What is worrying me in this conversations is the discrepancy between the stated aim in draft-brockners-proof-of-transit and the function provided by the draft. I'm not (at the moment ;-) picking hols in what the S**3 scheme achieves, but I am questioning whether it provides enough to be useful in the context stated. So draft-brockners-proof-of-transit says... > This document defines mechanisms to > securely prove that traffic transited the defined path. and > In certain cases regulatory > obligations or a compliance policy require operators to prove that > all packets that are supposed to follow a specific path are indeed > being forwarded across and exact set of pre-determined nodes. Shamir as applied in this document does not meet those criteria. It only shows that all members of a defined set of nodes were traversed at some point along the path. It neither guarantees ordering, nor does it protect against the insertion of additional nodes on the path. Other parts of the document are more clear about exactly this point. it is not clear to me, therefore, whether the intention is misstated in the text I quoted, or whether the proposed solution (while good at what it does) is inadequate for the purpose. BTW, think it is not clear in the text that the order of the polynomial must count the verifying node. I.e., the verifying node must also insert a point on the curve. This is needed because otherwise the full set of points is included in a message on the wire and the polynomial becomes vulnerable. Thanks, Adrian From: OPSAWG [mailto:[email protected]] On Behalf Of Tal Mizrahi Sent: 20 July 2016 05:30 To: Sashank Dara (sadara); Frank Brockners (fbrockne); [email protected] Cc: [email protected]; [email protected]; [email protected] Subject: Re: [OPSAWG] Question regarding Proof of Transit draft Hi Sashank, >[SD] The attack is valid only if the attacker can get away bypassing a service function/node. >For example, if the attacker bypasses a node and if POT determines it did not bypass is a valid attack on the system. I would phrase it this way: *Given* a packet that did not go through its desired path, the attacker can easily make you think that it *did* go through the desired path. Sounds like a very significant vulnerability. If a packet did not go through the firewall (for one reason or another), the attacker will make you think that it did go through the firewall. Cheers, Tal. From: Sashank Dara (sadara) [mailto:[email protected]] Sent: Wednesday, July 20, 2016 4:31 AM To: Tal Mizrahi; Frank Brockners (fbrockne); [email protected] Cc: [email protected]; [email protected]; [email protected] Subject: Re: Question regarding Proof of Transit draft The POT replacement attack (1.) is not an attack on the integrity. It is an attack on the path verification. This simple attack can cause the verifier to accept a packet that did not go through the firewall SF (even though it should). I believe this is exactly the problem you were aiming to address in this draft. [SD] The attack is valid only if the attacker can get away bypassing a service function/node. For example, if the attacker bypasses a node and if POT determines it did not bypass is a valid attack on the system. In the current state, there is no way an attacker can get away as we determine the exact path the packet travelled (aim of the draft) . I reiterate that the verifier needs to handle what to do with the path reconstructed ! We could emphasize this in our next draft , but it would be beyond scope of POT to determine what to do with the path constructed. IMO, It would be highly application specific. Regards, Sashank Thanks, Tal. From: Frank Brockners (fbrockne) [mailto:[email protected]] Sent: Tuesday, July 19, 2016 6:00 PM To: Tal Mizrahi; Sashank Dara (sadara); [email protected] Cc: [email protected]; [email protected]; [email protected] Subject: RE: Question regarding Proof of Transit draft Hi Tal, thanks for the summary. We'll provide more details on 2. Per my earlier point - 1. is an interesting discussion, given that we don't claim to provide integrity protection for the packet payload. Or in other terms - to be exact: What POT provides is a proof that the POT-header/meta-data transited all the required nodes. There is no association (and thus proof) provided for the additional data carried along with the POT-header - neither header nor payload. As a consequence, attacks which change the packet payload won't be detected/mitigated. We'll explicitly state this in the security considerations in the next rev of the document. What we could consider is linking the RND number to CRC across the packet payload or similar - but that way we'd restrict the applicability to deployments where the packet payload isn't changed across the path (which might not apply to certain deployment - e.g. WAN optimization / compression schemes). Do you think it is worthwhile to provide a solution for a deployment which is expected to not alter the packet payload? Thanks, Frank From: Tal Mizrahi [mailto:[email protected]] Sent: Dienstag, 19. Juli 2016 17:44 To: Sashank Dara (sadara) <[email protected]>; Frank Brockners (fbrockne) <[email protected]>; [email protected] Cc: [email protected]; [email protected]; [email protected] Subject: RE: Question regarding Proof of Transit draft Hi, To summarize my take on this thread: The