Transcript Slide 1
Policies and Protocols for
Interoperability, Cyber-Security, and Resilience
GRIDSCHOOL 2010
MARCH 8-12, 2010 RICHMOND, VIRGINIA
INSTITUTE OF PUBLIC UTILITIES
ARGONNE NATIONAL LABORATORY
Dr. John C. Hoag
School of Information and Telecommunication Systems
Consortium for Energy, Economics and the Environment
Ohio University
[email protected] 877-660-4624
Do not cite or distribute without permission
MICHIGAN STATE UNIVERSITY
Themes of Today’s Talk
General design elements like protocols, methodologies, security
mechanisms, and control systems
Standards – organizations, relevant documents, status
Research and practice “gaps”
GridSchool 2010
Mitra - 02
Outline of first part, systems design
Interoperability and protocols
Security
Methodology
Conferral of degrees
GridSchool 2010
Mitra - 03
Interoperability
We start here because of statute. Cyber security (CS) is more
important – now and in the future.
Hardware interconnects; software interoperates.
End-to-End Principle: systems must work together regardless of
What connects them
Their underlying “platforms”
Without interoperability, we simply bridge systems. Security infers
that we compartmentalize subsystems anyway.
GridSchool 2010
Mitra - 04
Interoperation, demonstrating end-to-end nature
GridSchool 2010
Mitra - 05
Interoperation: concealing non-dependent features
GridSchool 2010
Mitra - 06
Principles for Interoperation
Protocols exist for every software layer. Their scope includes
Syntax, structure of message
Semantics, meaning of elements in context
Per layer, there is a PDU, Protocol Data Unit
Fixed or variable length
Numeric address scheme, global or local
Payload
Optional internal error detection/correction scheme
Protocols can be de facto or de jure; open or closed/proprietary
“Show Me” rule: Internet-related protocols must work before
consideration as a standard
GridSchool 2010
Mitra - 07
Interoperability Design Factoids
TCP/IP and OSI are common layered protocol architectures
Principles
Communications substrate
• Stable and reusable
• Embedded in network devices (router, switch)embedded
End-to-End “Transport” services
• Part of operating system, reside in computers (hosts)
• Know context, are necessary for reliable messaging
Applications
Dumb Networks prevail: intelligence resides in endpoints/hosts
GridSchool 2010
Mitra - 08
Security, Prelude
What is necessary and sufficient?
Firewalls are necessary but not sufficient
Cryptography is necessary but not sufficient
Security Policies (rules) are necessary but not sufficient
Concept of “perimeter” is not valid due to mobility
Are security standards (or certifications) necessary or sufficient?
GridSchool 2010
Mitra - 09
Security, snapshot today
Largest threats
DDoS: distributed denial of service attacks; e.g., botnets
Exfiltration of data
No platform is invulnerable; some have higher value as targets (HVT)
Many platform services needlessly run “privileged” as superuser, thus
targets of exploits. Virus authors know this!
Solution: turn off unnecessary services
Solution: run as little as possible as superuser
Solution: compartmentalize access to all resources
Solution: self-check integrity for unauthorized code and config mods
GridSchool 2010
Mitra - 010
Security Tools and Techniques
Network security has two verbs: permit and deny (traffic)
Based on source or destination address, protocol, etc.
Tools and techniques (very important)
Stateless (context-free) and stateful deductive filters
Intrusion detection and prevention inductive alerters/filters
Distributed detection (probes) and directed response
These expire after use:
• Preferred crypto system, PKI, based on public+private key pairing
• Strong, situational, multi-phase credential authentication
– by user AND device AND session
GridSchool 2010
Mitra - 011
Recognized Systems Development Methodologies
Cf. “Lifecycle”
“Waterfall” [Each phase has work products subject to review]
Analysis, requirements, and specification
Architecture (feasible technology choices)
Detailed Design
Implementation, Test, Acceptance; Operation and Maintenance
Cyclic
Emphasis on prototype
Add features incrementally
Underlying goal is risk avoidance
Note: Neither type begins with standards development!
