[stephenmac98]

It's 2024, PKI best practices are well known and well documented, anybody still using a self-signed certs on their mail server (or anywhere) is either lazy or stupid.

Plenty of existing applications will refuse to connect to a self-signed certificate on the belief that allowing the end-user to confirm a certificate offers basically 0 protection against malicious actors.

[shakow]

There is no security hole if I am singing my own certificate for my own mails on my own server; it would mean that I do not trust... myself?

Now if I were to provide this as a commercial service, sure, my customers may be worried.

[stephenmac98]

"This is good enough because I don't expect anyone other than me will use it" is lazy

What would happen if you connected to your mail client today and you got prompted "Trust this certificate?" showing a certificate with the same subject as the one you generated? Most people would click trust and get MITM'ed

Allowing self signed certificates significantly lowers the bar when it comes to generating a new certificate which can closely resemble an existing certificate

Beyond that, the management of multiple trusted certificates creates all sorts of room for confusion in an environment. Presumably most services that you run, run over TLS, do you really maintain every certificate both on it's application and on everything which needs to connect to it? That's a huge amount more effort than signing all your PKI with an internal CA, the configuring your connecting applications to trust that CA

[akira2501]

> Most people would click trust and get MITM'ed

So accept self signed on first connection with a detailed panel showing the certificate fingerprint. Then after that require a more involved process to accept a new certificate.

> do you really maintain every certificate both on it's application and on everything which needs to connect to it?

These are client certificates, and in some cases, they're actually pretty awesome.

> than signing all your PKI with an internal CA

That's a single layer of abstraction away from a self signed certificate, because, your CA _is_ a self signed certificate in this scenario. You've taken any defense in depth and thrown it right out the window.

The purpose of software is to make things possible not enforce random pedantry.

[denkmoon]

>"This is good enough because I don't expect anyone other than me will use it" is lazy

is both a mischaracterisation of the argument, and wrong. It's not lazy, it's a choice with pros and cons. Just because you don't like it does not mean it is lazy. Again, issuing your own certificates is a choice.

Allowing self signed certificates does not "significant lower the bar". Did you know that all root certificates are self signed?

The management of multiple trusted certificates is basic administration for large private networks. Yes, TLS and certificate management can be complex, but that is not a good argument for disallowing it, and the idea that managing your own certificate trust is against "best practices" is ludicrous.

[anamexis]

It's not that you're trusting your own certificate, it's that you're trusting any self-signed certificate, leaving you open to getting MITM'ed.

[denkmoon]

Why would this oblige the client to trust any self-signed cert as opposed to trusting all certificates whose chain of trust can be established using the system's trust store? The reporter isn't asking for mail to automatically trust untrusted certificates, they have added them to the trust store.

[darknavi]

Can't you install trusted certificates by hand on iOS?

[anamexis]

Yes, and you still can. That's not broken.

[digitalPhonix]

Based on the bug report I think it is. Mail is not respecting the self signed certificate that’s been added to the user trust store.

[bdavbdav]

How? You add your signed cert explicitly to the trust store. You’re not just hitting “trust” every time.

[cpach]

Please keep in mind that a self-signed certificate is quite different from a certificate that is signed using a private CA.

The self-signed certificate has no link to a trust anchor. So it’s easy for Mallory to replace it with her own malicious certificate. It’s much harder for Mallory to replace a certificate that is tied to a CA.

[kortilla]

It’s 2024, we’ve seen countless examples of sophisticated hackers getting into all kinds of systems. Anybody who makes a blanket statement that you have to trust the public PKI is either lazy or stupid.

SSH has TOFU and it works very well if you don’t want a key infrastructure.

[mmd45]

explain how a pinned self signed cert is insecure. i don't see it. it would seem to be more secure than one signed by a public CA that's not pinned.

[stephenmac98]

I didn't say a "pinned self signed cert is insecure"

I said that self-signed certs are a lazy choice

I also said "allowing the end-user to confirm a certificate offers basically 0 protection" If an average user get's prompted to trust a certificate they will do so blindly At most, someone might look at the subject, but it's 0 effort for a malicious actor to generate a self-signed cert with the same subject, which will be sufficient to fool a decent chunk of users

Pinned certificates do relieve the above issue, but it is still a lazy choice that creates increased long term complexity in the configuration of multi-system environments Presumably most services that you run, run over TLS, do you really maintain every certificate both on it's application and on everything which needs to connect to it? That's a huge amount more effort than signing all your PKI with an internal CA, the configuring your connecting applications to trust that CA Using a CA also allows for use of CRLs or OCPS. If you have 20 devices configured to trust a given self-signed certificate, and that certificate leaks, you now have to update all 20 devices to remove that trust. If you used a CA and implemented either a CRL or OCSP, then you only have to update the respective impelmentation and all of yoru clients will immediately stop trusting that certificate.