proposed mechanism has two significant vulnerabilities that (in my understanding) are currently not addressed: 1. A man-in-the-middle can replace the POT of packet A with the POT of packet B. 2. It is possible to replay POTs within a certain time window, whose length is determined by the timestamp resolution. Sashank, thanks for agreeing to look into it further. I am looking forward to your insights on this. Regards, Tal. Link to the draft: https://tools.ietf.org/html/draft-brockners-proof-of-transit-01 From: Sashank Dara (sadara) [mailto:[email protected]] Sent: Tuesday, July 19, 2016 12:20 PM To: Tal Mizrahi; Frank Brockners (fbrockne); [email protected]; [email protected] Subject: Re: Question regarding Proof of Transit draft I want to ask a simple question: If the attacker attaches the POT of packet A (indicating the path through 1,3,5,6) to packet B, will the verifier accept packet B and believe that its path was indeed (1,3,5,6)? [SD] If the verifier is programmed to just validate the POT meta data against {1,3,5,6} then yes it accepts it. If the verifier is programmed to consult a policy database to cross check if the reconstructed path {1,3,5,6} is as per the policies then no , it drops it . But I see your point , that the parameters used in POT data donot consider the path or node-ids etc . We shall discuss this internally and get back. Also, We shall get back with more concrete numbers of the timestamp resolution and cache sizes (or other better approaches). Thank you so much for all the inputs. From: Tal Mizrahi Sent: Tuesday, July 19, 2016 10:28 AM To: 'Sashank Dara (sadara)'; Frank Brockners (fbrockne); [email protected]; [email protected] Subject: RE: Question regarding Proof of Transit draft Dear Sashank, I really appreciate the quick and detailed responses. >>>Lets take correct path taken by Packet A to be Path1 - ( 1,3,5,6) nodes. >>>Lets assume incorrect path to be Packet B i.e. Path2 - (1,2,3,6). >>> >>>If the attacker could take values from Path1 and reattach them to Path2 , >>>the reconstruction for PacketB would result in (1,3,5,6) instead of (1,2,3,6). >>>This could be compared with topology/policy db information for any policy >>>violations. POT does not enforce a particular path to be taken. >>So packet B skipped node 5 (e.g. the firewall SF), but the verifier believes it went through the correct path. >>Would you agree? >[SD] The verifier only constructs the path the packet took accurately, it is upto >the application to determine whether it violates any policies (I.e missing any function) >The bottom line is , an attacker taking a different path cannot get away with it ! >The whole intent of "proof-of-transit" is to prove the path packet has taken exactly. >POT does not define whether it good/bad path. It is upto high level applications on what >do with "reconstructed" path. I want to ask a simple question: If the attacker attaches the POT of packet A (indicating the path through 1,3,5,6) to packet B, will the verifier accept packet B and believe that its path was indeed (1,3,5,6)? >>Hmmm. Let's say we are talking about traffic at 10 Gbps, which is roughly >>1 million packets per second. That would mean the verifier has to store a million values of RND-2. >>Moreover, every time a packet arrives, the verifier would have to compare its >>RND-2 value with the 1 million stored values, in order to verify there is no replay. >>That does not sound feasible. >>Am I missing something here? >[SD] Second level precision is only an example, we could use as much precision as we want to reduce the number of values to be cached ! My point is that assuming reasonable resources are used, the mechanism is vulnerable to a replay attack. In order to prove me wrong, can you present concrete numbers of the timestamp resolution and the cache size? Otherwise, you may consider using a sequence number + sliding window (e.g., as in IPsec). Thanks, Tal. From: Sashank Dara (sadara) [mailto:[email protected]] Sent: Tuesday, July 19, 2016 9:56 AM To: Tal Mizrahi; Frank Brockners (fbrockne); [email protected]; [email protected] Subject: Re: Question regarding Proof of Transit draft >[SD] Ok. This is an interesting attack. > >Lets take correct path taken by Packet A to be Path1 - ( 1,3,5,6) nodes. >Lets assume incorrect path to be Packet B i.e. Path2 - (1,2,3,6). > >If the attacker could take values from Path1 and reattach them to Path2 , >the reconstruction for PacketB would result in (1,3,5,6) instead of (1,2,3,6). >This could be compared with topology/policy db information for any policy >violations. POT does not enforce a particular path to be taken. So packet B skipped node 5 (e.g. the firewall SF), but the verifier believes it went through the correct path. Would you agree? [SD] The verifier only constructs the path the packet took accurately, it is upto the application to determine whether it violates any policies (I.e missing any function) The bottom line is , an attacker taking a different path cannot get away with it ! The whole intent of "proof-of-transit" is to prove the path packet has taken exactly. POT does not define whether it good/bad path. It is upto high level applications on what do with "reconstructed" path. Hmmm. Let's say we are talking about traffic at 10 Gbps, which is roughly 1 million packets per second. That would mean the verifier