Identifying Airtel middleboxes that censor HTTPS traffic

by justDankin • 6 years ago

395 124 6 years ago

ThreeFx• 6 years ago

The TTL field isn't cryptographically protected though, so it should be very possible that there's an actor after Airtel which does the censoring, no?

justDankin9• 6 years ago

Using the TTL, we figured out that the censor kicks in at the kth hop. This kth box belonged to Airtel.

If a box was censoring after Airtel, we would have received a clean response (ICMP timeout) at hop k as well. Of course, the TTL itself can change during the run, but that wouldn't happen for so many cases :)

iso1631• 6 years ago

If you used an allowed domain name, did you get TTL expired messages back when using TTL=5/6/7/8?

davidhyde3• 6 years ago

Does Airtel block TLS 1.3 with the encrypted SNI extension enabled?

https://blog.cloudflare.com/encrypted-sni/

justDankin9• 6 years ago

I haven't been able to test that yet, neither am aware of any research which answers that question.

IMO the only way to do that would be to either (i) block the IP (high collateral blocking) or (ii) block TLS 1.3 itself (GFC does this).

A major blocker in answering this is finding a potentially blocked website that also supports TLS 1.3

davidhyde3• 6 years ago

Agreed and since TLS 1.3 is still work in progress the chances are slim that you will find a blocked website that meets your criteria. Great article, very accessible. Thanks!

detaro• 6 years ago

If they block ESNI, it's likely all ESNI requests would be blocked, because they can't tell if its a blocked site or not. (it'd be possible they only apply ESNI-blocking against IPs associated with blocked sites, but that seems unlikely)

tialaramex5• 6 years ago

This is wrong. The ClientHello message is not encrypted in TLS 1.3, so, the client has to announce any extensions in plaintext. Thus the Great Firewall blocks connections which say they want to do encrypted SNI.

TLS 1.3 works fine in China, but if you use TLS 1.3 with the earlier proposed encrypted SNI draft it is blocked. The Great Firewall can't tell which name you actually wanted, but it can tell you're encrypting the SNI and block that.

With the currently proposed Encrypted Client Hello with a GREASE-style dummy ECH on all connections (so the "real" Hello is sometimes in an encrypted block and sometimes that encrypted block was just noise), China would still be able to choose to block all ECH-enabled connections since their presence is detectable. This would break everything, but China can choose to do that. What happens next is a policy question.

If you want to sneak past nation state snooping you need something else, that's not what TLS is for. The TOR project does not directly offer this either, but they can help you find out how to connect to TOR in a sneaky way if that's necessary for you.

airtelblock3• 6 years ago

Setting esni to enabled in Firefox partially works.

But Airtel really, really wants to run scripts and show ads on blocked pages.

Duckduckgo - https://i.postimg.cc/SqkRhpRC/Mozilla-Firefox-29-09-20-w-PA....

Pirate Bay - https://i.postimg.cc/qMmwMXVY/t-29-09-20-9-Dv.png

dannyw• 6 years ago

DuckDuckGo is blocked in India??

Tepix• 6 years ago

How do they inject stuff into HTTPS pages?

codezero• 6 years ago

Heads up, that site was injected with a ton of malware/adware and redirects. Possibly their ad network got hosed, but that site doesn't seem safe unless you are locked down.

airtelblock3• 6 years ago

Duckduckgo is blocked in India with Airtel. HTTP url refuses to connect. I have to manually type https://duckduckgo.com to connect.

https://i.postimg.cc/SqkRhpRC/Mozilla-Firefox-29-09-20-w-PA....

justDankin9• 6 years ago

Yeah there was a huge outcry about it on Twitter some time back.

P.S. From the screenshot, it looks like you're trying to connect to [http]://duck.., hence shifting to https works.

This also hints towards a mixture of plain old http censorship and https censorship, which Airtel (in fact all ISPs) do randomly

airtelblock3• 6 years ago

This is not about the block a month ago. This is a new block.

When it was blocked a month ago, there was no notice. Now, there is a notice that it is a TRAI order. usually seen on sites that the govt themselves ask to block (piratebay, torrentz.eu, etc)

crazy_hombre• 6 years ago

That's weird. I just get a 301 redirect to the HTTPS version when I visit http://duckduckgo.com.

