This breaks setting of file timestamps in go-fuse, downstream bug: hanwen/go-fuse#58

What do you think should happen? It seems to me that having the kernel return EINVAL is correct.

tv.Sec should be negative but tv.Usec should be positive.

Testcase: https://gist.github.com/rfjakob/6d3359a97d22f32083bc

 t=1960-01-10 23:00:00 +0000 UTC 
 nsec=-314758800000000000 
 tv.Sec=-314758799 tv.Usec=-999999

tv.Sec and tv.Usec are both wrong. It should be:

tv.Sec=-314758800 tv.Usec=0

nsectotimeval should do

if (tv.usec < 0) {
tv.usec += 1e6
tv.sec --
}

the same holds for NsecToTimeSpec.

func main() {
name := "test.txt"
var utimes [2]syscall.Timespec

f:=  time.FixedZone("bla",1) 
atime := time.Date(1968, 06, 18, 0, 0, 0, 99, f)
mtime := atime
utimes[0] = syscall.NsecToTimespec(atime.UnixNano())
utimes[1] = syscall.NsecToTimespec(mtime.UnixNano())

if e := syscall.UtimesNano(name, utimes[0:]); e != nil {
    log.Fatal(e)
}

}
=>
[hanwen@pomba test]$ go run tv.go
2015/09/29 08:38:00 invalid argument
exit status 1

I can prepare a patch if you want.

@hanwen I think the rounding should also be the other way round if nsec is negative

I noticed that the Linux kernel rounds DOWN while we round UP, check out http://lxr.free-electrons.com/source/kernel/time/time.c#L422 . This is unrelated, but it may be an opportunity to fix this as well.

Ping? NsecToTimeval is still broken.

CL https://golang.org/cl/30705 mentions this issue.

Fixed by 6c517df