GridSchool 2010
Mitra - 012
Outline of second part, Standards
Standards activities
Relevant Standards
Comments
GridSchool 2010
Mitra - 013
Standards Activities, Public
NIST, EISA (2007)
“…responsibility to coordinate development of a framework … for
interoperability of SG devices”
Cyber Security Coordination Task Group (now CSWG)
• Work product: NISTIR 7168, second draft February 2010
NERC: Bulk Power, 8 CIP Standards
DHS
CSWG: Control Systems WG, for SCADA
NPPD: National Protection Programs Directorate
• CS&C: Cybersecurity and Communications
– NCSD: National Cyber Security Division: CERT alerts; STORM
• IP: NIPP, 18 sectors
NSA: crypto; NOC-of-last-resort; contributed SELinux
USDoE National Labs
LBNL: OpenADR
GridSchool 2010
Mitra - 014
Standards Activities, Non-Public
Hierarchy
ISO
IEC.ch
• TC57 “Power systems mgmt and associated information exchange”
ANSI
• IEEE
– Committee 802, LAN networking including wireless
– Committee P1901, PLC, ISP-neutral
– Committee P2030, SG interoperability
EPRI UCA, Intelligrid
DNP.org: DNP3 protocol
Google.org PowerMeter
Manufacturer Alliances: HomePlug, ZigBee, Wi-Fi
GridSchool 2010
Mitra - 015
DNP
Evolution of MODICON, Ladder Logic, Programmable Load Controller
Effective for automation, closed loop control (PID)
Scope: substations, relays, etc., pre-”IED”
“outstation” PC is gateway from PLC nodes to network, SCADA
Upstream connections via dialup and LAN
Application-layer protocol of coded messages
Device addresses custom to utility
Parameter selection and values unique to device instance
Polled or event-driven traffic
Comment: can be kept confidential with integrity, subject to DoS
Comment: essentially implementation-specific, non-standard
GridSchool 2010
Mitra - 016
IEC 62351, Security Standards for TC57 Protocols
Rule: Not-enough expected processor power means the standard will
embrace weak authentication and weak encryption.
Part 3: TCP/IP traffic
End-to-End with authentication, well-encrypted TLS (SSL)
Part 4: ICCP messaging
Both secure (TLS) and non-secure nodes supported
Part 5: for IED/RTU class equipment: substations, relays, etc.
DNP3 TCP/IP LAN version, TLS but without authentication
• Symmetric/shared keying
• Single User per node (no concept of software multiplexer)
Serial version, not enough CPU for encryption; weak authentication
Part 6: for Peer-to-Peer comms including Web services (SOA)
App: relay control < 4ms, thus no delay allowed for encryption
Part 7: SNMP wannabe, lacked MIBs as of 2007
GridSchool 2010
Mitra - 017
Minimum protocol approach mandated
GridSchool 2010
Mitra - 018
SCADA Case Study, Thailand, 2008
Wiwat Amornimit, BSEE, Metro Electricity Authority, Bangkok
SCADA/DMS with some RTUs
Protocols
To RTUs/IEDs in substations for relays, meters
• IEC 60870-5 via LAN and serial comms
• DNP3, security governed by IEC 62351-5
To Control Centers (backup and one neighbor)
• Company-owned fiber running SONET; also TETRA radio
• IEC 60870-6 / TASE / ICCP (SSL-ish) over TCP/IP
GridSchool 2010
Mitra - 019
SCADA, Requirements and Expectations
Horowitz, et al., IEEE PES Magazine, March/April 2010
Future T&D systems to mitigate outage exposure based on better (but
sparse) VERY timely information, new logic, and advanced relays
Title: State Estimator, now one per control area
Incorporate new technologies:
Phase Monitoring Units (PMUs)
• Synchronized to 1 microsecond, sending timestamped info
• Should locate on 1/3 of system buses
Automatic calibration by PMUs to compensate for known errors
“Incomplete Observability” approach produces optimal model results
Comment: Not clear where to locate new “zone boundaries”
GridSchool 2010
Mitra - 020
NIST 7628 SG CS Strategy and Requirements
SGIP-CSWG (Interoperability Panel), 369 Members
Second draft, February 2010, 305 pages
Public Comments, 238pp, reduced herein to 10 slides
Individuals submitted comments on behalf of: public interest
groups (esp. regarding privacy), industry and trade associations,
utilities (esp. telcos), DHS, and Microsoft, among others.
Comments included: recitals, rants, requests, specific responses.
Comments ranged from merely editorial to overstatement.
No responses beyond North America
NIST prepared and published responses inline
• Important concerns were escalated
GridSchool 2010
Mitra - 021
NIST 7628 Comments (1): EEI
• The Draft NISTIR should be revised to make clear that it does not
create Smart Grid Cyber Security “requirements,” but rather is a
strategy document intended to facilitate the development of such
requirements through the SGIP/SGIPGB processes.