Santosh83• 6 years ago

I also get redirected to the HTTPS site. I think different ISPs block inconsistently.

xrisk2• 6 years ago

You should consider using HTTPS :)

gammarays_• 6 years ago

Please read the article :)

xrisk2• 6 years ago

I did. It was just a suggestion for OP to make his blog accessible via HTTPS.

kube-system• 6 years ago

I think they're talking about the blog itself not being configured to use HTTPS. I had to add an exception in my HTTPS everywhere extension to read the article.

justDankin9• 6 years ago

Haha yeah, I've been procrastinating for the last 5 years now

galliher2• 6 years ago

This is pretty clever! The reset within airtel_103.224.212.222_fullhd720.com.pcap arrives with IP time-to-live of fifty-seven while the segment carrying synchronize | acknowledge flags arrived with a time-to-live of forty-four. So without any active probing, and some educated guesses around default IP time-to-live values @ 1<<[6..8] you could could conclude that the reset originated fourteen hops closer to the capture than other packets in same five-tuple defined "flow".

justDankin9• 6 years ago

Yes! That's a great observation

However, the reason I went for probing the entire path is because the TTL itself can be spoofed

galliher2• 6 years ago

Agreed, it's flimsy. Certainly a bit more effort for them spoof it correctly though. Would need to watch traffic on the path back per flow to isolate the number of prior decrements to the TTL leading up to MitM, and then store that value until such time that it sees an SNI it cares about / it's time to generate a reset.

segfaultbuserr• 6 years ago

Tor Project closely cooperates with The Open Observatory of Network Interference. OONI provides a standardized testsuit that automatically probes a wide range of websites for evaluating the censorship status in different parts of the world, and volunteers can contribute data by running an instance.

But the objective of the project is only to identify governmental and ISP-wide censorships. And its capabilities for checking protocol-level censorship techniques are rather limited [0]. Perhaps we need a similar tool or fork for probing middleboxes and censorship in private networks.

[0] https://ooni.org/nettest/

myself2482• 6 years ago

Oooo.

How does that compare with something like RIPE ATLAS? Is there a list of such monitoring projects somewhere?

tyingq2• 6 years ago

Some older research suggests Airtel uses Netsweeper: https://citizenlab.ca/2018/04/planet-netsweeper-section-1-me...

Netsweeper is a Canadian company in the business of content filtering: https://www.netsweeper.com/

aeyes• 6 years ago

Another possible supplier is this company: https://www.ipoque.com/

They openly state that they supply ISPs with DPI hardware, I talked to them in person 15 years ago and they had no problem to admit that they also supply government institutions.

walrus0113• 6 years ago

My job overlaps with microwave and millimeter wave RF engineering somewhat. R&S also has no qualms about selling high-end spectrum analysis equipment to authoritarian regimes. Probably done through middlemen. For instance, you can find the Iranian government using their equipment in Tehran to hunt down things they don't like.

To be fair, there's probably less than ten manufacturers of their category of spectrum analysis gear (for commercial/non-military use) in the world, and their stuff is top quality.

amitheonlyone• 6 years ago

This is completely off-topic but the strike-through on links had me confused for a good few minutes. I was not clicking on those because I thought the link were not valid today and hence it has strike-through.

Semaphor2• 6 years ago

What strike-through?

folmar• 6 years ago

All links have a thin red line through them: https://pasteimg.com/images/2020/09/29/strika.png

walrus0113• 6 years ago

I wonder what they intend to do (other than just blocking entire IP ranges of non-Indian hosting providers, which I would not be surprised by) when things increasingly move to TLS1.3 with ESNI.

justDankin9• 6 years ago

Well, they could block TLS1.3 entirely (which would force hosts to drop down to 1.2 for connections)

GFC does this, I really hope it doesn't happen here

tialaramex5• 6 years ago

The Great Firewall does not block TLS 1.3. You may have seen headlines which claim it does, but they're based on a report that actually says it doesn't. Remember journalists probably know even less than you do about most things they write about!

