samba pdc with winbind has duplicate domain sam unix accounts
Problem has been found by LOCAL-NSSWRAPPER test on the buildfarm while testing master (problem is not discovered by this test in v3-5-test as "winbind enum users = yes" is not activated in v3-5-test for make test)
getent passwd | grep -i administrator
is what I get on a samba pdc with ldapsam and nss_winbind and "winbind enum users = yes".
(in my case I already had created some users, that is why the uid of the 2nd account is 10104, from a fresh install it may differ)
Created attachment 5365 [details]
smb.conf for my test
winbind on a dc running tdbsam has the same problem of course
I just very briefly tried to reproduce it (with master)
with tdbsam. But no success yet.
I might find time to test again later...
easiest way to demonstrate this is to fetch master with 2eb83f29648c7647cff4ac6eb38830ae35fcbcb5 including and call:
TESTS=posix_s3 POSIX_SUBTESTS="RPC-SAMR-LARGE-DC LOCAL-NSS-WRAPPER" make test
OK, reproduced it with Günther's new test in master.
It does not seem to be completely deterministic though.
As non-root I get 300 duplicate group entries, while
running as root, I get 2 duplicate entries for root.
This is on ubuntu 9.10, 64bit.
Günther seems to get different figures.
Volker has patches for this.
Created attachment 5396 [details]
patch by volker
proposed patch by volker.
While the patch seems to fix the issue for volker in a "real world environment", the make test run still fails on my box with the same dups.
Assigning to Günther for now, to review / decide...
0e9882a65e91e4ab5 in master fixes it for me.
Hmmm. Now WINBIND-STRUCT fails because we don't return anything in for GETPWENT :-(
I think we need to decide what make test is supposed to expect...
Phantastic! Thanks so much Volker for fixing this!
Karolin, please pick 0e9882a65e91e4ab5 for 3.5.0 final
We can make sure to pass WINBIND-STRUCT again later (the failure of that is based on mixed expectations between s3 and s4 winbind).
(In reply to comment #13)
> Phantastic! Thanks so much Volker for fixing this!
> Karolin, please pick 0e9882a65e91e4ab5 for 3.5.0 final
Pushed to v3-5-test.
> We can make sure to pass WINBIND-STRUCT again later (the failure of that is
> based on mixed expectations between s3 and s4 winbind).
Re-assigning to Günther for fixing the test.
Ok, WINBIND-STRUCT test fixed as well.
Hey, buildfarm is getting green again!