RFR: 8244336: Restrict algorithms at JCE layer

[Valerie Peng]

On Fri, 25 Jul 2025 19:45:04 GMT, Sean Mullan <mullan at openjdk.org> wrote:

>> This enhancement introduces a new security property "jdk.crypto.disabledAlgorithms" which can be leveraged to disable algorithms for JCE/JCA crypto services. For now, only Cipher, KeyStore, MessageDigest, and Signature services support this new security property. The support can be expanded later to cover more services if needed. Note that this security property is meant to disable algorithms irrespective of providers. If the algorithm is found to be disabled, it will be rejected before reaching out to provider(s) for the corresponding implementation(s).
>> 
>> A few implementation notes:
>> 1) The specified security property value is lazily loaded and all changes after it's been loaded are ignored. Invalid entries, e.g. wrong syntax, are ignored and removed. The algorithm name check is case-insensitive. If a disabled algorithm is known to has an object identifier (oid) by JDK, this oid and its aliases is also added to the disabled services.
>> 2) The algorithm name checking impl is based on the sun.security.util.AlgorithmConstraints class, but without the decomposing and different constraints.
>> 3) The hardwiring of NONEwithRSA signature to RSA/ECB/PKCS1Padding cipher in java.security.Signature class is removed. Instead, this is moved to the provider level, i.e. SunJCE and SunPKCS11 provider are changed to claim the NONEwithRSA signature support. Disabling one will not affect the other. 
>> 
>> CSR will be filed once the review is wrapping up.
>> 
>> Thanks~
>> Valerie
>
> src/java.base/share/classes/java/security/KeyStore.java line 1846:
> 
>> 1844:             for (Provider p : Security.getProviders()) {
>> 1845:                 for (Provider.Service s : p.getServices()) {
>> 1846:                     if (s.getType().equals("KeyStore") &&
> 
> I think the check should only be done if the code determines after probing the keystore that it is of the disabled type, so move this check down, probably just before line 1856.

No need to probe if the particular keystore is disabled seems more efficient? Anyhow, I made the suggested change as it has the merits of finding out the result of the probe through debugging.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/26377#discussion_r2241724785