In this case the report says the Great Firewall was determined to block the following specific combination:

* A ClientHello for TLS 1.3 that * Includes the 0xffce extension value (used for experimenting with an earlier SNI draft)

If you add a 0xffce extension full of random noise, the Great Firewall blocks it. If you use the same random noise but pick a different extension value (do not do this in production code - those aren't for your meddling!) the Great Firewall doesn't interfere at all.

We have yet to discover what happens if a big bang release of Encrypted Client Hello (the current iteration of the encrypted SNI work) just deluges the Great Firewall with ECH connections. But we do know TLS 1.3 has been used successfully for years from China.

You also mention this idea that it would "force hosts to drop down to 1.2 for connections".

It is hard to tell what you intended here, it would of course be possible to force the humans using a computer to downgrade, or to disable encryption, or to cease using a computer altogether, perhaps you could put a gun to their heads for example.

But TLS 1.3 has an anti-downgrade design. A [edited to add] modern TLS 1.3 capable web browser which connects to a TLS 1.3 capable web site but finds that the connection has been negotiated as TLS 1.2 instead will reject the connection as clearly under attack, you cannot reach that site until the problem is remedied. I think you would notice if all TLS 1.3 capable sites (about a third of popular sites) suddenly did not work from China, even the Chinese government might struggle to silence such confusion and dismay from their people.

walrus0113• 6 years ago

On a meta level, this is one of the reasons why I tell every junior/entry level person I encounter in the ISP business the following: Ethics is important in network engineering. You can and should refuse to do things that cause measurable harm to the Internet. You should understand why certain things are bad, and should make a conscious choice not to aid and abet them.

It is regretful that organizations like NANOG, RIPE, ARIN, APNIC and others do not take a stronger stance (on a global geopolitical level) against censorship of the Internet, and attempts to create walled gardens and national firewalls.

The people who possess the equivalent of 'enable'/'configure' on the core routers of gigantic ASNs have real power to refuse to further harm the Internet.

We have seen a number of discussions on HN about ethics in software engineering. A quick search for posts with 'ethics' in the subject line turns up a number of things. In my opinion, ethics in the ISP, telecom and network engineering industry is equally important.

jacquesm2• 6 years ago

This goes for any position with responsibility for and access to end user data. For instance: administering a mail server or hosted mail account for the company, dealing with the fall-out of security incidents and so on.

joosters5• 6 years ago

All well and good, but are these kind of 'middleboxes' unequivocally unethical? For example, some ISPs might want to block highly illegal content - let's use the typical examples, e.g. child porn sites, malware domains, and so on. It's not inherently unethical (or, at least, there are plenty of reasonable people who would say it is ethical) to install a middlebox that will make it more difficult for users to access these sites.

So now your company has got a content blocker installed. What exactly are your network engineers meant to do now? Demand personal refusal over any additions to the sites that these boxes will block? That seems highly unlikely to happen, and how could that even work in practice? Are all the engineers meant to vote on blocks, and only those unopposed sites get added to the list?

Can ethical network engineers usefully oppose content blocking?

walrus0113• 6 years ago

let's use the typical examples, e.g. child porn sites, malware domains

a futile game of whack-a-mole that only serves to make politicians feel good, and so they can claim they're "doing something" about social threats.

malware domains can be adequately addressed at the application level through things such as: https://www.google.com/search?channel=fs&client=ubuntu&q=goo...

commandlinefan2• 6 years ago

some ISPs might want to block highly illegal content

There is no way - not even a theoretical way - to allow blocking of illegal content (for any definition of illegal) that won't allow for blocking of any other arbitrary content. Censorship is binary. You can accept either none of it, or all of it.

cryptonector• 6 years ago

ISPs should help LEAs find illegal content all day long, but developing this sort of blocking technology is not the same thing, and neither is using it. There are legitimate uses of it though, namely in employers' networks, for example.

rhizome• 6 years ago

How do you propose to determine that a new, never-seen-before URL hosts child porn? In the US, viewing child porn is a strict-liability offense, meaning you are guilty of a serious felony just by looking at a page with the image(s) on it.

