Don't trust the local machine's system time. #1453
Conversation
This can never happen I think. When joining an age instance, the client always sends a
According to my notes, none of the fan servers support Perhaps |
Previously, when asking for the server's time, the calculation was done by basically offsetting the local system time by some previously calculated delta. This was basically useless because the user can easily change the system time out from under us and cause huge deltas. The attempt to fix things in H-uru#173 by using `WM_TIMECHANGE` was a good idea but ultimately incorrect because `hsTimer` is a monotonic timer. At this point, it uses `std::chrono::steady_clock` as its backend, so attempting to use `hsTimer` to recalculate the server/client time offset is useless. Therefore, this changes us to simply save the last time value that we received from the game server and the monotonic time that we received that update. This is definitely not perfect because we don't consider things like round trip time for the message and time spent in the update loop. However, it *is* a clear improvement in that now the KI will always display what the server thinks Mountain time is, and it should be impossible to perform system clock based exploits (such as instantly baking pellets)... without any platform specific code!
Co-authored-by: dgelessus <dgelessus@users.noreply.github.com>
Previously, when asking for the server's time, the calculation was done by basically offsetting the local system time by some previously calculated delta. This was basically useless because the user can easily change the system time out from under us and cause huge deltas. The attempt to fix things in #173 by using
WM_TIMECHANGEwas a good idea but ultimately incorrect becausehsTimeris a monotonic timer. At this point, it usesstd::chrono::steady_clockas its backend, so attempting to usehsTimerto recalculate the server/client time offset is useless.Therefore, this changes us to simply save the last time value that we received from the game server and the monotonic time that we received that update. This is definitely not perfect because we don't consider things like round trip time for the message and time spent in the update loop. However, it is a clear improvement in that now the KI will always display what the server thinks Mountain time is, and it should be impossible to perform system clock based exploits (such as instantly baking pellets)... without any platform specific code!
Future work: currently, we are implicitly depending on the server to send us periodic
plNetMessages with the time sent value to synchronize the time. If that's never done, then we simply return the client's time (yikes!). This is a decent assumption in all current content because we will receiveplNetMsgGameMessagewithplServerReplyfor region enters on spawn andplNetSDLMsgfor the Age state (when an Age has any non-excluded SDL states), however, there is little guarantee that the server will continue to send us these messages if we are in an Age by ourselves doing nothing. One option would be to send an innocuous message such asplNetMsgGetSharedStateperiodically. I have not implemented that here because it is not immediately apparent whether or not the various server types implement this message.