[cifs-protocol] [EXTERNAL] Windows Server 2025 PKINIT regression - TrackingID#2412190040009154

[Michael Bowen]

[DocHelp to bcc]

Hi Alexander,

Thanks for your question about Windows Server 2025 and Kerberos. I've created case number 2412190040009154 to track this issue, please leave the number in the subject line when communicating with our team. One of our engineers will contact you soon.

Best regards,
Michael Bowen
Sr. Escalation Engineer - Microsoft® Corporation

-----Original Message-----
From: Alexander Bokovoy <ab at samba.org>
Sent: Thursday, December 19, 2024 4:26 AM
To: Interoperability Documentation Help <dochelp at microsoft.com>
Cc: cifs-protocol at lists.samba.org
Subject: [EXTERNAL] Windows Server 2025 PKINIT regression

[Some people who received this message don't often get email from ab at samba.org. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]

Hi Dochelp,

I believe we are seeing a regression in how Windows Server 2025 handles Kerberos PKINIT, probably due to algorithm agility rewrite.

Sometime ago we have updated MIT Kerberos implementation of PKINIT to use sha256WithRSAEncryption in supported CMS types and removed sha1WithRSAEncryption to be able compliant with FIPS 140-3.

The commit https://github.com/krb5/krb5/commit/cbfe46ce20f3e9265baa9c648390148c739ab830
is part of MIT Kerberos 1.20 or later releases.

This change worked well for Windows Server versions prior to Windows Server 2025 release. With Windows Server 2025, the request is rejected (packet 8 from ad2025.pcap in attached archive):

Kerberos
    Record Mark: 106 bytes
        0... .... .... .... .... .... .... .... = Reserved: Not set
        .000 0000 0000 0000 0000 0000 0110 1010 = Record Length: 106
    krb-error
        pvno: 5
        msg-type: krb-error (30)
        stime: Dec 18, 2024 15:22:36.000000000 CET
        susec: 926640
        error-code: Unknown (79)
        realm: WIN2025-UO83.TEST
        sname
            name-type: kRB5-NT-SRV-INST (2)
            sname-string: 2 items
                SNameString: krbtgt
                SNameString: WIN2025-UO83.TEST


We built a custom version of MIT Kerberos which adds both sha256WithRSAEncryption and sha1WithRSAEncryption to the list of supported CMS types and still signed with sha256WithRSAEncryption, it failed again. The corresponding packet exchange can be seen in ad2025_sha1.pcap in the attached archive.

Both variants work against Windows Server 2019, so to us this looks like a regression in Windows Server 2025 implementation.

If this is not a regression and instead it is an intentional change, could you please make sure MS-PKCA and other corresponding documents get updated with a proper logic of the changes.

--
/ Alexander Bokovoy