Re: [chrony-dev] Fw: leap seconds correction

[Marek Behun <kabel@xxxxxxxxxxxx>]

On Wed, 12 Feb 2014 11:09:43 +0100
Miroslav Lichvar <mlichvar@xxxxxxxxxx> wrote:

> On Wed, Feb 12, 2014 at 02:14:06AM +0100, Marek Behun wrote:
> > > >  http://www.ucolick.org/~sla/leapsecs/right+gps.html
> 
> The last two paragraphs on that page imply that chrony smears the leap
> second, that's not true. Similarly to ntpd, it just tells the kernel
> to insert/delete the leap second and the Linux kernel does it by
> stepping the clock back/forward by one second at midnight.
> 
> > I was thinking about how to do this the best way and I have decided
> > to write a compatibility library (which will override time
> > getting/setting system calls with LD_PRELOAD), so the code of
> > ntp/chrony or any other software that needs to think system clock is
> > represented in classic POSIX timestamp doesn't need to be changed.
> > You can find it at https://github.com/elkablo/time2posix .
> 
> That's pretty cool.
> 
> Did you test how it works during leap second? I suspect the ambiguity
> in the UTC->right conversion when leap second is inserted could cause
> some troubles. In linuxptp (PTP uses TAI) the problem is avoided by
> suspending the operation during that second. I think chronyd and ntpd
> would need to do that too.
> 
> Also, you might need to intercept recv() and similar to adjust the
> timestamps in the SO_TIMESTAMP and SO_TIMESTAMPING control messages.
> 

I tried to simulate a leap second with little success - I tried
running ntpd and changing kernel's adjtimex() status to STA_INS,
but it didn't do anything observable. Can chrony simulate a leap second
from a local leap second table file?

Currently my code deletes STA_INS and STA_DEL bits in adjtimex() status
when setting time status and sets STA_INS/STA_DEL bits (and
return value of adjtimex() - TIME_INS, TIME_DEL, TIME_OOPS or TIME_WAIT)
depending on the system time and leap seconds table.

Thank you about the SO_TIMESTAMP/ING control messages, I didn't know
about them. Till now I have only intercepted time setting functions and
utmp/utmpx functions. File system API functions like utime, stat and
others will also need to be intercepted.

Marek

Attachment: signature.asc
Description: PGP signature