net: time unit used for TCP keep-alive on DragonFlyBSD should be seconds

[panjf2000]

After CL 542275 was merged into the master branch, I played around with this new feature along with investigating some known issues like #64251. As a result of which, I discovered a few issues and potential improvements on different platforms.

For the starter, DragonFlyBSD switched the time unit for TCP keep-alive from milliseconds to seconds since v5.8 while Go still uses milliseconds up to now (CL 542275 wasn't aware of it):

go/src/net/tcpsockopt_dragonfly.go

Lines 13 to 20 in aaf8e84

	func setKeepAlivePeriod(fd *netFD, d time.Duration) error {
	// The kernel expects milliseconds so round to next highest
	// millisecond.
	msecs := int(roundDurationUp(d, time.Millisecond))
	if err := fd.pfd.SetsockoptInt(syscall.IPPROTO_TCP, syscall.TCP_KEEPINTVL, msecs); err != nil {
	return wrapSyscallError("setsockopt", err)
	}
	err := fd.pfd.SetsockoptInt(syscall.IPPROTO_TCP, syscall.TCP_KEEPIDLE, msecs)

I think that we need to keep Go in sync with that considering that Go makes no guarantee to support any older versions of DragonFlyBSD other than the latest one:

Generally only the latest release version only. We have a builder, but it’s not the most stable of our ports.

References:

• DragonFly TCP
• Change tcp keepalive options from ms to seconds
• MinimumRequirements for DragonFlyBSD

[gopherbot]

Change https://go.dev/cl/565315 mentions this issue: net: harmonize the time units used for TCP keep-alive on DragonFly and other UNIX's by seconds