Created attachment 9230 [details] proposed patch to fix this in master The code changes in the below patch did not make a special case for Deleted objects, and the inbound replication code did not correctly set up the invocationID (regression in the move to join.py). The result of both of these is that we break replication with Windows 2008R2 domains, and submit a corrupt object for replication into the domain. commit d3aad891c5759f66bd891cb47866d908a0562a8a Author: Andrew Bartlett <abartlet@samba.org> Date: Fri May 31 20:01:17 2013 +1000 dsdb: Prune deleted objects of links and extra attributes of replicated deletes When an object is deleted, the links to be removed are not propogated, you have to watch out for them manually! We do this by calling back into the originating update delete code(ie what is called if you ldb_delete() locally) so that any extra attribute found locally and not on the remote server becomes removed remotely too. We currently do the same with links, but that isn't strictly correct, but for now our getNCChanges server code filters these out, so only the usn is bumped. Andrew Bartlett Signed-off-by: Andrew Bartlett <abartlet@samba.org> Reviewed-by: Stefan Metzmacher <metze@samba.org>
Is there a chance to get review until tomorrow (freeze for 4.0.10)?
The patches are not yet finished. We now have the first set of dbcheck fixes, but more still needs to be finished.
To be clear: the issue is not in 4.0, only 4.1 (fortunately).
Created attachment 9240 [details] patches cherry-picked from master for 4.1 Complete patch series, including tests for 4.1. Compressed because it would be 41MB plain, due to the exported tdb.
Comment on attachment 9240 [details] patches cherry-picked from master for 4.1 xzcat <file> | git am applies it...
Pushed to autobuild-v4-1-test.
Pushed to v4-1-test (included in rc4). Closing out bug report. Thanks!