On 19 August 2013 20:35, Chad <innocentkiller(a)gmail.com> wrote: Congrats to everyone for getting this going. Is there a workaround available for people behind the Great Firewall to log into projects in languages other than those that are exempted? If so, what is the best way for those individual users to contact Operations or whoever, outside of IRC? I'm fairly certain some of those users may not want to have to publicize their locations. I see mention of an email address: could that be created before the change please? (I'm guessing it might be some special user permission similar to IP-block exemption, but that may just be my non-technicalese speaking.) Risker

On 19 August 2013 21:09, Ryan Lane <rlane32(a)gmail.com> wrote: Okay, perhaps I wasn't clear. What I am referring to are editors from China or Iran who regularly log into projects that will be covered with HTTPS, as we know that HTTPS is (at least sometimes) blocked in those countries. Remember that you're including Commons, Meta, and all English projects - and yes, it is the right thing to do. But we do have a non-negligible number of users (including administrators and stewards) who will need to have a way to access these projects. Do you have a way to exempt them? Risker

On 19 August 2013 21:31, Ryan Lane <rlane32(a)gmail.com> wrote: Well, see. The problem is that the home projects for these editors are not Farsi or Chinese, they are English or German or Commons or Meta (and a few other languages too - and that is just the editors I personally have had contact with). So, is your recommendation to these editors that they log in using the Chinese or Farsi project and then traverse to their home wiki from there? Risker/Anne

On 19 August 2013 23:07, Chris Steipp <csteipp(a)wikimedia.org> wrote: Thanks, Chris. I recommend you add that to the applicable pages. Is it certain that they won't run into a problem when moving from Farsi/Chinese to an HTTPS project? Has this been tested and verified? That is, they won't suddenly find themselves logged out on Commons because they're not using HTTPS? Risker

On Mon, Aug 19, 2013 at 9:32 PM, MZMcBride <z(a)mzmcbride.com> wrote: Just to clarify a bit: this is a moot problem, because disabling HTTPS in your user preferences does *not* disable login over HTTPS. Once this is deployed, *all* logins (as in absolutely all of them on whatever projects it is enabled on) will be over HTTPS. *-- * *Tyler Romeo* Stevens Institute of Technology, Class of 2016 Major in Computer Science www.whizkidztech.com | tylerromeo(a)gmail.com

On Tue, Aug 20, 2013 at 10:34 AM, MZMcBride <z(a)mzmcbride.com> wrote: My point is that it doesn't matter what your user preference is. Whether it's false or true, you still have to log in over HTTPS. In other words, the user preference has no effect on your ability to use the site. *-- * *Tyler Romeo* Stevens Institute of Technology, Class of 2016 Major in Computer Science www.whizkidztech.com | tylerromeo(a)gmail.com

On Tue, Aug 20, 2013 at 1:12 PM, Greg Grossmeier <greg(a)wikimedia.org> wrote: Indeed, and in those case they just leave their HTTPS preference turned off (since the default is off anyway). *-- * *Tyler Romeo* Stevens Institute of Technology, Class of 2016 Major in Computer Science www.whizkidztech.com | tylerromeo(a)gmail.com

To clarify, the default value for this HTTPS option is false, meaning you have to explicitly turn it on in order to force HTTPS. In other words, the only functional change being made by this deployment is that *login* on certain projects will be over HTTPS. So for those who do not have HTTPS, they will have to log in through a project that does not have secure login enabled. And once they do log in, they should be fine thereafter. *-- * *Tyler Romeo* Stevens Institute of Technology, Class of 2016 Major in Computer Science www.whizkidztech.com | tylerromeo(a)gmail.com On Tue, Aug 20, 2013 at 1:54 PM, Risker <risker.wp(a)gmail.com> wrote:

On Tue, Aug 20, 2013 at 11:07 AM, James Alexander <jalexander(a)wikimedia.org>wrote;wrote: It is turned on by default when $wgSecureLogin is enabled. That will be the case come tomorrow, yes. Yes, this is correct. No, it will put you back on HTTPS as that was the default. You have to turn the preference off. Preferences are local, so unless the local preference has been set to false, you would end up on HTTPS. Correct. I'm beginning to think there's a disconnect between what we coded and what people expect. The preference is *on* by default which I think is what's going to cause problems. We can change defaults before tomorrow so I think we should all be clear. -Chad

