Control Plane Resilience: The Method of Strong Detectio n Raj Kumar Vishal Misra Dan Rubenstein...

23
Control Plane Resilience: The Method of Strong Detection Raj Kumar Vishal Misra Dan Rubenstein Allerton, 9/28/06
  • date post

    20-Dec-2015
  • Category

    Documents

  • view

    223
  • download

    0

Transcript of Control Plane Resilience: The Method of Strong Detectio n Raj Kumar Vishal Misra Dan Rubenstein...

Control Plane Resilience: The Method of Strong

DetectionRaj Kumar

Vishal MisraDan Rubenstein

Allerton, 9/28/06

Routing Protocols with Misconfigurations

• Routing Protocols in “friendly” environments are well understood, e.g.,

– Link State: global knowledge, centralized approach

– Distance Vector (a.k.a. Bellman-Ford): known to converge (quickly), adapt to changes, etc.

– BGP (Path-Vector): some problems in converging when routes change, significant literature evaluating/understanding

• Critical Assumption for correctness: Nodes follow the proper protocol procedure

• Q: What happens when nodes don’t follow the protocol like they’re supposed to?

History Shows: Misbehaving nodes can be a big problem

• The infamous BGP AS 7007 Incident:

• Consider routes to node 8765

7007

5165

4345

7074

6957

2134

AS # 7007’s Distance

2134 2

4345 1

5165 3

6957 2

7074 1

… …

8765 8

8765

Traffic goes where it is supposed to

Nodes don’t always “behave”

• The infamous BGP AS 7007 Incident:

7007

5165

4345

7074

6957

2134

AS # 7007’s Distance

2134 1

4345 1

5165 1

6957 1

7074 1

… …

8765 1

8765

Traffic enters “black hole”

Theory to detect “Bad” Nodes

• Rules:– “Bad” nodes can cheat, “Good” nodes can attempt to detect the bad nodes

– “Good” nodes must limited to information provided by the routing protocol

• Want to exchange additional info, modify the protocol

• Challenge: When can a good node determine something isn’t right?

Can I tell if my neighbors are giving me the correct information?

A Node’s Info: Its State• A node’s state is its (only) view of the network

– e.g., Distance-Vector (a.k.a. Bellman-Ford)Dest/ Neighbor

A B E

A 0 1 12

B 1 0 7

C 7 13

8

D 5 9 6

E 9 6 0

F 12 15

13

G 4 9 2

A

B

C

E

G

F

D

Note our convention: (I,J) in state table reports node I’s distance to J (not local node’s distance to J through I)

Detection• Assume: Routes have stabilized (routing protocol inactive)

• Q: For routing protocol P, given a good node’s state, what misconfigurations can it detect/observe within the network?

• Note: A node can’t always detect a problem

N X Y1 1

D(X,Y) = 3N X Y

1 3

An undetectable misconfig at node N:

Prior Work: “Weak” Detection

• Process for constructing a weak detection method:

– Find a property that a node’s state should exhibit

– Check the property in a node’s state

– Declare misconfiguration in network if property is violated

• A detection method is “Weak” if it fails to identify a misconfiguration that is detectable using another method (on same state)

A Weak Detection Method: Symmetry

• In an undirected graph, D(X,Y) = D(Y,X)

– Here, D(A,B) = 1

– But D(B,A) = 4

• Using symmetry, found a misconfiguration

• So why is Symmetry weak?

Dest/ Neighbor

A B E

A 0 1 12

B 4 0 7

C 12 13

8

D 5 9 6

E 9 6 4

F 12 15

13

G 4 9 2

Another Weak Detection Method: Triangle Inequality

[DMZ’03]• Triangle inequality should

hold: D(X,Z) ≤ D(X,Y) + D(Y,Z)

• Violated here:

– D(B,E) = 3

– D(B,A) = 1

– D(A,E) = 1

– D(B,E) > D(B,A) + D(A,E)

• Note: symmetry property not violated

Dest/ Neighbor

A B E

A 0 1 1

B 1 0 3

C 12 13 8

D 5 9 6

E 1 3 0

• Example shows why detection via symmetry is weak: failed to identify a detectable misconfiguration

• So why is triangle inequality weak?

