The autobuild test system (on sn-devel-104) has detected an intermittent failing test in the current master tree.
The autobuild log of the failure is available here: http://git.samba.org/autobuild.flakey.sn-devel-104/2016-05-08-0803/flakey.log The samba build logs are available here: http://git.samba.org/autobuild.flakey.sn-devel-104/2016-05-08-0803/samba.stderr http://git.samba.org/autobuild.flakey.sn-devel-104/2016-05-08-0803/samba.stdout The top commit at the time of the failure was: commit 6379737b7ddc6ccb752238c5820cc62e76a8da17 Author: Uri Simchoni <u...@samba.org> Date: Thu May 5 23:40:22 2016 +0300 heimdal: encode/decode kvno as signed integer This patch changes the encoding/decoding of kvno (key version number) in blobs and packets to signed integer, for compatibility with Windows. Reportedly, MIT Kerberos does the same. This patch effectively reverts commit 1124c4872dfb81bec9c4b527b8927ca35e39a599 in the heimdal tree. According to the Kerberos spec (RFC 4120 5.2.9), the kvno field in encrypted data object is an unsigned integer that fits in 32 bits. The Heimdal Kerberos component bundled with Samba conforms to this. However, Windows deviates from the standard and encodes kvno as a signed integer, and this creates interoperability issues. ASN.1 DER has no special encoding for unsigned integer. A 32-bit unsigned integer is encoded as a signed integer, so while a signed 32-bit integer (covering the range of -0x80000000..0x7fffffff) is encoded using up to 4 bytes, an unsigned integer (covering 0..0xffffffff) could require 5 bytes. Normally, kvno for a given account starts at 1 and increments on password changes. Kerberos defined this as unsigned because there's no meaning for negative version numbers, so the standard writers figured 4 billion versions is better than 2 billion. It was not expected for a kvno to really go past 0x7fffffff and the disctinction usually does not matter. However, RODCs use kvnos which have the most-significant bit set. In Active Directory, RODCs have a private secret for the krbtgt, because the assumption is that the RODC is less secure, and recovering the domain krbtgt secret from the RODC would compromise the security of the entire domain. The kvno field is being used to identify the private krbtgt account that owns the key - the upper 16 bits are the RODC id, and the lower 16 bits identify the key version number for this specific RODC. It's common to have an RODC id greater than 0x8000, and therefore to have a kvno larger than 0x7fffffff, which would be DER-encoded using 5 bytes. Windows encodes kvno as signed integer - basically taking the 32 bits and treating them as a signed integer rather than an unsigned integer. This means that in Windows a kvno can always be encoded using 4 bytes, and Windows DCs reject a kvno encoded using more than 4 bytes without even generating an error response (the DC assumes it's an attack). Heimdal re-encodes the TGT when it creates a TGS request. Obviously it cannot decode and encode the encrypted parts but it does re-encode the plain parts, which include the kvno. That leads to a 5-byte kvno in the TGS request, which is rejected without an error response. BUG: https://bugzilla.samba.org/show_bug.cgi?id=11900 Signed-off-by: Uri Simchoni <u...@samba.org> Reviewed-by: Stefan Metzmacher <me...@samba.org> Reviewed-by: Ralph Boehme <s...@samba.org> Autobuild-User(master): Ralph Böhme <s...@samba.org> Autobuild-Date(master): Sat May 7 21:14:21 CEST 2016 on sn-devel-144