[OpenAFS-devel] linux 2.6.11.4 and bogus 'setting clock back'

[Troy Benjegerdes]

On Wed, Mar 16, 2005 at 07:42:57PM -0800, Russ Allbery wrote:
> Kris Van Hees <aedil-afs@alchar.org> writes:
> 
> > The large number is (from the look of it) a regular time value (number
> > of seconds from epoch), e.g. 1111023402 is "Thu Mar 17 01:36:42 2005".
> 
> Which indicates, given the message, that the client thinks the server has
> a time of 0.

Yes, this seems to be what is going on...

> >> afs: setting clock back 10 seconds (of 1111023402, via 147.155.137.11 in
> >> cell scl.ameslab.gov); clock is still fast.
> 
> Kris almost certainly knows this, but for other people reading, the "of X"
> string in this message in AFS has historically always indicated the amount
> of time discrepancy between the server and the client.
> 
> The solution is certainly to run afsd with -nosettime and use ntpd; AFS
> time setting has a variety of very odd problems.

But will -nosettime solve the problem of thinking the server has a 0
time? I knew -nosettime was a bogon, but at the moment I'd think the
bogus server time is the real problem..

We should probably print a more informative warning than just "can't
contact server".. when run with -nosettime, does afsd log any messages
about the time being off?