To update 10th-gen Honda Civics, Honda ships updates on specially-formatted USB drives. They're essentially Android 4.2.2rc1-era recovery packages with some Honda-added version checks (which can be spoofed). The packages are signed with the publicly-known AOSP test key, so with physical access to the front USB port you can sign and flash your own package for arbitrary code execution on the headunit. This doesn't require root/su. I've run it end-to-end on my own 2021 Civic and separately confirmed an official EU update file carries the AOSP test-key signature. Tooling and writeup in the post.
A number of other cars' infotainment systems are also based on ASOP. I remember downloading updates for my Hyundai which were also essentially Android images
The head units themselves are very dated and simply could not run recent versions of Android. I have a 2020 and I'm always eyeing up the after market units which are all better in every way.
I’ve heard product managers proudly proclaim their firmware was signed using the corporate internal signing service (good).
Of course, the question explicitly being asked (related to internal mandate) was if the firmware was signed — not if the firmware update process actually checked the signature (it certainly did not).
Wonder how good the rest of the security is. The head unit is likely hooked up to a CAN gateway, can it call into telematics. Maybe find some novel way to abuse carplay/aa to call home.
Seeing more and more projects eschew code docs with the idea that "well architected code can be queried by LLMs" and stick to more functional runbook style docs. It really is unlikely that at any given point all of the docs of a project are up to date with the code.
I'm generally aligned with this, but it is predicated on the whole "well architected" code part.
The test can show intended use, show interesting corner cases, and I know it is up to date because it is constantly running and passing.
I think that is a huge underrated benefit of adding a lot more testing.
If I think a developer is going to ask a question of how something works, or about a corner case, isn't that deserving of a test, so they can just see proof of the answer to their question immediately rather than trying to re-derive it?
You know what, you are right on the money with that. I think if you expand to include functional/smoke/e2e tests, that covers pretty much everything documentation is supposed to be.
Just by running them you can measure if they are in or out of sync with the code (well, if they were written correctly).
Hyundai head units at one point used an RSA key you got by googling “RSA key” (no joke: https://programmingwithstyle.com/posts/howihackedmycar/ ), an honestly even more amazing mistake since it required effort rather than just a default.
Of course, the question explicitly being asked (related to internal mandate) was if the firmware was signed — not if the firmware update process actually checked the signature (it certainly did not).
I'm generally aligned with this, but it is predicated on the whole "well architected" code part.
The test can show intended use, show interesting corner cases, and I know it is up to date because it is constantly running and passing.
I think that is a huge underrated benefit of adding a lot more testing.
If I think a developer is going to ask a question of how something works, or about a corner case, isn't that deserving of a test, so they can just see proof of the answer to their question immediately rather than trying to re-derive it?
Just by running them you can measure if they are in or out of sync with the code (well, if they were written correctly).