RFC 5245, "Interactive Connectivity Establishment (ICE): A Protocol for Network Address Translator (NAT) Traversal for Offer/Answer Protocols", April 2010

Note: This RFC has been obsoleted by RFC 8445, RFC 8839

Note: This RFC has been updated by RFC 6336

Errata ID: 3619
Status: Rejected
Type: Technical
Publication Format(s) : TEXT
Reported By: Ashish Kundu
Date Reported: 2013-05-13
Rejected by: Gonzalo Camarillo
Date Rejected: 2013-05-15

Section Section 4, 5 says:

Missed candidate pair in ICE standard

It should say:

Scenario: X is caller, Z is callee. X is behind a non-full-cone 
(such as symmetric) NAT, Z is behind a full-cone NAT. 

ICE standard: Section 2.1 of RFC5245 describes the addresses that 
are collected as candidate addresses: (local address, server-reflexive 
address, TURN relay address).  For X: (X:x, X1:x1, Yx:yx), and for Z: 
(Z:z, Z1:z1, Yz:yz). 

Missed candidate pair in ICE standard: 
1. X:x sends a connection check message to the Z1:z1 (as part of the 
process in Section 2.2 of the standard) 
2. Since X is behind a non-full cone NAT such a symmetric one, NAT of 
X maps X:x to X2:x2, sends the message to Z1:z1
3. Z is behind a full-cone NAT, so packets received at Z1:z1 address 
is forwarded to Z:z by the NAT

Since X is behind a non-full cone NAT such a symmetric one and Z is 
behind a full-cone NAT, connection from X:x to Z1:z1 would be via a 
server-reflexive address X2:x2 of X, which is not a candidate address 
for X as specified by ICE. X2:x2 should be a candidate address of X, 
which however can only be determine when X sends a message to Z. The 
pair (X2:x2, Z1:z1) provides a direct connection option between X and Y.

Conditions on which X2:x2 is a valid candidate address:
1. One of the peers (Z) is behind a full-cone NAT, else step 3 above 
does not succeed.
2. X2:x2 is unique, i.e., different from X1:x1 (already covered by 
Section 2.1) if and only if one of the peers is  behind a non-full-cone 
NAT.

So I think there should be two stages in the candidate collection process:
A: Section 2.1 -- candidate addresses independent of the other clients
B: collection of the candidate pairs with respect to the peer, such as 
X2:x2 and Z2:z2, if any. 

B consists of the following steps including 1, 2, and 3:
4. Z:z determines if X2:x2 from which it received the message is a 
different address than in the candidate set of X.
5. If 4 is true, then send an OK message to X2:x2 that it received the
message with X2:x2 XOR-encoded.
6. X:x receives the OK message in 4, then X:x determines X2:x2 as its 
new candidate address.

If X:x decides to establish the connection via X2:x2, it sends ACK 
message to Z2:z2.

Notes:

This feedback for improvement of ICE candidate gathering and decision process was sent to Dr. Rosenberg on Nov 09, 2012. However, since I have not received any response from him over my next two followups and this e-mail, I thought it should be reported via this method.

This is not an error mesage, but a method to improve the candidate gathering and decision process of ICE.
--VERIFIER NOTES--

Report New Errata