Hey Mike,
On Wed, 2006-01-18 at 13:49 -0600, Michael C Thompson wrote:
Hey all,
I'm not sure if anyone else has seen this, or if its been brought up
before (though I think
not), but I've discovered a problem with trying to have audit filter
on fields with negative
values. I suspect this is due to a difference in kernel space and user
space, given the
results I've seen below, but here are the particulars:
On zSeries and on xSeries, we have noticed that we are incapable (in
some situations) of
filtering messages when the filter value is negative. On zSeries, this
seems to be true for all
fields, while on xSeries, its true if the field is a1,a2,a3.
We have explicity tested -9 and -1, but I believe this code will
extend to all manner of
negative values because seems to be related to the representation of
these values in
the different architectures (32 v 64). I have not tested it on a
32-bit only platform, if someone
has the ability to that (should take all of 3minutes) that would
probably be useful :)
Below is all of my test information.
Thanks,
Mike
What kernel are you testing on? I just checked the latest kernel
(lspp.6) and this does look like a problem:
struct audit_field {
u32 type;
u32 val;
u32 op;
};
We only allow unsigned val(ues). Eek
-tim