• Since the Use Cases in Appendix A are not comprehensive, this
appendix should be revised to make clear that the Use Case are
not mandatory. Furthermore, the Use Cases in Appendix A of the
Draft NISTIR should be revised to eliminate statements that imply
broad and general requirements. The following are some nonexhaustive examples of such statements that should be either
eliminated or qualified.
GridSchool 2010
Mitra - 022
More (2): EEI, referring to the Open SG on AMI
• “The most secure option would allow for one way
communications from the NAN to the HAN and not allow
data to flow from the HAN to the NAN.”
• The requirements identified in the OpenHAN SRS establish
the need for two way communications between the NAN and
HAN to meet the industry’s long term functional goals. The
addition of two way communications between the NAN and
the HAN introduces additional risk for unauthorized access to
the AMI system
• For these reasons, compartmentalization of the AMI system
and boundary protection should be employed
GridSchool 2010
Mitra - 023
More (3): NERC
…NERC notes that there is no such thing as full cybersecurity.
That is, there is no “cyber security model” that, once
achieved, will ensure full protection of the Smart Grid.
There currently is no cyber security model that will accomplish
complete security of the Smart Grid. Therefore, it is important
that NIST understand that there are different levels of
maturity involving the Smart Grid, and integration of new
parts and pieces into the Smart Grid could present cyber
risks because there is no industry-accepted cyber security
strategy.
GridSchool 2010
Mitra - 024
More (4): Verizon
The Smart Grid communications between the AMI meter and
the utility company should not include communications from
the home area network (HAN). The HAN’s untrusted
environment introduces additional vulnerabilities into the
communications infrastructure, and communications from HAN
to the utility company should not be permitted to be sent
through the meter.
The AMI meter will likely have limited processing capability and
not be able to provide the robust routing, filtering, and security
mechanisms that are necessary to protect the communications
channels and content.
GridSchool 2010
Mitra - 025
More (5): Verizon
…all AMI devices have a unique key to identify each device.
Including the same key on all devices from a single manufacturer
(and separately authenticating the messages) is insufficient to
protect against spoofing. Single-key schemes have been
demonstrated repeatedly to be weak against cryptographic attacks.
DNS services should not be used within the AMI system. Although
they are easy to use and implement, DNS services have a history
of being highly susceptible to compromise.
The AMI system must also be protected against unauthorized
changes to the software in the device. In other words, it must
employ some type of integrity protection or tamperproof
mechanism, and be able to fail to a known state should the
firmware be altered via unauthorized access.
GridSchool 2010
Mitra - 026
More (6): DHS NPPD NCSD CSSP
None of the communication protocols currently used (primarily
Distributed Network Protocol (DNP3) and sometimes International
Electrotechnical Commission (IEC) 61850) are typically
implemented with security measures, although IEC 62351 (which
are the security standards for these protocols) is now available but
implementation adoption and feasibility is not yet clear.
Many devices have no notion of a user or a role making security
management a challenge.
Many of the SCADA Masters may have no way to add security
without complete replacement
The information exchange requirements between the DMS and the
AMI head-end, except for outage information, are not known.
GridSchool 2010
Mitra - 027
More (7): IEEE
This requires a definition of what constitutes a security event in a
power system.
IEC62351-8 recommends time limited credentials which would be
one solution to the issue of revocation without a centralized security
manager.
"Many of the SCADA Masters may have no way to add security
without complete replacement". This is dangerously general and
probably not true..
Sections 3.5-3.11 ignore remote access connections for device
maintenance and configuration. This is a grave oversight.
GridSchool 2010
Mitra - 028
More (8): AT&T
RATHER THAN ALL INDIVIDUAL COMPONENTS PROVIDING
DEFENSE AGAINST DENIAL OF SERVICE, THE AMI ARCHITECTURE
SHOULD BUILD IN CAPABILITIES THAT DEFEND AGAINST
INDIVIDUAL METERS BEING COMPROMISED AND, SHOULD THAT
OCCUR, THAT STEPS CAN BE TAKEN TO ISOLATE THE BREACH
RE: AMI:
• THE TERM “OPERATIONAL SYSTEM BOUNDARY” AND THE
GENERAL TERM “BOUNDARY(IES)” ARE NOT DEFINED.
• THE PROTECTION SHOULD BE APPLIED TO THE NETWORK AND
THE APPLICATION LAYER.
• THE PATH IS ONLY ONE PART OF THE ENTIRE CONNECTION
AND THE PATH ITSELF MAY BE VIRTUAL OR PHYSICAL.