In Summary: Using an internal CA offers all the potential protections of pinned certificate, with a number of additional useful security options like OCSP or CRLs Using Self signed certificates creates more work when handling certificate leaks or certificate rotation Using a CA is the industry standard practice, I highly doubt there is a single outward facing project by a major company using a directly self-signed certificate. BUT A self signed certificate is lower effort on the initial setup

Lazy

[kortilla]

You need to calm down and take a step back to realize not everyone needs to support 20 devices or even 2. What you’ve suggested is a ridiculous blanket statement assuming everyone is setting up things for a fleet of clients.

[mmd45]

for the use case of a single user IMAP server this is all way, way, too complicated and buys you nothing in terms of security. it's completely analogous to why we dont use CAs to validate openssh host certificates.

[Twisell]

Yes it's a analogous using CA is still a higher bar, but it would arguably be better to also use CA to validate openssh host certificates for all the reasons he listed above.

So maybe we should ask ourselves why can't we just figure out a way to improve handling of CA? Thanks to Let's Encrypt https coverage dramatically improved, now is maybe the time for more people to switch to self CA.

I agree though that promoting adoption through good tooling and pedagogy would be a nicer approach than Apple slap on the wrist.

[lxgr]

It really only is for bad practical reasons, that all coincidentally make it harder and harder to self-host stuff locally without paying a few dollars a month or year here and there to various rent seekers.

"Just use Letsencrypt" really is the correct answer for 99% of use cases, but good luck if you find yourself with one from the 1%. You'll get an army of people mindlessly parroting "best practices" and will assume you're incompetent/lazy if you can't find a way to make them work for you.

[User11110]

Internal CAs and self signed certificates are different. You can still generate a CA, sign your certificates, import your own CA into your phone and have that verify your certificates. You don't need Letsencrypt. But you'll learn in time.

[lxgr]

Thanks for the condescension, but I know how to do all of this. I've done it before. And because of that, I can first hand attest that it's way too complicated.

No non-sophisticated user is able to run their own local CA, and that's why their NAS, IoT setup etc. all run over HTTP only, which in turn has implications for available web APIs (thanks to "secure origin only" policies and no exemption for local IPs/zeroconf domains) and many other things.

It also doesn't work for at least modern Android apps, since Android no longer makes user-provided CA certificates available to (non-browser) apps anymore, I believe, unless they're compiled with a special debugging parameter. On iOS it's still possible, but I'm not sure how long it's going to stay that way.

[stephenmac98]

If a user set up a NAS they should be capable of googling

"Openssl how to set up a CA" > First link fully explains it https://arminreiter.com/2022/01/create-your-own-certificate-...

"How to import CA into iPhone" > First link fully explains it https://www.ibm.com/docs/en/mpf/7.1.0?topic=certificates-ins...

"Android app customize trusted CAs" > First link fully explains it https://developer.android.com/privacy-and-security/security-...

The barrier to entry on PKI isn't that it's hard, it seems to be that people just can't be bothered, PKI is among the most google-able tech processes out there

[lxgr]

Setting up a NAS means buying one on Amazon and plugging it in.

You're completely out of touch with the majority of the userbase of these products if you think even one in 10 NAS users will set up their own CA using OpenSSL (in a secure way that doesn't expose themselves to being MITMed even on public sites such as that of their bank down the road).

[digitalPhonix]

How? An internal CA is just a self-signed certificate that you’ve told your device to trust; and to trust other certificates signed by it.

Somewhere you still need to trust a self-signed certificate.

[cpach]

You can guard the root certificate better than the leaf certificate. For example, you can keep it offline in an air-gapped environment.

[lxgr]

Or is operating a local-only mailserver not connected to the larger internet? I guess that's a lazy or stupid thing too, these days...

I'm a fan of having TLS on by default for everything on the Internet, but I'm seriously annoyed by the collateral damage to local self-hosted services the implementation of that has caused.