On Tue, Aug 20, 2013 at 11:36 AM, Tyler Romeo <tylerromeo(a)gmail.com> wrote: Did you read the patch? If $wgSecureLogin is true, prefershttps is also true. This is core. -Chad

On Tue, Aug 20, 2013 at 11:31 AM, Chad <innocentkiller(a)gmail.com> wrote: Thanks Chad, that's a lot of help. Yeah, this seems to contradict what I thought Ryan was saying above and what I was under the impression for. The bad use case for here (as describe by Risker for example) is a mainland china user from zhWiki logging in (through http) but now not being able to visit enWiki logged in at all (because it will force them to https and https is blocked). I know Ryan used the term 'home wiki' some up in his emails. My interpretation when reading the thread was that it actually meant the wiki you were logging into (which I think is fine) and not the 'home' wiki that is marked in the CentralAuth interface (though I can't figure out where that's actually marked in the database...). If we are using that 'CentralAuth' definition we're going to have a lot of false negatives, a significant amount of people are marked off as their home being enWiki or somewhere else because it was the first place they created an account. We've never really used 'home wiki' to mean anything other then first account. James James Alexander Legal and Community Advocacy Wikimedia Foundation (415) 839-6885 x6716 @jamesofur

The vast majority of people we serve with Wikipedia and friends don't have accounts and don't log in, and won't be affected in any way by this change. IMO it's simply unacceptable to leak authentication tokens or account passwords in cleartext; allowing any form of login over HTTP is dinosaur behavior and we'd be crazy to let it continue, whether for "some sites" only or all. We should require HTTPS for all logins on all sites in all languages all the time. Note that there are plenty of projects producing and maintaining block-circumvention tools, which is what folks who are running afoul of government censorship should be looking into if they need to log into a blocked site, or read blocked pages. It's a bit out of scope for Wikimedia to fix China's internet, though it'd be nice if we gave some recommendations on tools. Or would that just get us more blocked? -- brion On Tue, Aug 20, 2013 at 12:46 PM, George William Herbert < george.herbert(a)gmail.com&gt; wrote:

Wikipedia was blocked ENTIRELY in China for years; people interested in *reading* as well as contributing used circumvention tools (VPNs etc) to more securely access the site, and just got generic errors if they didn't. This is an acceptable trade-off which we've allowed the Chinese government to make for us before, and here we're talking about a much smaller effect (on contributors only). Again, it's not our business to fix China. China has to fix China. -- brion On Tue, Aug 20, 2013 at 1:15 PM, George William Herbert < george.herbert(a)gmail.com&gt; wrote:

Brion Vibber wrote: I completely agree with China fixing China. But I think the Wikimedia community has to decide whether this trade-off is acceptable. I'm not sure it's a decision that sysadmins and developers can make alone. China, but not HTTPS. If we change all sites to require HTTPS for logged-in users, we'll certainly increase site security and enhance the user experience for most users, but is that worth losing every zh.wikipedia.org contributor who lives in China? Or do we expect anyone blocked from HTTPS to simply edit without an account? I think the concern here is that some projects may be decimated (in terms of number of contributors) if HTTPS is forced for all users. MZMcBride

On 20 aug. 2013, at 23:21, Bartosz Dziewoński &lt;matma.rex(a)gmail.com&gt; wrote: Can I just say that IF there is still this much discussion and confusion going on even at the level of the developers, that I feel really uncomfortable with this being deployed in the next 24hours. This all just feels way too rough. And it smells like this is gonna create yet another deploy shitstorm within the communities. I wouldn't like to be in the shoes of the liaisons and ambassadors tomorrow.... DJ

