recursive mutexes in appl_head winbindd_cm.c?

[Martin Pool]

On 14 Jan 2003, jra at dp.samba.org wrote:
> get_mutex:
> 	ServerReqChallenge
> 	ServerAuthenticate2
> release_mutex:

Yes, that's what we meant.

I hypothesized to ab that in NT there is some kind of table
indexed by IP (or client name?) holding the challenges.  I wonder why?

<speculation>

Perhaps there's no per-tcp-connection data area easily accessible from
their RPC server, so when it sends out a challenge it can't easily be
associated with that TCP socket.  I'd guess that if it's a generalized
RPC server then the challenge can't be stored on the stack either.
The obvious thing, then, is to have a table shared between all threads
in the RPC server process, indexed by IP and storing the
currently-outstanding challenge.

This suggests an alternative implementation for us: rather than using
the mutex to make this a critical section, we instead have a
machine-wide tdb storing the outstanding challenge from each server.
Whoever gets around to authenticating next will remove the challenge
from the tdb and use that.

Mind you I can't see any way in which this is better than the current
approach, and it's probably more fragile.  :-)

--
Martin