On Thu, 11 Jan 2024 14:09:37 GMT, Sean Mullan <mul...@openjdk.org> wrote:

>> This enhancement simplifies and improves the performance of the Comparator 
>> that the PKIX CertPathBuilder uses to sort candidate certificates.
>> 
>> [RFC 5280](https://www.rfc-editor.org/rfc/rfc5280#section-4.2.1.1) requires 
>> that certificates include authority and subject key identifiers to 
>> facilitate cert path discovery. When the certificates comply with RFC 5280, 
>> the sorting algorithm is fast and efficient. However, there may be cases 
>> where certificates do not include the proper KIDs, for legacy or other 
>> reasons. This enhancement targets those cases and has shown an increase in 
>> performance of `CertPathBuilder.build` by up to 2x in tests involving 
>> certificates that do not contain KIDs. Specific changes include:
>> 
>> - Removed and simplified some of the steps in `PKIXCertComparator.compare` 
>> method. Some of these steps were not a good representation of common 
>> certificate hierarchies and were overly expensive to perform. 
>> - Several methods in `X500Name` and `Builder` have been made obsolete and 
>> thus removed.
>> - `X500Name` has been changed to use shared secrets instead of reflection to 
>> access non-public members of `X500Principal`, and vice-versa.
>> - The `CertificateBuilder` test code has been enhanced to set reasonable 
>> defaults for serial number and validity fields of a certificate
>
> Sean Mullan has updated the pull request incrementally with one additional 
> commit since the last revision:
> 
>   Add more comments. Remove unnecessary import.

Some additional information on this fix for future reference. I removed 3 steps 
from the Comparator:

     * 3) Issuer is a descendant of a trusted subject (in order of
     *    number of links to the trusted subject)
     *    a) Issuer: ou=E,ou=D,ou=C,o=B,c=A        [links=1]
     *    b) Issuer: ou=F,ou=E,ou=D,ou=C,ou=B,c=A  [links=2]
     *
     * 4) Issuer is an ancestor of a trusted subject (in order of number of
     *    links to the trusted subject)
     *    a) Issuer: ou=C,o=B,c=A [links=1]
     *    b) Issuer: o=B,c=A      [links=2]
     ...
     * 6) Issuer is an ancestor of certificate subject (in order of number
     *    of links to the certificate subject)
     *    a) Issuer:  ou=K,o=J,c=A
     *       Subject: ou=L,ou=K,o=J,c=A
     *    b) Issuer:  o=J,c=A
     *       Subject: ou=L,ou=K,0=J,c=A
     
These steps were removed because it assumes the PKI hierarchy is very rigid, 
and each intermediate CA is a direct child of it's parent in the DN namespace. 
More commonly, there is some structure, but not so exact.  For example, a CA 
with a subject DN of “CN=Root CA, O=OpenJDK, C=US” may issue a certificate to a 
sub-CA named “CN=JDK  CA, OU=JDK, O=OpenJDK, C=US”. Here the subCA is not a 
direct descendant of the root CA, but they both share a common namespace of 
"O=OpenJDK, C=US" (which will be found by step 5 which was retained).

-------------

PR Comment: https://git.openjdk.org/jdk/pull/17248#issuecomment-1887289722

Reply via email to