struct timex and Linux adjtimex()

[Konstantin Belousov]

On Wed, Dec 02, 2020 at 09:23:54PM -0800, Cy Schubert wrote:
> An NTP developer at approached me about the prospect of a system call to 
> add or subtract time from the system clock, returning the current time 
> after the update. The two options were:
> 
> 1. A new syscall, similar to clock_settime() and clock_gettime() that 
> adds/subtracts the delta, in a struct timespec, via a call to 
> kern_clock_settime(). Then atomically returns the current time as if 
> clock_gettime() was immediately called.
> 
> 2. His colleague didn't think this was appropriate for their needs. He 
> wants a FreeBSD implementation of the Linux adjtimex() syscall.
> 
> Though non-standard option 1 is least intrusive on existing applications.
> 
> Option 2, implement the Linux adjtimex():
> 
> 1. The Linux adjtimex() appears to be a re-implementation of ntp_adjtime() 
> with the addition of ADJ_SETOFFSET.
> 
> 2. ADJ_SETOFFSET adds or subtracts the time specified in a struct timeval 
> within the timex struct. Unfortunately the FreeBSD timex struct contains no 
> timeval. To implement this would require versioned symbols to maintain 
> backward compatiblity.
There are two options, as far as I see:
1. Implement new syscall, which would take extended struct timex.
   ntp_adjtimex() perhaps should be kept for backward compatibility.
   [It does not matter where struct timeval is placed in the updated
   struct timex, see below].

2. Extend existing struct timex with struct timeval at the end of
   the structure, but do not require it for any mode except when
   ADJ_SETOFFSET is specified in modes.  In other words, syscall
   should copyin legacy struct timex, and if ADJ_SETOFFSET is set,
   copying struct timeval from the address right after timex.

IMO option 1 is relatively simple and low overhead.  Option 2 is just
more work for a little gain. If you go the route 1, please add some padding
at the end of extended struct timex so potential future modifications do
not require a new syscall.

Add me to the review anyway.

NB. compat32 for ntp_adjtime(2) seems to be completely broken.  It is probably
a good moment to fix old syscall, since you would need to implement compat32
shims for the new one anyway.

> 
> Up for discussion is:
> 
> 1. Are we, FreeBSD, interested in implementing the Linux adjtimex() 
> syscall? (I would take on the task to author it should we feel this is a 
> worthwhile project. OTOH I prefer not to spend the time working on this if 
> the community feels otherwise.)
> 
> 2. From a cursory scan through the tree it appears that ntp is the only 
> internal consumer of struct timex. There are probably others in ports and 
> other third party software such as openntpd and chrony. A versioned symbol 
> should satisfy old applications which might use the previous timex struct.
> 
> Are we interested in the Linux adjtimex(2) for FreeBSD?
> 
> 
> -- 
> Cheers,
> Cy Schubert <Cy.Schubert at cschubert.com>
> FreeBSD UNIX:  <cy at FreeBSD.org>   Web:  https://FreeBSD.org
> NTP:           <cy at nwtime.org>    Web:  https://nwtime.org
> 
> 	The need of the many outweighs the greed of the few.
> 
> 
> _______________________________________________
> freebsd-arch at freebsd.org mailing list
> https://lists.freebsd.org/mailman/listinfo/freebsd-arch
> To unsubscribe, send any mail to "freebsd-arch-unsubscribe at freebsd.org"