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

On Fri, Sep 14, 2018 at 5:19 AM Paul Moore <paul(a)paul-moore.com&gt; wrote: > On Fri, Aug 24, 2018 at 8:00 AM Ondrej Mosnacek <omosnace(a)redhat.com&gt; 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. > > > > Next, it adds two functions to the audit interface: > > - audit_tk_injoffset(), which will be called whenever a timekeeping > > offset is injected by a syscall from userspace, > > - audit_ntp_adjust(), which will be called whenever an NTP internal > > variable is changed by a syscall from userspace. > > > > Quick reference for the fields of the new records: > > AUDIT_TIME_INJOFFSET > > sec - the 'seconds' part of the offset > > nsec - the 'nanoseconds' part of the offset > > AUDIT_TIME_ADJNTPVAL > > op - which value was adjusted: > > offset - corresponding to the time_offset variable > > freq - corresponding to the time_freq variable > > status - corresponding to the time_status variable > > adjust - corresponding to the time_adjust variable > > tick - corresponding to the tick_usec variable > > tai - corresponding to the timekeeping's TAI offset > > I understand that reusing "op" is tempting, but the above aren't > really operations, they are state variables which are being changed. I remember Steve (or was it Richard?) convincing me at one of the meetings that "op" is the right filed name to use, despite it not being a name for an operation... But I don't really care, I'm okay with changing it to e.g. "var" as Richard suggests later in this thread. > Using the CONFIG_CHANGE record as a basis, I wonder if we are better > off with something like the following: > > type=TIME_CHANGE <var>=<value_new> old=<value_old> > > ... you might need to preface the variable names with something like > "ntp_" or "offset_". You'll notice I'm also suggesting we use a > single record type here; is there any reason why two records types are > required? There are actually two reasons: 1. The injected offset is a timespec64, so it consists of two integer values (and it would be weird to produce two records for it, since IMO it is conceptually still a single variable). 2. In all other cases the variable is reset to the (possibly transformed) input value, while in this case the input value is added directly to the system time. This can be viewed as a kind of variable too, but it would be weird to report old and new value for it, since its value flows with time. Plus, when I look at: type=TIME_INJOFFSET [...]: sec=-16 nsec=124887145 I can immediately see that the time was shifted back by 16-something seconds, while when I look at something like: type=TIME_CHANGE [...]: var=time_sec new=1537185685 old=1537185701 type=TIME_CHANGE [...]: var=time_nsec new=664373417 old=789260562 I can just see some big numbers that I need to do math with before I get an idea of what is the magnitude (or sign) of the change. > > old - the old value > > new - the new value > > > > Signed-off-by: Ondrej Mosnacek <omosnace(a)redhat.com&gt; > > --- > > include/linux/audit.h | 21 +++++++++++++++++++++ > > include/uapi/linux/audit.h | 2 ++ > > kernel/auditsc.c | 15 +++++++++++++++ > > 3 files changed, 38 insertions(+) > > A reminder that we need tests for these new records and a RFE page on the wiki: > > * https://github.com/linux-audit/audit-testsuite I was going to start working on this once the format issues are settled. (Although I probably should have kept the RFC in the subject until then...)

> * https://github.com/linux-audit/audit-kernel/wiki I admit I forgot about this duty, but again I would like to wait for the discussions to settle before writing that up.

> paul moore

Ondrej Mosnacek <omosnace at redhat dot com>