Informative article but it meanders about for way too long.
In some circumstances, Windows resets its clock based on the ServerUnixTime field of incoming TLS handshakes, for reasons that are not completely clear
OpenSSL puts random numbers in ServerUnixTime
Problem!
Disable via HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\W32Time\SecureTimeLimits
Their official advice is to disable STS when using NTP.
As for the explanation, I think it was just an example of bad decisions compounding on themselves.
Oh no, it’s difficult to sync time because the secure communication layer doesn’t work when our time is already out of sync. That’s okay, we’ll use a totally other dubious mechanism instead of fixing that.
Oh no, the dubious mechanism is giving us bad results sometimes. That’s okay, we’ll introduce weird heuristics to attempt the impossible problem of determining whether the dubious mechanism’s output is trustworthy.
Oh no, the heuristic fails sometimes. That’s okay, “We agree that the overall direction of technology with the adaption of TLS v1.3 and other developments in this area could make Secure Time Seeding decreasingly effective over time, but we are not aware of any bugs arising from their use. This technology direction also makes heuristic calculation of time using SSL/TLS far less attractive when compared to deterministic, secure time synchronization.”
Informative article but it meanders about for way too long.
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\W32Time\SecureTimeLimitsSee? That didn’t take long.
So does that mean one needs to run NTP as well as the domain-based time sync, for when the donation based one fails?
Seems weird. I wonder why they’re so cagey about it
Their official advice is to disable STS when using NTP.
As for the explanation, I think it was just an example of bad decisions compounding on themselves.