On Tue, Aug 20, 2013 at 3:57 PM, Tyler Romeo &lt;tylerromeo(a)gmail.com&gt; wrote: Time out... We do not have a lack of secure login. That was solved a long time ago (many years). I've been using it since about when it was made available first (secure. ...). This is going from "secure is default and available" to "you cannot access other than secure", unless you know a non-secure exempted wiki you can log in to first. The people with a firewall (national, corporate, whatever) that blocks HTTPS deserve some warning that something bad is going to happen to them, and that they can mititate that using (X), before it hits. Again - it is entirely reasonable to shift the stance towards all secure. This will affect some people (I don't know how many). They have not been warned and the workaround is not intuitive. It's not a normal or reasonable to affect some number of users like that with no warning. This will be the third attempt at deploying this in the past six months, so It was not clear to me that this would have that wide an effect, or I for one would have been saying something months ago. I said exactly how significant I feel it is immediately upon my understanding what the effects will be. I understand your frustration, but again, the impact on those users is (to me) a blocker bug. It being discovered and made visible this close to deploy time is unfortunate. We should (later) have a conversation about feature descriptions and notifications on the tech list so that discoveries like this aren't last minute. That does not affect that it should be a blocker bug. Those affected people deserve notification and information on the workarounds. That does not mean "don't roll this out" but "don't roll it out until it's adequately publicized long enough that nobody is surprised and unable to find the workaround". A week or two weeks of adequate notice should be fine. -george

how this all is going to work. So let's take a fairly simple and common use case, and work out how we're going to keep these users editing. The use case I suggest we work out is "English Wikipedia editor who lives in China and wants to edit English Wikipedia using his logged-in account." So: - Can this user log in on English Wikipedia? (based on this discussion, no) - Can this user log in on an HTTPS-exempt Wikipedia (based on this discussion, any Chinese or Farsi project, yes) - How will they know that they have to go there? - When they try to go from Chinese Wikipedia to English Wikipedia, what happens? - Do they get an HTTP page, or an HTTPS one? (the latter of which is not accessible) - Do they remain logged in as HTTP? or do they get logged out because the enwp page is HTTPS by default? - If they manage to get as far as English Wikipedia, they can change their user preference to HTTP once that is enabled. Will that also allow them to log in under HTTP so they don't have to go to another project? Now, I can personally think of a dozen editors to whom this *specific* user case applies, and I know at least another dozen to whom similar user cases apply in relation to other projects - and I really don't know that many people. I get the importance and value of this action plan. At the same time, there is no benefit to the movement to de facto block existing contributors or ghettoize new ones. Essentially banning Iranian Wikimedians from editing anything but Farsi projects is contrary to the core principles of the Wikimedia movement; deliberately creating a situation where editors need to break the law of their land to participate on Commons or Meta is a significant problem. It's bad enough when that is caused by external forces outside of WMF control; it's disturbing when it is being caused by WMF itself. Risker

On Tue, Aug 20, 2013 at 7:22 PM, Erik Moeller &lt;erik(a)wikimedia.org&gt; wrote: Excellent. Thank you. Yes; in the medium term (past the next few days / week) the direction here is clearly good from a user privacy / censorship and monitoring point of view. -- -george william herbert george.herbert(a)gmail.com

On Tue, Aug 20, 2013 at 8:04 PM, MZMcBride &lt;z(a)mzmcbride.com&gt; wrote: There's not a lot of great options for us to actively bypass censorship methods and anything we do will likely result in us being completely blocked by doing it. Maybe what we're doing is appeasement, but realistically we have no political power against China. The editors from mainland China had a discussion with some of us at Wikimania and they said that Wikipedia is basically unknown in China because Baidupedia is what shows up in the search results and Wikipedia does not. They actually spend a great deal of time trying to make Wikipedia known to readers in hopes of strengthening the editing community. If we were fully blocked again in China, it wouldn't cause any political fuss. - Ryan