Weakness of Triangle Inequality

• Suppose graph edge lengths are all 1

• No violation of symmetry or triangle inequality

Dest/ Neighbor

A B

A 0 2

B 2 0

C 3 1

D 3 3

A

BC

C is distance 1 from B

D

D is distance 3 from both A & B: nowhere to put connecting edge

A and B are our neighbors

Where to place edges?

“Strong” Detection• A detection method is “strong” if it always detects detectable

misconfigurations

• More formally, Let

– μ be a method to detect misconfigurations

– C = N be the set of valid networks (what the network might look like)

– NR: the actual network (Note NR є C)

– sn(N) is state of node n when the routing protocol is executed correctly (and stabilized) within a network N є C

– s’n(NR) be the state actually computed at node n (possibly with misconfigurations) in network NR

• μ is a strong detection method if one of the following holds whenever s’n(NR) ≠ sn(NR):

– Detected: μ detects that sn(NR) ≠ s’n(NR)

– Undetectable: No method μ’ exists that can detect sn(NR)≠s’n(NR)

A High-Complexity Strong Detection Algorithm

• Input:

– State s’n(NR) of node n for the “real” but unknown network NR

– Description of set of allowable networks, C = N

• Algorithm: For each N є C

– Compute sn(N)

– If sn(N) = s’n(NR) then return MISCONFIG UNDETECTABLE (N might be the valid network)

• If no N є C matches, then misconfiguration detectedAlgorithm Complexity is Ω(C), often

huge!

Low-Complexity Strong-Detection

• Q: Can Strong Detection be achieved with low complexity?

• A: Sometimes: we show how to do it for Bellman-Ford (a.k.a. Distance Vector)

Strong Detection for D.V.• Input at node n:

– S’n(NR): a single node’s (steady state) state table that reports each neighbor’s (supposed) distance to all nodes

– Set C of all allowable networks

• defined by Axy: Axy is the set of allowable lengths of edges between node x and y

• E.g., Axy = [0,3) U [4,4] U [7,100]

Dest/ Neighbor

A B E

A 0 1 12

B 4 0 7

C 12 13

8

D 5 9 12

E 9 6 4

F 12 15

13

G 4 9 2S’n(NR)

Strong Detection in D.V. at a node, n

• Take node n’s state, s’n(NR)

• Use this state to build the canonical graph, M є C

• Simulate D.V. on M to generate simulated state sn(M)

• We will prove:

– If sn(M) ≠ s’n(NR), then misconfiguration detected

– Else, either there is no misconfiguration, or it is undetectable (using node n’s state) because M might be the actual network

Dest/ Neighbor

A B E

A 0 1 12

B 4 0 7

C 12

13

8

D 5 9 12

E 9 6 4

F 12

15

13

G 4 9 2

Dest/ Neighbor

A B E

A 0 1 12

B 4 0 7

C 12

13

8

D 5 9 12

E 9 6 4

F 12

15

13

G 4 9 2

s’n (NR)

sn(M)

G

A

B C

EF

D

G

A

BC

n E

F

D

M

n

Creating the Canonical Graph, M for an undirected

network• For each pair of nodes

(x,y):

– Create edge (x,y) with length exy = smallest value in Axy ≥ maxm є V(n) |d(m,x) – d(m,y)|

– exy = ∞ if all values in Axy too small

• Consider state table on left

– eCD ≥ max(|12-5|, |13-9|, |8-12|) = 7

– If ACD = [1,1] U [4,6] U [8,10], then eCD = 8

Dest/ Neighbor

A B E

A 0 1 12

B 4 0 7

C 12 13

8

D 5 9 12

E 9 6 4

F 12 15

13

G 4 9 2

Proving Strongness of the Canonical Graph Method

• N: a network for which sn(N) = s’n(NR), when such a network N exists

• M: the canonical graph constructed by n from s’n(NR)

