Re: [PATCH ghak10 v5 1/2] audit: Add functions to log time adjustments

[Paul Moore]

On Thu, Sep 13, 2018 at 9:59 AM Ondrej Mosnacek <omosnace@xxxxxxxxxx> wrote:
> On Mon, Aug 27, 2018 at 6:38 PM Steve Grubb <sgrubb@xxxxxxxxxx> wrote:
> > On Monday, August 27, 2018 5:13:17 AM EDT Ondrej Mosnacek wrote:
> > > On Mon, Aug 27, 2018 at 9:50 AM Miroslav Lichvar <mlichvar@xxxxxxxxxx>
> > wrote:
> > > > On Fri, Aug 24, 2018 at 02:00:00PM +0200, Ondrej Mosnacek wrote:
> > > > > This patch adds two auxiliary record types that will be used to
> > > > > annotate
> > > > > the adjtimex SYSCALL records with the NTP/timekeeping values that have
> > > > > been changed.
> > > >
> > > > It seems the "adjust" function intentionally logs also calls/modes
> > > > that don't actually change anything. Can you please explain it a bit
> > > > in the message?
> > > >
> > > > NTP/PTP daemons typically don't read the adjtimex values in a normal
> > > > operation and overwrite them on each update, even if they don't
> > > > change. If the audit function checked that oldval != newval, the
> > > > number of messages would be reduced and it might be easier to follow.
> > >
> > > We actually want to log any attempt to change a value, as even an
> > > intention to set/change something could be a hint that the process is
> > > trying to do something bad (see discussion at [1]).
> >
> > One of the problems is that these applications can flood the logs very
> > quickly. An attempt to change is not needed unless it fails for permissions
> > reasons. So, limiting to actual changes is probably a good thing.
>
> Well, Richard seemed to "violently" agree with the opposite, so now I
> don't know which way to go... Paul, you are the official tie-breaker
> here, which do you prefer?

The general idea is that we only care about *changes* to the system
state, so if a process is setting a variable to with a value that
matches it's current value I see no reason why we need to generate a
change record.

Another thing to keep in mind, we can always change the behavior to be
more verbose (*always* generate a record, regardless of value) without
likely causing a regression, but limiting records is more difficult
and more likely to cause regressions.

-- 
paul moore
www.paul-moore.com