That’s an interesting point. As an app developer, I’d assumed that would be handled automatically by the OS.
What’s the best way to test for certificate validity? (In my case I’m interested in iOS, but the same concern must exist on all platforms).
What’s the best way to test for certificate validity? (In my case I’m interested in iOS, but the same concern must exist on all platforms).
Which is alarming.
We MITM and certificate validation works correctly.
I deal with this virtually every day.
- Is your root self-signed only? (Root CAs haven't been allowed to be self-signed only since roughly 2007 according to the principles of most browser root CA policies for public Roots. All public roots today are cross-signed among each other.)
- Does your root certificate have a valid revocation chain? Can you query up-to-date revocation information on it? (Modern Roots all have to have working revocation information, and Root CAs have been revoked in internet history, you cannot blindly trust your device's Root CA store over time without up to date revocation lists.)
Those are just two warnings I see most often from my dev tools on the MITM infrastructure I'm forced to deal with it. I know that this is compromising my security stance as a developer, and I know that turning off/ignoring those trade offs is a risk I directly pass on to users of anything I build. I've felt it a responsibility of professional ethics to pass on this concern to others in my company. I have debated many times whether if the right Root CA CVE or Self-Signed Certificate CVE comes across my dash if I will have to attempt to exercise the company's "Stop Work Authority" and refuse to continue development while being MITMed in a way that the company's security/safety infrastructure will not understand how to handle, but remains on my radar because I'm a professional and worrying about such things is my job.
Running a Root CA is a huge responsibility, and still has a ton of risks for the "real" Root CAs. (Just look at the recent battle between browser security teams and Symantec, for instance, over generating bad certificates.) Running a corporate MITM has all the same responsibility, with an even worse risk if you get it wrong (your entire company's device footprint has a single point of failure). It's such an incredible vulnerability/risk that whatever tiny gain it gives companies in surveillance over SNI sniffing and endpoint/device-deployed auditing tools is never worth the risk of subjecting so many developers to badly MITMed developer environments, especially some of the developers most at risk (bank software, health software, etc) of passing on the software equivalent of a bad MITM plague given the worst happens. I cannot imagine the blasé with which Corporate America has MITMed itself can be seen as anything but an incredible folly, if not today than certainly tomorrow (hopefully not after the worst happens).
It's fucking scary how far they're willing compromise security internally and externally to avoid extra work and maintain control.
- Everyone works on 8GB windows machine and sticky keyboard
- Remote desktop
- "You wanna install your IDE? Yeah contact IT, gonna take few days"
- Can't install any of cli tools
- Atlassian suite
- Spend half a day in meetings
- Scrum
Then they complain how hard it is to get a good developers...
However, I hear other companies like the configurability too...and their crappy processes were so easily configured that it became a crappy tool for their poor developers.
Often libraries have options to specify trusted root CAs as well as options to disable validation per host and/or globally. I've never come across any library that would have any of these options enabled by default.
With that said, apps like banks should still not be putting trust into the OS or anything else. Certificate pinning is a good and should be utilised, especially for sensitive systems such as banks' apps.
OpenSSL for years only provided some fairly hairy code if you actually wanted to do dnsName matching, which you absolutely should do. What that means is, lots of software was written (say, 10+ years ago) in which OpenSSL is checking that your peer has a "real" certificate but it doesn't care which one. A certificate for we-are.literally-thieves.example ? Cool, that's issued by a trusted CA and so it's fine. Oh you thought you were connecting to my-real-bank.example? You didn't ask me to check the name on the certificate and I don't bother providing a sensible API to do so anyway.
Here's their actual documentation:
> Versions prior to 1.0.2 did not perform hostname validation. Version 1.0.2 and up contain support for hostname validation, but they still require the user to call a few functions to set it up.
Modern (1.1 onward) releases of OpenSSL provide a sane API which checks names you give it, so if you tell OpenSSL to connect to my-real-bank.example it realises you don't think certificates for other names are OK. But the old ones didn't do that and the ones 10+ years ago expected you to grok PKIX (the Internet's agreed way of coercing the X.509 standard intended for the X.500 series Directory into a way to certify things on the Internet) or else give up.
So, if my company installed a CA on my phone that they issued in-house, and MiTM my traffic, they can spoof certs and most software will accept it.
To be really safe, you should pin the certificate to ensure that your code only trusts a specific certificate or specific authority, so, for e.g. it was signed by Let's Encrypt X with fingerprint Y and not, say, Digicert Z with fingerprint A.
If you're writing the backend and frontend, yor could go one step further and embed your own CA in the app and follow secure practice for managing the private key and issuing certificates to your infrastructure.
Building your own PKI is always potentially the safest option, and in practice it will usually be the least safe and most unreliable. The main attraction of your own PKI should not be the safety/ security you likely won't actually achieve in practice but other conveniences. For example your PKI can issue a 20 year cert. Maybe it shouldn't, but it can and that might work better for you than certificates which expire and introduce exciting last minute changes.
We specialise in this sort of thing at my work, I'm not suggesting anyone does this without first understanding the risks you mention above as well as the long term commitment required.
But done right, it is the most secure approach.
The most common, and automatic, is the verification of the chain of trust. On iOS this happens automatically if you use the standard network APIs against an HTTPS URL.
You can take it a step further and avoid MITM attacks where the middle party is able to mint trusted certs by doing something called certificate pinning. This is a manual verification that the certificate used by the server you’re connecting to has certain properties that you know match your API server’s.
A frequent problem on android that might lead some to throw validation under the bus, is that certificate validation is much less robust than the one in browsers, particularly on old devices. A typical scenario is that the certificate of one of your backend approaches EOL, ops dutifully obtain a new one and it checks out nicely on all browsers. But a whole bunch of older Androids that might make up a quarter of your user base of you are unlucky has never heard of the root certificates involved so the app becomes unusable. A similar situation can arise if you have clients that check revocation (good) but don't check for alternative signature chains like a modern browser would do (not so good).
The correct way to solve these situations is extending the server configuration with another certificate chain that is valid on the devices in question, no doubt about that. But when the app is not a core use of the backend and there server is not run by the same organization, breaching the wall of "but it works in all browsers, clearly the error must be on your side" defense can be quite hard. Nontechnical leadership will be extremely tempted to do the writing thing.