It shouldn't be this hard to e.g. host web server on my local network that browsers grace with "trusted website APIs", but it really is. Why on earth do I need to set up Letsencrypt (and by extension at least DNS) on a local website if I want to be able to use a game pad on it, for example!? https://developer.mozilla.org/en-US/docs/Web/Security/Secure...

We absolutely need a localhost and local domain exemption for both TLS/X.509 certificate validation and web APIs. For example, TOFU seems like a much better model for that use case than trying to bend the "public Internet" model until it fits. SSH has had considerable success in that model, for example.

[stephenmac98]

You have to consider the rarity of your use case vs the use case they're defending against.

How often do you think someone tries to connect their gamepad to a local server? Not never, but the total amount of users doing it is probably high tens or low hundreds at most

Compare that to how often gamepad users try to connect to a malicious website - probably hundreds or ever thousands of times a day.

Loosening certificate validation further expose the many less than competent users to risk, and the potential impact both on the customer base and on the product's reputation are significantly higher than the risks of making it cost a couple bucks a year to allow your gamepad to connect to a local server.

For something like a computer, there is a legitimate argument for allowing the user to bypass SSL/TLS restrictions (after some resistance) because laptops are used for development.

I can almost guarantee that the gamepad developers had an options for certificate validation bypass in it's developer options, but they're not gonna expose that needlessly when it offers no benefit, but exposes their customers to additional risk. Your gamepad is likely not a development device after all

[lxgr]

Any malicious website can access my gamepad, since it can trivially get a Letsencrypt certificate – the only requirement for getting "secure origin" API access.

What exactly is this restriction preventing me from, then? (And what does a malicious website do with my gamepad data anyway?)

> Not never, but the total amount of users doing it is probably high tens or low hundreds at most

Yes, I'm fully aware that local hosting is rare in the grand scheme of things, but I think you're vastly underestimating the potential. It's currently not even possible to do much better even as a commercial NAS provider, and these are somewhat popular.

A big part of that also seems like a chicken and egg problem: Fewer and fewer users do it because it's getting harder and harder, thanks to browser standards and OS defaults being largely driven by stakeholders that have no interest in it becoming easier.

Yes, none of this is an evil conspiracy; it's just a question of incentives and priorities in the end. I just find it so sad how willingly even a "hacker" audience here embraces the move towards more and more centralization, on more than one dimension. (Peer-to-peer vs. client-to-server, "trusted CA only" vs. trust on first use, cloud vs. self-hosting etc.)

[stephenmac98]

Allowing self-signed certificates creates a higher risk for MITM attacks. Sure you can trivially get a letsencrypt certificate once you register a DNS entry, but you can't trivially get a letsencrypt certificate which validates google.com

If you control the local network it's trivial to redirect traffic intended for elsewhere, like "google.com", and trivial to have the server it redirects to present a certificate with "google.com" in it's subject or SAN.

What would happen on a laptop is you would be hit with a certificate validation error because it was self signed, and on the laptop you have the ability to bypass it, but that ability to bypass is very dangerous. Most users will not properly check a certificate before clicking to trust it.

As far as what could be done, "this is a low value device to an attacker" is not a security measure, but beyond that I'm sure that people have bought games on a gamepad, and anything which involves financial transactions has the potential for malicious behavior with severe consequences

[lxgr]

Then only allow self-signed certificates for literal IPs or those on .local (and other private/reserved TLDs).

Right now, .local is completely impossible to encrypt, as well as impossible to use “secure origin” APIs on, which is a shame.

[stackskipton]

.local also hasn't been best practice since 2005. Current recommendation, because of Certificates is to use internal only subdomain of domain you have control over.

[diogocp]

> We absolutely need a localhost and local domain exemption for both TLS/X.509 certificate validation and web APIs.

localhost is already considered a secure origin.

Local networks are horribly insecure; easily the most likely place for a MITM attack.

[lxgr]

Ah, that’s good – it’s been a while since I last had to work around that.

And I generally agree on local networks being insecure. So how about making them more secure instead of marginalizing them even more?

TOFU for TLS certs on .local (for Zeroconf, and maybe something else/new for local DNS) would be a huge step forward from unencrypted and unauthenticated HTTP. Such sites could even still be displayed with a broken padlock or whatever HTTP gets these days to not create any false expectations by users.