Visiting Speaker Talk
— 2:30pm
Location:
Panther Hollow Conference Room 4101
-
Collaborative Innovation Center
Speaker:
Deepak Garg
,
Professor, MPI-SWS
https://www.mpi-sws.org
Runtime enforcement of dynamic policies in data-oriented systems
Speaker: Deepak Garg
Professor, The Max Planck Institute for Software Systems
Location: CIC, Panther Hollow Conference Room 4101
Runtime Enforcement of Dynamic Policies in Data-oriented Systems
Many services today collect, index, process and serve sensitive user
data. It is in the interest of both service providers and users that
data, whether that collected directly or derived indirectly, always be
used in accordance with applicable privacy polices. However, ensuring
bug-free compliance with policies without imposing significant runtime
overhead is very difficult in practical, large-scale software
systems. To start, most relevant policies are dynamic, meaning that
they depend on runtime information, so purely static enforcement is
difficult, if not impossible. Other methods like fine-grained
taint-tracking using hardware or software techniques have
prohibitively high overhead.
In this talk, I will describe an approach to enforcing dynamic
policies end-to-end by relying on coarse-grained runtime
monitoring. The idea comes from coarse-grained information flow
control---rather than examine code statically or monitor code
execution at runtime, we monitor process I/O at runtime. Using this,
we approximate data flows coarsely, and then enforce access and
declassification policies, end-to-end. This is precise enough for
practical data retrieval pipelines, and has only low to moderate
overhead. We reduce the runtime overhead even further by using system
lean periods to predictively check policies along (coarse-grained)
flows that may arise in the future. Our policy enforcement is grounded
in a customized distributed monitoring algorithm for linear temporal
logic (LTL).