[Ntp] Re: ntpd-rs: RefID/LI incorrectly reported for source-less stratum-1 config (root cause + patch); also a draft-ID interop note

[Danny Mayer <[e-mail]>]

This is an IETF Working Group. If you have problems with an 
implementation you need to contact the development team for that 
implementation. They should have information on how to do that on the 
site where you got the implementation and in the documenation that came 
with the code.

Danny

On 8/1/26 4:44 AM, Pranutha Sundeep wrote:
>
> Hello,
>
> I'm a PhD researcher at the University of Texas at Dallas, working on 
> cross-protocol time synchronization accuracy (NTPv4, NTPv5, PTP). 
> Through my work managing a production network, I've had the 
> opportunity to test this across a fleet of ~90 HPE Aruba AOS-CX 
> switches. As part of that work I've been running ntpd-rs as an 
> NTPv5-capable reference source (GMR6000 grandmaster -> ptp4l/phc2sys 
> -> ntpd-rs, source-less, local-stratum=1), and identified and fixed a 
> bug that seems worth sharing with the group.
>
> *Symptom: *a correctly configured source-less ntpd-rs instance 
> (local-stratum=1, clock authority is phc2sys, not an upstream NTP/NTS 
> source) correctly reports Stratum=1 to clients, but always emits LI=3 
> (Unsynchronized) and RefID="XNON", regardless of the configured 
> reference-id. This matches upstream issue #1718.
>
> *Root cause:* in ntp-proto/src/system.rs, 
> |NtpSnapshot::from_used_sources()| unconditionally initializes 
> |reference_id = ReferenceId::NONE|, only overwriting it if 
> |used_sources.peek()| finds an active source. It never falls back to 
> |synchronization_config.reference_id| when the source list is 
> legitimately empty (i.e., stratum-1 with an external clock authority). 
> Separately, |update_used_sources()| overwrites the entire snapshot on 
> every call, clobbering any one-time fix applied at startup in 
> |NtpManager::new()|. The LI bug is a related but separate path, in 
> |update_time_snapshot()|.
>
> *Proposed Fix:* thread a |configured_reference_id| parameter into 
> |from_used_sources()|; when |used_sources| is empty and |local_stratum 
> == 1|, initialize |reference_id| from 
> |synchronization_config.reference_id| instead of |ReferenceId::NONE|. 
> A parallel fix gates the LI value the same way.
>
> *Verification: *patched, rebuilt, and confirmed at the wire level -- 8 
> captured NTPv4 response packets to a production AOS-CX switch all show 
> LI=0 (NoWarning), Stratum=1, RefID='PTP ' as expected. On the switch 
> itself, the NTP association went from a permanent reach=0/.INIT. state 
> to reach=7 (climbing) with REF-ID=.PTP.. -- its first successful NTP 
> association in this deployment. Zero regressions observed.
>
> I'm happy to open a PR against ntpd-rs with this if useful.
>
> And on a separate note, a smaller observation regarding draft ID 
> mismatch. While testing against a real ntpd-rs NTPv5 server, I found 
> it identifies as "draft-ietf-ntp-ntpv5-08". A client sending a 
> non-matching draft ID (e.g. -09) has its packets silently dropped, 
> with the only trace being a TRACE-level log line ("Mismatched draft ID 
> ignoring packet!"). From an interop-robustness standpoint this seems 
> worth a note somewhere (e.g. a louder log level, or a 
> response/rejection rather than a silent drop), since it's easy for an 
> operator or implementer to mistake for a network issue.
>
> Happy to share more detail, packet captures, or the patch itself if 
> useful.
>
> Best,
>
> Pranutha Malickal
>
> PhD Researcher, Computer Engineering, University of Texas at Dallas
>
> Network and Security Manager, Dallas Theological Seminary, Dallas
>
>
> _______________________________________________
> ntp mailing list [e-mail]
> To unsubscribe send an email [e-mail]