<quote name="MZMcBride" date="2013-08-21" time="00:23:04 -0400"> We have some preliminary data that we collected over the weekend on the ability of users to access an https resource (zero byte gif hosted on upload.wikimedia). The numbers still only live in a google doc (sorry!), and they have a lot of caveats. The caveats (really important to read): https://docs.google.com/a/wikimedia.org/document/d/1Y2vs8lpevv9PtH_dp3P5hZe… (one really important caveat is we don't even list countries which had less than 100 requests total, as that would be too much noise in the data.) https://docs.google.com/a/wikimedia.org/spreadsheet/ccc?key=0Ams-fyukCIlMdF… (Be sure not to miss the tabs at the bottom which show you a map view of the data). As China and Iran are the outliers there, here's a spreadsheet with them zero'd out so you can more easily see the middle gradients: https://docs.google.com/a/wikimedia.org/spreadsheet/ccc?key=0Agte_lJNpi-OdF… Additionally, to see if any changes have a major effect on the ability of people to log in, we've started parsing out the successful centralauth autentications and will have a nice Ganglia graph tomorrow. We also parsed out some historical data on those going back a week or more to have a better idea of what "normal" is. Our numbers here are "successful logins per hour" which should be a decent metric to watch. Thanks much to Ori and Dario for working on the HTTPS capability numbers, and Ori and Robla for getting the successful login numbers. Greg -- | Greg Grossmeier GPG: B2FA 27B1 F7EB D327 6B8E | | identi.ca: @greg A18D 1138 8E47 FAC8 1C7D |

On 21 August 2013 00:51, Greg Grossmeier &lt;greg(a)wikimedia.org&gt; wrote: Thank you for sharing this data, Greg. I am surprised to see 5 additional countries with more than 10% failure rates, and another 11 with more than 5%, although these tend to also have higher than average margins of error. It would be interesting to see how the numbers stabilize over time. Risker

On Aug 20, 2013, at 9:43 PM, Greg Grossmeier &lt;greg(a)wikimedia.org&gt; wrote: Thanks, Greg. Is there any chance that monitoring could track success of login if someone is redirected from HTTP to HTTPS? The redirects should be easy to spot. -george Sent from Kangphone

If it was six months ago, I would suggest we hand over a unique random cookie with the redirect and verify on the HTTPS side that the cookie showed up, to make sure that it worked. And then only keep a success/fail log for IP block, perhaps, no user data. That would seem privacy neutral. Too late now to do that, though. Sent from Kangphone On Aug 20, 2013, at 10:24 PM, Greg Grossmeier &lt;greg(a)wikimedia.org&gt; wrote:

On 08/20/2013 11:05 PM, Risker wrote: I'm sorry Risker, but you've got this backwards. Making a long-overdue /minimal/ fix to our login process is not political advocacy. Compromising the security and privacy of our editors for the sake of a government's censorship policies, *is*. The very idea that editors with checkuser or oversight might even be *able* to login in cleartext over an Internet we *know* is monitored by entities that are demonstrably hostile to privacy is worrying enough on its own without introducing additional flaws in the process. I don't even agree that an exception should be made to allow cleartext logins from regions which are even *more* hostile to privacy than the United States; and I would have advocated that no account with bits should be allowed to do so regardless of location. Nevertheless, engineering has been bending over backwards to accommodate as many editors with crippled Internet access as is possible; inventing bogeymen and quoting misapplied bits of policy around ("promotional use"? Really?) is an extraordinary show of bad faith. -- Marc

On Tue, Aug 20, 2013 at 9:20 PM, Risker &lt;risker.wp(a)gmail.com&gt; wrote: No, but the point is that there's no apolitical choice here. Actively suppressing a standard, long overdue security measure in order to ensure that a country's censorship practices do not interfere with editing and access to Wikipedia is a political choice. Not doing so in full awareness of the consequences is a political choice. We cannot claim ignorance or neutrality either way. The real question is what political choices serve Wikimedia's mission best in the short and long term, and I agree that that's a discussion that extends beyond the technical dimension, so is somewhat OT here. Erik -- Erik Möller VP of Engineering and Product Development, Wikimedia Foundation

On Mon, Aug 19, 2013 at 5:35 PM, Chad &lt;innocentkiller(a)gmail.com&gt; wrote: Changes all staged for deployment, will probably merge right before we throw the switch Wednesday. https://gerrit.wikimedia.org/r/#/q/status:open+project:mediawiki/core+topic… -Chad