For a password modify operation (ldap modify or extop) we call new_passwdPolicy/delete_passwdPolicy 4 or more times (if all password checking is selected). I think it would be much more efficient to just call new_passwdPolicy once and stuff the policy object in the pblock, and have the different cases use the one in the pblock if present. This should work fine for all cases but the per user password policy case if that policy is being set at the same time as the password is being modified. In that case, we need to be very careful about the order of applying the change.
*** Bug 229290 has been marked as a duplicate of this bug. ***
more from Pete: A call to check_account_lock() (providing a slapi_entry) goes on to check password expiry etc. with a call to new_passwdPolicy() passing in the DN of the entry. new_passwdPolicy() then searches the db for the entry (that we already have). This occurs in the bind code path so any optimization here would be good. There should be a way to pass in the entry to avoid the search. new_passwdPolicy() seems to get called a lot in the password management code, and there look to be other places that might be able to optimize out the search. new_passwdPolicy() also has this comment /* If we're doing an add, COS does not apply yet so we check parents for the pwdpolicysubentry. We look only for virtual attributes, because real ones are for single-target policy. */ I'm not sure why this would be true - if an entry could be passed in in from the add code this shouldn't be an issue I think, and the whole policy searching algorithm could be ripped out and rely on COS for adds too (might be an issue with regards to whether the add op has succeeded yet or something).
Let's revisit this in 8.2 when we examine the password policy code
This falls under the category of "expose password policy to plug-ins"
Upstream ticket: https://fedorahosted.org/389/ticket/147
This was fixed in 389-ds-base-1.3.0.0-1.