GridSchool 2010
Mitra - 029
More (9): ATT
NIST SHOULD TAKE STEPS TO CERTIFY ALGORITHMS FOR
SMALLER INTELLIGENT ELECTRONIC DEVICES (“IEDS”) WITH
PROCESSORS THAT HAVE LIMITED COMPUTING POWER AND
ARE NOT CAPABLE OF SUPPORTING ADVANCED
ENCRYPTION STANDARD (“AES”).
AUTHENTICATION AND INTEGRITY OF AMI DEVICE-TO-DEVICE
COMMUNICATION SHOULD BE PERFORMED AT THE
APPLICATION LAYER.
GridSchool 2010
Mitra - 030
More (10): Google.org PowerMeter
Comments to California PUC
“… We understand why there is a desire to delay HAN enabled
devices until the standard settles a bit more to help avoid stranded
assets - both for consumers as well as utilities. However,
consumers should not wait an indeterminate amount of time before
they start seeing the benefits of their investment in the smart grid.
The standards process for home area networking has made
significant progress since the Commission approved the AMI
rollouts. At least one nationally recognized standard for HANs has
already been released (Zigbee 1.0) and a second version (Zigbee
2.0) with improved networking and security protocols should be
available next year.
GridSchool 2010
Mitra - 031
More (11) USDOE INL
White paper, 2009
AMI Security, as it currently stands, is insufficient to protect the
national power grid from attack by malicious and knowledgeable
groups (Carpenter; Goodspeed), 2009.
Attackers can extract data from the memory of these devices
including keys used for network authentication and how the device
memory can be modified to insert malicious software. Once the
device is connected, it can be used to attack other parts of the SG by
communicating through the network. Attacks that originate with an
AMI wireless network device can lead to direct control systems
compromise.
GridSchool 2010
Mitra - 032
More (11) Microsoft
Previous attempts to bolt security on late in the lifecycle (after
development and deployment) have not worked in the past in
other sectors and will not work in the Smart Grid.
As technology is designed and developed for the Smart Grid is
it important that vendors and integrators learn from the body of
knowledge available such as secure software engineering
practices including developer training, organizational
processes, and tools.
GridSchool 2010
Mitra - 033
More (12) USDOE INL / Idaho
There must be a coordinated and ongoing effort to secure the SG
that includes the full development lifecycle. The development
lifecycle includes requirements, design, implementation,
verification, validation, procurement, installation, operations, and
maintenance. A failure in any phase of the lifecycle leads to
defects, which leads to vulnerabilities that can be exploited by a
skilled attacker.
GridSchool 2010
Mitra - 034
Third part, research and practice gaps
G1: Ad hoc methodology
G2: Current efforts not capturing SCADA requirements
G3: Current efforts not capturing security requirements
G4: Current product offerings
G5: Public policy, w.r.t. risk and consequential damages
GridSchool 2010
Mitra - 035
Black Swans (exist)
Black Swans and the Impact of the Highly Improbable, N.N. Taleb
Math Induction – it worked yesterday and today, it will tomorrow
• No, this demonstrates over-confidence in models
Software reliability is like mechanical reliability
• No, software flaws are latent and inert, waiting for activation
Rare events are Gaussian, distributed along a bell curve
• No, they follow a Power-Law, “heavy-tailed” distribution
Silent Evidence
Not the known knowns, or the unknown knowns, or even the known
unknowns …. but the unknown unknowns!
NIST’s critics may have registered a false negative
GridSchool 2010
Mitra - 036
Layers of Importance and Order in SG Design
GridSchool 2010
Mitra - 037
Where designers should focus
System State
Both distributed and hierarchical
Define modes of operation; parameters; state transitions
Address zones and locality, cf. micro and macro
THEN can develop information architecture, transaction protocols
End-to-end, authenticated transactions
Connection-oriented
Device, User, and session duration!
AMI/Retail mode
Separated, “air-gapped” from SCADA
Security operations
GridSchool 2010
Mitra - 038
Closed-loop and Open-loop Perspectives on SG
GridSchool 2010
Mitra - 039
Grand Challenges
At what hierarchic levels is system state to be maintained?
What are the variables of system state?
Can we extend the SG to the home without a set-top box (STB)?
Should regulators be concerned with…
Yet another network overbuild?
Retail SG (HAN) “essential” and prudent?
Obsolescence of devices?
Moral hazard?
Consequential damages?
GridSchool 2010
Mitra - 040
Special credit to Robert E. Burns, JD
Additional resources will be made available on this site:
http://www.its.ohiou.edu/grid
Contact Dr. John C. Hoag, [email protected] for more information
GridSchool 2010
Mitra - 041