masarati@aero.polimi.it wrote:

No. Some overlay along the lines apparently mucks with e_private in the entry that back-ldap is sending. It's probably slapo-pcache.

You're right. Following the debug traces, it's merge_entry which is passing the entry it got from the proxy directly to be_add(). But bdb_add() takes the given entry and inserts it directly into its entry cache, instead of calling entry_dup(). Since this particular entry came from back-ldap's stack, this whole chain of events is doomed to fail.

I'll fix this by calling entry_dup() in pcache merge_entry().