There are also civil and criminal liability concerns at the corporate level by assuming the responsibility for constructing and/or maintaining these filters.

kitteh4• 6 years ago

It'd be nice if we could address some things like BCP38 (anti spoofing), RPKI, route filtering and folks who knowingly support infrastructure that's used for outbound ddos (c2s and regular hosts), spam and malware phishing. Plenty of hosting shops in US and Canada have these problems. That seems a bit more within our reach whereas an ISP in India is more than happy to pay a vendor to implement middlebox packet molesters.

walrus0113• 6 years ago

Absolutely agreed. It is really easy to be a shitty, lazy colocation/hosting/VPS hosting company. It is somewhat more effort and more difficult to be a proper one. Margins are so thin in the hosting business that it would be own version of a personal hell... I feel a lot better about operating symmetric gigabit last mile residential services.

ev12• 6 years ago

I've been dealing with a ban evader/forum shock image spammer for months now, and the place he is buying proxies from is actively doing BGP hijacking on resources owned by AT&T, Windstream, hospitals and universities - for the primary purpose of carding and fraud. I haven't managed to get anyone knowledgeable at those companies to figure out how to pressure the small upstreams (that are not those T1s) to stop it.

foobiekr• 6 years ago

This is a great position. As someone who works in SP on the vendor side, I think it's great someone is carrying the torch.

Kaknut• 6 years ago

Really loved your article buy why no SSL/HTTPS? It's free afterall.

cnst3• 6 years ago

HTTPS is not free. It has a very significant management/maintenance and compatibility overhead, which is unavoidable by the very nature of HTTPS.

sroussey2• 6 years ago

Which can be automated.

ac29• 6 years ago

On the scale of a personal blog, its approximately 0 minutes per year to maintain HTTPS certificates using Lets Encrypt with something like certbot (or use Caddy, which handles it on it own).

commandlinefan2• 6 years ago

We have SNI because it's needed to support virtual hosting, which we really only need because IPv4 addresses are scarce. If we could ever get to IPv6, SNI could be retired completely.

mindslight• 6 years ago

Sure, but the security vulnerability of unencrypted SNI is still present with IP based vhosts - the destination IP precisely identifies the site.

occamrazor• 6 years ago

True, but then IP based blocking would be highly effective.

hedora• 6 years ago

I wish the world would adopt tcpcrypt. SSL provides two services, encryption, and authentication.

tcpcrypt just does the encryption part. Once the connection is established, userspace on both sides can invoke an ioctl that provides a session nonce. If the nonce matches on both sides, the connection is not man-in-the-middled. It’s easy to confirm the nonce matches: Both sides sign it and send it to the other party.

This has two big advantages:

(1) operating systems can opportunistically encrypt traffic for unmodified legacy applications and network protocols, allowing the endpoints to detect mass surveillance without requiring any certificates.

(2) Like newer versions of SSL, it encrypts the information currently sent in cleartext in the SNI, preventing the type of censorship in the article.

PowerBar2• 6 years ago

That would still rely on

A) some kind of secret that only the server knows that the client can verify in order to ensure it's not trading nonces with the MITM.

B) A way for the client to ensure that the nonce isn't being passed through a second tcpcrypt session between the MITM and the server with the connection being in cleartext between the 2 tcpcrypt streams.

Currently the best supported method of implementing both A and B is certificates, which means you may as well use TLS.

Dylan168072• 6 years ago

which means you may as well use TLS

Even if you don't authenticate at all, it makes it much more expensive to intercept all these connections.

And TLS lacks a way to automatically apply it to all connections.

Also I don't understand what scenario you're outlining with B.

totaldude87• 6 years ago

Theoretically, lets say am in India, using a tor browser, or an opera browser wit inbuilt VPN, would I see different results? my point is to see whether a VPN of sort can circumvent, in that case, what if all those browsers decides to create inbuilt vpn (for connection origins)..

does that makes sense?

dj_mc_merlin• 6 years ago

A VPN would prevent SNI snooping since traffic over it is encrypted. Of course, the ISP could block the TLS handshake between you and the VPN server if that's how your VPN functions.

Tor would similarly work.

Identifying Airtel middleboxes that censor HTTPS traffic

Related Stories