52b7cbcb78c1 - tcp: improve handling of stopped timers

While tracking down a hung postgresql auto-failover issue ([1] awesome tool btw) I found that almost 1/3 of our ~ 170 FreeBSD 15.0 VMs show negative timers.

This was patched in CURRENT & STABLE already [2] but not in 15.0 nor 15.1, which I plan to upgrade to Very Soon Now.

I'm not entirely clear what the impact is. Does this mean that these TCP connections are stuck in some way, unable to progress? That is more or less the symptom of what I observed in the applications.

15.x does have new sysctl max retransmission timeouts [3], do any of these provide a way to influence (or unblock) the runtime behaviour? AFAICT not.

If not, this patch [2] seems errata-worthy; I can do the paperwork if agreed.

So far my approach is to use tcpdrop to unstick any connections with a negative time out.

A+
Dave

[1]: https://pg-auto-failover.readthedocs.io/
[2]: https://reviews.freebsd.org/D58484
[3]: https://reviews.freebsd.org/D50891