On 11/9/20 10:23 AM, John Wood wrote:

Hi,
Thanks for the typos corrections. Will be corrected in the next patch
version.

On Sun, Nov 08, 2020 at 08:31:13PM -0800, Randy Dunlap wrote:

quoted

So an app could read crash_period_threshold and just do a new fork every
threshold + 1 time units, right? and not be caught?

Yes, you are right. But we must set a crash_period_threshold that does not
make an attack feasible. For example, with the default value of 30000 ms,
an attacker can break the app only once every 30 seconds. So, to guess
canaries or break ASLR, the attack needs a big amount of time. But it is
possible.

So, I think that to avoid this scenario we can add a maximum number of
faults per fork hierarchy. Then, the mitigation will be triggered if the
application crash period falls under the period threshold or if the number
of faults exceed the maximum commented.

This way, if an attack is of long duration, it will also be detected and
mitigated.

What do you think?

Hi,
That sounds reasonable to me.

thanks.
-- 
~Randy