Fake leap seconds being announced by NTP servers, still causing linux crashes | Hacker News Reader

Fake leap seconds being announced by NTP servers, still causing linux crashes

lists.ntp.org

51 points0x013 comments

No need to panic; this event happend a few days ago, August 1. And no one's certain, but it seems just as likely this is a bug in NTP as it is a deliberate attempt to crash unpatched Linux servers. The way ntpd manages the leap second flag is pretty complex and error prone.

Link to thread view of NTP list discussion: http://lists.ntp.org/pipermail/questions/2012-August/thread....

From http://lists.ntp.org/pipermail/questions/2012-August/033626....

There are over 10 stratum one's that advertise LI=1 as of Wed Aug 1 14:18:51 UTC 2012. Unless this changes another false leap second could occur on August 31, 2012

So yes this was happening on July 31st but may happen as well on August 31.

There is a need to worry, because AFAIK at least Debian hasn't provided patches yet, so come September servers might still be crashing.

http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=679882

The fix needs backporting large patchset, yet the next leap second isn't announced. So no need to rush... just wait for a while and do QA. Typical mentality.

An idea: Setup a fake NTP server that inserts leap seconds randomly (for example, once a week) (forward and backward). It could be use with test servers to test if they are working correctly, before a real leap second crash the production server.

Leap seconds only occur at the end of a month, so an NTP client that accepted such leapseconds would be non-compliant.

Exactly how incompetent are linux coders? After all, the "problem" is extremely trivial.

Feel free to patch it, I'm sure the world of linux coders would be happy to have someone who finds this sort of matter trivial in their midst.

Well, I'm not a linux coder, but I'll give it a try and report back if I'm successful.

I think it's already been patched. https://lkml.org/lkml/2012/7/17/546 (navigation on the left of the page)

Looks like a long, 11-part patch set. Would be interesting to hear back from 'bjork' after his try to see if it is still trivial ;)

https://lkml.org/lkml/2012/7/1/25 Sounds like a race condition which is pretty awkward to avoid.

DateTime code bugs happen all the time, there are so much things to consider and it's hard to test, because the problem conditions might not occur for years to come. Remember this? http://news.cnet.com/new-years-hangover-for-zune-users/ (also I too am guilty of producing a similar bug in a software a few years ago)

Reply on news.ycombinator.com