Trouble with Bind/overlays/callbacks

[Howard Chu <hyc@symas.com>]

In slapd just about all of the operation handlers send their results from the 
backend directly to the client. The notable exception is for the Bind 
operation, which sends failures directly to the client, but has to return 
successes to the frontend to allow final processing to occur there. This 
difference causes a fair bit of complexity when writing Bind handlers in 
backends and overlays.

In 2.3 we ran into other sequencing issues that could only be solved by moving 
some of the Bind cleanup into a callback (connection_bind_cb). We have an 
opportunity now to move all of the postprocessing into that callback, allowing 
us to remove all special case handling of Binds in the backends and overlays. 
Of course, making this change will touch every backend that handles Bind 
operations. It may impact overlays that intercept Binds as well.

In particular, overlays that insert response callbacks on Bind operations can 
potentially break in the current code. I.e., if an overlay inserts a callback 
at the head of the stack (instead of at the end) then currently it will simply 
not get called in the case of a successful Bind, because the overlay framework 
tears down the callback stack before it returns to the frontend. If the 
callback was inserted using malloc'd memory then it will cause a memory leak 
as well.

I plan to file an ITS on this for RE24, not sure how soon I'll get to the 
actual changes though.
--
  -- Howard Chu
  Chief Architect, Symas Corp.  http://www.symas.com
  Director, Highland Sun        http://highlandsun.com/hyc/
  Chief Architect, OpenLDAP     http://www.openldap.org/project/