart features turned way off (security holes such as WPS). For any AP that you haven't personally set up and vetted, and especially open public APs, a VPN connection is absolutely mandatory. And I still wouldn't trust it 100%.


There have been proposals to use IEEE 802.11u for access points to signal that they allow EAP-TLS using only server-side authentication. [1]

It's more of an implementation and UI problem. The required standards already exist, no need to wait for WPA3.

[1] https://en.wikipedia.org/wiki/IEEE_802.11u#EAP-TLS

[2] https://web.archive.org/web/20131126183610/http://riosec.com...


> That is currently, name an AP the same as another and a machine that had the AP name saved will try to connect to it.

That's on purpose. At work, for instance, we have two APs with the same name, connected to the same wired network. A machine which originally was connected to one of them is supposed to try to connect to the other one when its signal is stronger. Break that mechanism, and you'll break all wifi networks larger than a few rooms.


I know this is on purpose and I use it myself to great extent, but provisioning each legit AP with an asymmetric key pair to be checked by the client proves this is one of the APs you intend to connect to, while still allowing for the roaming functionality.

Public areas could even physically display the AP's key fingerprint alongside the SSID (+ WPA key when not open).


> Another feature will strengthen user privacy in open networks through individualized data encryption.

WPA1 can arguably be forgiven a lot of shortcomings given its circumstances, but I never understood why this wasn't in WPA2. I understand that the station can't authenticate the AP in this case, but it still seems like it's strictly better for the traffic to be encrypted. What am I missing?


The same reason browsers don't allow self-signed certificates, I guess. Those security people think a false sense of security is worse than no security.


The false sense of security is what WPA2 gives anyway. Coffee house has a password on their wifi? Great, then only someone with the password can impersonate the AP, which is only... everyone.

You can't tell everybody something and still use it as a secret.

What they should be doing is putting the public key of the AP on the wall as a QR code.

Or the real solution, which is for everything to be end-to-end encrypted regardless.


> What they should be doing is putting the public key of the AP on the wall as a QR code

Except if the attacker has his own QR code stickers.


If it’s inside a coffee shop or other physical locations the attempt is likely to be caught. Or, put the code inside a photo frame making it easier for staff to distinguish.


> I never understood why this wasn't in WPA2

Perhaps they thought that this would be taken care of at a higher protocol layer. ISTR that things like IPSEC opportunistic encryption was hyped a lot at the time.


You can have per-device stream encryption with WPA2, just use 802.1x. If you accept any user/password you'll have a public wlan where people can't eavesdrop.


WPA-PSK already does this (per-device stream encryption). The per-stream encryption is specified during the four way handshake via the ANonce and SNonce values [1].

However, this is only true if the handshake is not intercepted or snooped on by a third party. If it is, then the attacker knows enough to decrypt your data.

802.1x provides extra protection by ensuring the initial PSK is not know to the attacker, making the initial snoop of the handshake ineffective.

[1] https://en.wikipedia.org/wiki/IEEE_802.11i


Now if only I could run `apt-get update && apt-get upgrade` on my router's firmware to support this.


Fine, but in addition everything you do should be encrypted end to end as well as you can. Use httpseverywhere. Use privacy badger.


Privacy Badger forces you to send the DNT header, making your browser far more vulnerable to fingerprinting. Just use the relevant filter lists in uBlock Origin.


How does DNT make you more vulnerable to fingerprinting? Because there's fewer people that DNT than who don't, or am I missing something else?


Yes, because it's an option you specifically have to set as a user, most people don't bother (or don't know about it), so simply by setting the flag, you're making yourself easier to identify.


You can disable the DNT check, and as far as I can tell, I am not sending a DNT flag at all, despite having Privacy Badger installed.


How does one do this? I found no relevant setting in privacy badger's settings, and even though dnt is disabled in my browser's settings, I'm still sending the dnt header with privacy badger on (after disabling Privacy Badger my browser no longer sends it, so it's definitely Privacy Badger).


Disable the option "Check if sites comply with EFF's Do Not Track policy" in Privacy Badger's settings, that should do it.

I'm on Firefox 57, also using uBlock Origin with every tracking-related filter list (I presume that may make a difference).


According to this page, I have DNT on: https://www.mozilla.org/en-US/firefox/dnt/ (How else can I confirm?)

The settings say that DNT is either always on or on when using tracking protection. So do you have tracking protection off and rely on uBlock instead? (I don't use privacy badger.)


According to that site, I also have DNT enabled, however https://amiunique.org/fp says it is off ("NC" == not set).

I have tracking protection in Firefox set to off, as it uses the same Disconnect.me lists that you can enable in uBlock Origin. I guess Privacy Badger just forces through DNT, as per EFF's stance on the matter. I think it's somewhat counterproductive.

Nonetheless, DNT on/off is absolutely secondary to simply blocking all tracking through uBO and Privacy Badger.


Either NC doesn't mean what you think it means or the website is wrong. It says NC for me too, but I can clearly see the header being sent in both dev tools and in wireshark.


Disabling the option doesn't change a thing for me. DNT is still sent.


Can anyone comment on the efficacy of the CNSA algorithm they are planning to use?


CNSA isn't an algorithm. It's the NSA's update to what used to be called "Suite B". It's a set of algorithms that NSA recommends for use in National Security Systems (classified systems that use unclassified algorithms).

Currently, that's AES-256, ECDH over P-384, ECDSA over P-384, SHA-384, and then also RSA/DH with a >3072 modulus.


So is WPA3 defective by design, then? I sure don't trust modern NSA-designed algorithms.


These are publicly designed algorithms that are widely used. It just so happens that the NSA also recommends then.


P-256 and P-384 have some "NSA" stigma associated with them though.


Which of the above do you think are "NSA-designed algorithms"?

The real NSA algorithms are part of "Suite A", which you don't have to worry about trusting because they aren't public. (I've chosen not to use things like the NIST curves for ECDSA, though.)


Technically: SHA-2 and P-curve ECDH.


Thanks, it was mostly a rhetorical question for the user who seems to think that the NSA designed all of them. (For some reason, I've met a lot of people who think NSA designed AES, for example.)


Just in time for CES!




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: