linux-kernel - Re: 2.6.19-rc2 and very unstable NTP

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20061027113049.7236.qmail@science.horizon.com>
Date:	27 Oct 2006 07:30:49 -0400
From:	linux@...izon.com
To:	linux@...izon.com, zippel@...ux-m68k.org
Cc:	johnstul@...ibm.com, linux-kernel@...r.kernel.org
Subject: Re: 2.6.19-rc2 and very unstable NTP

> To analyze the kernel behaviour I would at least need the information fed 
> to the kernel, e.g. by tracing the ntp daemon via "TZ=utc strace -f -tt 
> -e trace=adjtimex -o ntpd.trace ..." and the ntp peer logs of same time 
> period. Kernel boot messages and config might help as well.

Actually, I just found it!  It's kind of a combination bug.

I had hacked NTP_MINPOLL to 2 in the ntpd source, and was using that
value on the pps source.  But ntpd would copy that value, minus 4,
into ntv.constant (and set ADJ_TIMECONST), and adjtimex() would complain
and return EINVAL.

It wouldn't accept the passed-in values, but would still fill in the
structure with the current values.

But ntpd didn't notice or log this error, and would log the *returned*
consstant.  Which was oddly dissociated from the expected one.

The failure to check the return code is an ntpd bug.

Now, in the kernel, I wonder what the allowed range should properly be.

I notice that on the input side, you have
	time_constant = min(txc->constant + 4, (long)MAXTC);
but on the output side, it's
	txc->constant      = time_constant;

This is clearly inconsistent and wrong, but which side should be fixed
is an interesting question I haven't got an answer to yet.

For now, I've just hacked the kernel to allow values of time_constant
down to 0 (txc->constant >= -4), as nothing will break with that
value.  (The only complicated one is ntp.c:296, and SHIFT_PLL is 4,
while SHIFT_NSEC is 12, so it all works.)

(Also, is MAXTC high enough for ntpv4?  I think it's in "-4" space,
so it should be time_constant = min(txc->constant, (long)MAXTC) + 4;)

That was an interesting piece of debugging... I was all over the map
before finding it.  And your suggestion of "strace -e adjtimex" was
indeed exactly what found it in the end.

So thanks!
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/