• fxy: length of edge (x,y) in N (when the edge exists)

• exy: length of edge (x,y) in M (edges always exist)

• dG(x,y): shortest path distance from x to y in network G

• Assume: all edges have positive length (easy to extend when edges can also have length 0)

• High Level Sketch of Proof:

– If N exists where sn(N) = s’n(NR), then sn(M) = sn(N) = s’n(NR)

– If N does not exist, then sn(M) ≠ s’n(NR)

Bounds on exy

• Lemma 1: If sn(N) = s’n(NR) for some N є C and edge (x,y) exists in N with length fxy, then exy ≤ fxy

• Proof: In N, x & y’s distances to any neighbor v must differ by at most fxy, i.e.: For each neighbor v, |dN(v,y) – dN(v,x)| ≤ fxy

• Hence maxm є V(n) |d(m,x) – d(m,y)| ≤ fxy

• Recall exy = smallest value in Axy ≥ maxm є V(n) |d(m,x) – d(m,y)|

• Since N є C, we have fxy є Axy and so exy ≤ fxy

n xyfxy

v

• Lemma 2: If sn(N) = s’n(NR) for some N є C, then dM(v,x) ≤ dN(v,x) for all neighbors v and all nodes x

• Proof: by contradiction. Select x with smallest dN(v,x) where dM(v,x) > dN(v,x)

• Let y be a preceding node on a shortest path from v to x in N: fxy is the edge connecting y to x on this path, so dN(v,y) < dN(v,x) and |dN(v,x) – dN(v,y)| = fxy

• dN(v,y) < dN(v,x), hence y not blue dM(v,y) ≤ dN(v,y), so dM(v,y) ≤ dN(v,y) < dN(v,x) < dM(v,x)

• fxy = |dN(v,x) – dN(v,y)| < |dM(v,x) – dM(v,y)| ≤ exy

v x

Distance from v in N

yfxy

Blue nodes t satisfy dM(v,t) > dN(v,t)n

Contradicts Lemma 1 (which states exy ≤ fxy)!!

exyxy

in M:

• Lemma 3: If sn(N) = s’n(NR) for some N є C, then dM(v,x) ≥ dN(v,x) for all neighbors v and all nodes x

• Proof: by contradiction. Select x with smallest dM(v,x) where dM(v,x) < dN(v,x)

• Let y be the node preceding x on a shortest path from v to x in M where edge exy connects y to x on this path: hence dM(v,y) < dM(v,x) and exy = dM(v,x) - dM(v,y)

• dM(v,y) < dM(v,x), hence y not blue dM(v,y) ≥ dN(v,y)

• Hence exy = dM(v,x) - dM(v,y) < dN(v,x) - dN(v,y) = | dN(v,x) - dN(v,y) |

v x

Distance from v in M

Blue nodes t satisfy dM(v,t) < dN(v,t)y

exyn

xyin N:

But exy = maxm |dN(m,x) – dN(m,y)|, and

maxm |dN(m,x) – dN(m,y)|≥ |dN(v,x) – dN(v,y)| !! Contradiction!

The Main Result

• Some N є C produces state sn(N) = s’n(NR) sn(M) = s’n(NR)

• Proof:

Follows from Lemma 2 (dM(v,x) ≤ dN(v,x)) and Lemma 3 (dM(v,x) ≥ dN(v,x))

If no N є C produces state s’n(N), since M є C, M cannot produce state = s’n(N)

• In other words, only need to check if sn(M) = s’n(NR)

• Complexity: O(|V|3)

– Construct the canonical graph, M

– Simulate Bellman-Ford

– Compare State Tables

Extensions / Future Directions

• Same idea works for:

– Directed graphs

– Using state info from a set of trusted nodes

• Similar canonical graph construction works for path-vector variants

• Future Directions:

– Identifying the offending node (not just its existence)

– Performing Strong Detection for other routing protocols (Ad-hoc network, geographical positioning)