Re: RFC: time_namespaces(7) manual page

[Thomas Gleixner <tglx@linutronix.de>]

Andrei Vagin <avagin@gmail.com> writes:
> On Sat, Apr 04, 2020 at 01:08:50PM +0200, Michael Kerrisk (man-pages) wrote:
>>    /proc/PID/timens_offsets
>>        Associated  with  each  time namespace are offsets, expressed with
>>        respect to the initial time namespace, that define the  values  of
>>        the  monotonic  and  boot clocks in that namespace.  These offsets
>>        are exposed via the file  /proc/PID/timens_offsets.   Within  this
>>        file,  the  offsets  are  expressed  as  lines consisting of three
>>        space-delimited fields:
>> 
>>            <clock-id> <offset-secs> <offset-nanosecs>
>> 
>>        The clock-id identifies the clock whose offsets are  being  shown.
>>        This field is either 1, for CLOCK_MONOTONIC, or 7, for CLOCK_BOOT‐
>>        TIME.  The remaining  fields  express  the  offset  (seconds  plus
>>        nanoseconds)  for the clock in this time namespace.  These offsets
>>        are expressed relative to the clock values  in  the  initial  time
>>        namespace.   In  the  initial time namespace, the contents of this
>>        file are as follows:
>
> I think we can mention that offset-secs can be negative, but
> offset-nanosleep has to be 0 or positive.

I assume you meant offset-nanosecs :)

That aside, there are also limitations in place.

  1) Negative offsets which would offset time into negative space are
     rejected, i.e. its enforced that

          now(CLOCK) + offset[CLOCK] >= 0

     This is necessary as the kernel expects and also enforces that time
     cannot be negative.

  2) Positive offsets which would offset time above KTTIME_SEC_MAX / 2 are
     rejected, i.e. it's enforced that

          now(CLOCK) + offset[CLOCK] <= KTIME_SEC_MAX / 2

     That is done to prevent that clocks wrap around if the offset would
     bring it close enough to the wrap around point.

     The cutoff value is a pretty arbitrary choice (~146 years). So to
     hit this you'd need a system which has an uptime of > 146 years,
     which is pretty much unrealistic.

Thanks,

        tglx