[tor-project] Cross-team OONI and Tor Browser project: default bridge checking

[Arturo Filastò]

On 22/11/2016 17:14 +0000, isabela wrote:

> On 11/22/16 10:10, Georg Koppen wrote:
>> Roger Dingledine:
>>> Ah ha, right, your experiments with measuring blocking of default bridges
>>> totally tie into this one.
>>>
>>> I think the things that I want beyond what you want are:
>>>
>>> * Actually seeing if Tor works using the obfs bridge. In this case,
>>> the bridge address is accepting the TCP connection and then immediately
>>> hanging up -- it's like there's some proxy in front of the obfsproxy,
>>> so *something* answers, but it sure isn't obfsproxy. The goal here would
>>> be to learn if it's broken, rather than to learn if it's being interfered
>>> with, so maybe that will make measurement less complicated.

Yes I agree this would be useful, though we think the first step would
be to just do a simple TCP connection scan. We are actually going to be
pushing this test to all users of ooniprobe and we currently don't list
obfsproxy (or obfs4proxy) as a hard dependency of ooniprobe which would
be a requirement to have the full test run there.

We can change that in future version, but we should first do the simple
thing so we can get it out soon and start being useful, rather than wait
more to have something more fully featured.

This is part of the following ticket:
https://github.com/TheTorProject/ooni-probe/issues/614

There are pending pull requests for adding testing via a simple TCP
connect and we will be rolling this out as part of the 2.1.0 release due
and the end of this month.

The relevant pull requests are:
https://github.com/TheTorProject/ooni-probe/pull/682
https://github.com/OpenObservatory/ooni-resources/pull/1

Question for Tor Browser team:
This is the list of bridges we are going to be testing as part of this
first iteration:
https://github.com/OpenObservatory/ooni-resources/pull/1/files#diff-4534a6062fbf50f7e59b953fb772d809

This list is based on what David gave us to test and I think it includes
most bridges from the current TBB (though it doesn't seem to include all
obfs3 bridges) plus some other bridges that I believe you plan to start
using in the future.

Is there something else you would like to have tested and added to this
list?
>>> * I want the Tor Browser team to get involved with knowing how the tests
>>> work, knowing that they're happening, and having some way of noticing
>>> when the tests say that something has broken. Otherwise there will be
>>> knowledge inside some OONI data set somewhere that says a bridge stopped
>>> working, but we won't have anybody in place to close the loop and *do*
>>> something about it.
>> I am fine with getting to know how those tests work and that they are
>> happening. I think even getting a notice when something is broken is a
>> good thing. But I don't think we (Tor Browser team) should be the ones
>> hunting down folks to fix their bridges. Ideally the tests would show
>> the bridge is not working anymore and the operator gets directly a
>> notification about it. There should be no need to put the Tor Browser
>> team or a member of any other team in the middle.
>>
Yes I agree that it would be great if we had some way of notifying
bridge users directly, however we currently don't have support for
notifications inside of our data processing pipeline.

It is something we have planned, but it will not be available immediately.

I guess we should understand what should be done based on these results
once the first iteration is done.

>>> Though actually, there's some room for us to realize that it should
>>> instead be the Network team that steps in here. Or perhaps for both the
>>> Tor Browser team and the Network team to say that this topic isn't in
>>> their area. That outcome would be great to recognize and tackle too,
>>> since it needs to be in *somebody's* area.
>> Well, OONI seems to be in a good position to do the checking, if we want
>> to do that.[1] The question is what happens afterwards, in case a bridge
>> is down? As I said above notifying folks should not involve any other
>> team than the one that did the measurement.
>>
>> I am not sure yet what to do if we get a report that a default bridge is
>> not working anymore and a Tor Browser release is about to get built. But
>> maybe that case can get discussed and decided when we have reliable and
>> continuous measurements and notifications in place.

Something also to consider is that certain bridges may not be working
only from specific locations. That is OONI tests will not just tell you
if a bridge is DOWN, but will actually tell you in which countries a
certain bridge is or is not working.

I wonder if it would make sense to at some point consider integrating
this knowledge into Tor Browser itself. That is we could have some logic
as part of tor launcher that makes the user input their country and
depending on the country they input we know if we should advise them to
use bridges (because we know vanilla tor to be blocked there) and if so
which bridges are OK to be used there.

>> Georg
>>
>> [1] Ideally we would not need the measurement from anybody at all but
>> the bridge operator would set up the respective infrastructure to get
>> notified as soon as things are not working anymore. But I guess that
>> would raise the bar for deploying a respective bridge considerably to
>> the extent that we end up with less bridges we can ship in Tor Browser.
>> Which is not desirable.
>>
>>> --Roger
> I really like the idea and I agree OONI is in the best place to perform
> these tests.
>
> Like Georg pointed out we have to think what to do once the test detect
> a bridge is not working. From the above I think we have two paths:
>
> 1. operator - we should notify the operator for sure, if we can do it in
> an automate way that would be great. If not, we should centralize such
> notifications in a place and have a person in charge of reviewing those
> and sending a note to the operator.

My suggested step 0 is that we first get these measurements out and
somebody looks at the results and checks to see how useful they are and
actionable.

The concern here is that anything automated can potentially lead to
false positives and we don't want to be flooding bridge operators with
notices of things that are not necessarily true.

I also think that the monitoring of whether a bridge is up or down is
slightly out of scope with what OONI does, though it's also something
you can infer from the data as well if you take a certain vantage point,
maybe one Tor sets up, to be unfiltered.
> 2. product - how the product should behave giving such information is
> another question. I believe the product should be smart enough to
> 'whitelist' the bridges it receives notifications from ooni that they
> are not working, so they are not counted as options for it to use until
> otherwise. Which leads to the need of sending another notification when
> they are working again.

Yes, I really like this idea!

I wrote some ideas on how I would see this work above.

Really looking forward to moving this ahead!

~ Arturo