Hi Brad - The output of the DH calculation needs to preserve leading zeros.
The RFC unfortunately doesn't specify the Integer to Byte conversion primitive, but both X9.42 and the equivalent text in NIST SP800-56A (appendix C.1) make it clear that the conversion from Integer to Byte string ends up with a Byte string of length N where N is the smallest integer where 8^N >= the number of bits of the prime field. Those "RAW" bytes are the input into the KDF functions and have to exactly match to get a key output value that matches. (E.g. hash of "00 01 02 03" is different than the hash of "01 02 03" even though both represent the exact same integer). In any event, the RFC isn't controlling - the RFC is a profile of the actual standard X9.42. Mike At 02:00 AM 3/1/2012, Brad Wetmore wrote: >On 2/21/2012 5:33 PM, Valerie (Yu-Ching) Peng wrote: >>Brad, >> >>Can you please review the fixes for the following 2 bugs: >> >> * 7146728: Inconsistent length for the generated secret using DH key >> agreement impl from SunJCE and PKCS11 >> o http://cr.openjdk.java.net/~valeriep/7146728/webrev.00/ >> >> This impacts both SunJCE provider and SunPKCS11 provider. The >> implementations are inconsistent within SunJCE provider itself >> between the engineGenerateSecret() and >> engineGenerateSecret(byte[], int). Given that RFC 2631 specifies >> the leading 0s must be preserved so the generated secret has as >> many octets as the prime P, > >Just to be clear here, you're referring to Section 2.1.2 of 2631, which is >just one of the DH Key agreement variants (based on X9.42) for generating >Keying Material from secret keys obtained from a "raw" DH calculations, and is >then subject later SHA1 manipulations, right? This method provides >motivation/incentive to output our secret keys with the same lengths, but I >don't think this RFC makes any claims that the general output of "raw" DH key >agreement operation must be the same length. > >I'll take another look over the code tomorrow. > >Thanks, > >Brad > > > I have changed both SunJCE and >> SunPKCS11 provider to do so. When testing against Solaris and >> NSS libraries, Solaris preserves the leading 0s while NSS trims >> it off, thus similar handling is also needed in SunPKCS11 provider. >> >> * 7130959: Tweak 7058133 fix for JDK 8 (javah makefile changes) >> o http://cr.openjdk.java.net/~valeriep/7130959/webrev.00/ >> >> Instead of using the -Xbootclasspath, switching over to use >> -boothclasspath for consistency with the backported changes in >> the update releases for earlier JDK, e.g. 7u. >> >>Thanks, >>Valerie >>