has to store a million values of RND-2. Moreover, every time a packet arrives, the verifier would have to compare its RND-2 value with the 1 million stored values, in order to verify there is no replay. That does not sound feasible. Am I missing something here? [SD] Second level precision is only an example, we could use as much precision as we want to reduce the number of values to be cached ! Sashank From: Sashank Dara (sadara) [mailto:[email protected]] Sent: Tuesday, July 19, 2016 8:33 AM To: Tal Mizrahi; Frank Brockners (fbrockne); [email protected]; [email protected] Subject: Re: Question regarding Proof of Transit draft Inline .. We have two consecutive packets, A and B: - Packet A that went through the correct path and has a correct POT. - Packet B that did not go through the correct path and does not have a correct POT. Now the attacker performs a 'mix and match' attack, by taking the correct POT from packet A and attaching it to packet B. The attacker also terminates packet A. There is no replay here, because the verifier receives the correct POT only once. The problem is that the correct POT happens to arrive with packet B. L Thus, packet B appears okay to the verifier, even though it did not go through the correct path. Is there a way to mitigate this attack? [SD] Ok. This is an interesting attack. Lets take correct path taken by Packet A to be Path1 - ( 1,3,5,6) nodes. Lets assume incorrect path to be Packet B i.e. Path2 - (1,2,3,6). If the attacker could take values from Path1 and reattach them to Path2 , the reconstruction for PacketB would result in (1,3,5,6) instead of (1,2,3,6). This could be compared with topology/policy db information for any policy violations. POT does not enforce a particular path to be taken. It just reconstructs the particular path taken by the packets. Do you mean that there is a one-second-vulnerability for replay attacks? One second is practically forever. [SD] Not really , the verifier could cache the RND-2 numbers used in the time slice of one second and flush off after every second. There is no one-second-vulnerability as such, if the verifier caches those values. Effective replay prevention is typically performed using a sequence number, with a sliding window to allow out-of-order packets. Wouldn't it be possible to incorporate such a sequence number into your mechanism? [SD] Time stamp is essentially at high level sequence number (at seconds level)! The challenge with using complete RND-2 as sequence number is that the differential analysis of CML values (across packets) becomes very easy and predictable. that's reason we recommend using ( Timestamp(/ sequence number) + RND ) From: Sashank Dara (sadara) [mailto:[email protected]] Sent: Tuesday, July 19, 2016 1:11 AM To: Tal Mizrahi; Frank Brockners (fbrockne); [email protected]; [email protected] Subject: Re: Question regarding Proof of Transit draft Dear Tal, Thank you for your interest in our work. More inline with [SD] .. Right, I am referring to the former. The integrity check is not the issue. Let's say we have: - Packet A that went through the correct path and has a correct POT. - Packet B that did not go through the correct path and does not have a correct POT. The attacker can 'launder' packet B by replacing the (incorrect) POT of packet B with the correct POT of packet A (and drop packet A). Thus, packet B is verified correctly, even though it did not go through the correct path. [SD] This is valid scenario and we have certain in-built mitigation techniques and recommendations for the same. There are two scenarios here Partial Replay of POT data (Replaying intermediate CMLs) Attacker cannot reuse few intermediate node's POT values (from older packet traces) as it would disrupt the POLY-3 construction. On the other hand if the attacker tries to replay entire sequence of CML values, he cannot still , because notice that verifier also has a secret share of RND-1 and participates in the reconstruction of RND-3. Verifier's share of RND-3 is never on wire , so attacker cannot just observe the packet traces to reuse and replay it. Total Replay of POT data (Replaying complete RND-2) Another trick, attacker could do is by completely reusing RND-2 (but not intermediate CML values) In order to prevent the "Reply Attacks" , we recommend that the RND-2 (generated per packet) be a combination (I.e. RND2 = "Time Stamp + RND number") So in case a passive attacker tries to replay one of the correct but older RND-2, the verifier first could check the current timestamp against the timestamp retrieved from RND-2. If the retrieved timestamp is older than current timestamp we drop/raise a flag ! There is still a small window , where the attacker could replay it within the valid time slice itself but by carefully choosing the time slice we can make it nearly impossible for the attacker to replay. For example , if the Time Stamp chosen is 32 bits , we could get one seconds level precision in the time window . So it is highly impossible for the attacker to replay within such a small time at such a high packet rates. Also , the verifier could cache, if needed, certain number of previously used RND-2 to mitigate replay attacks. Hope this clarifies.
_______________________________________________ OPSAWG mailing list [email protected] https://www.ietf.org/mailman/listinfo/opsawg
