Processing saved packet may overflow cubic cc #1163
Conversation
Saved packets are process with a timestamp that correspond to time they have been received. This may overflow the calculation of the time spent in ca.
|
This should fix bug 1716588 (only the crash reports that involve cubic). This is triggered when saved-packets are process (see this crash report). |
There was a problem hiding this comment.
Looks good.
We have quite a few time-traveling tests that could have tripped this, but they all use Reno. Maybe we should consider switching our default CC in tests (maybe using a feature flag so that we can just build and run tests under two different configurations).
| .map_or(min_rtt, |t| now + min_rtt - t) | ||
| .map_or(min_rtt, |t| { | ||
| if now + min_rtt < t { | ||
| min_rtt |
There was a problem hiding this comment.
It might be worth commenting here that this only happens when processing old packets that were saved and replayed with old timestamps. In that case, using a fixed value is safe.
It's a shame that we can't use saturating_sub() and all that with Duration yet. Those could make this slightly nicer.
Saved packets are process with a timestamp that correspond to time they have been received. This may overflow the calculation of the time spent in ca.