rpc_server: be more strict with the incoming assoc_group_id

Allow 0 and 0x12345678 only.
This fixes the RPC-HANDLES test.

metze
(This used to be commit c123e597cc84685abf2b0d3564e1a26d80bbef2f)

diff --git a/source4/rpc_server/dcerpc_server.c b/source4/rpc_server/dcerpc_server.c
index a2ca8979815a0a61c215445000f91aae70836407..ac36825acde4f8e3c6183fe0d04089e77c94cba6 100644 (file)
--- a/source4/rpc_server/dcerpc_server.c
+++ b/source4/rpc_server/dcerpc_server.c
@@ -543,20 +543,20 @@ static NTSTATUS dcesrv_bind(struct dcesrv_call_state *call)
        uint32_t context_id;
        const struct dcesrv_interface *iface;
 
-#if 0
-       /* It is not safe to enable this check - windows clients
-        * (WinXP in particular) will use it for NETLOGON calls, for
-        * the subsequent SCHANNEL bind.  It turns out that NETLOGON
-        * calls include no policy handles, so it is safe there.  Let
-        * the failure occour on the attempt to reuse a poilcy handle,
-        * rather than here */
-
-       /* Association groups allow policy handles to be shared across
-        * multiple client connections.  We don't implement this yet. */
-       if (call->pkt.u.bind.assoc_group_id != 0) {
+       /*
+        * Association groups allow policy handles to be shared across
+        * multiple client connections.  We don't implement this yet.
+        *
+        * So we just allow 0 if the client wants to create a new
+        * association group.
+        *
+        * And we allow the 0x12345678 value, we give away as
+        * assoc_group_id back to the clients
+        */
+       if (call->pkt.u.bind.assoc_group_id != 0 &&
+           call->pkt.u.bind.assoc_group_id != 0x12345678) {
                return dcesrv_bind_nak(call, 0);        
        }
-#endif
 
        if (call->pkt.u.bind.num_contexts < 1 ||
            call->pkt.u.bind.ctx_list[0].num_transfer_syntaxes < 1) {