are any problems in a validation, a SERVFAIL will be returned. This may not
be correct in all cases (and will be fixed), but it leaves the server in a
much more consistent state after failures.
dns_dispatch_create() no longer exists. dns_dispatch_createtcp()
and dns_dispatch_getudp() are the replacements. _createtcp() takes
a bound, connected TCP socket, while _getudp() will search for
a sharable UDP socket, and if found, attach to it and return a
pointer to it. If one is not found, it will create a udp socket,
bind it to a supplied local address, and create a new dispatcher
around it.
dns_dispatch_remove{request,response}() no longer take the dispatch
as an argument.
query-source can now be set per view.
The dispatch manager holds onto three memory pools, one for
allocating dispatchers from, one for events, and one for
requests/replies. The free list on these pools is hard-coded,
but set to 1024. This keeps us from having to dig into the
isc_mem_t the pools draw from as often.
dns_resolver_create() and dns_view_createresolver() require that
valid dispatchers be passed in; dispatchers are no longer created
for the caller.
should have been set to NULL when the top of a level was being rotated but
instead it was set to the node which pointed down to the level of the deleted
node.
rotate_left and rotate_right no longer take a parent parameter, since they
don't need it with the existence of parent pointers.
dns_rbt_deletefromlevel now takes a pointer to the pointer of the root of
the level as a parameter so that it doesn't have to rediscover the root that
its caller (dns_rbt_deletenode) already discovered.
dns_rbt_deletefromlevel did some (minor) pointless work with the sibling of the
deleted node before the color fixup loop was entered; it does so no more.
forward function declarations changed to ISC style.
involved extensive restructuring of the reference counting of
zones and related objects.
Zones now attach to their views. To avoid a circular dependency that
would keep views from ever shutting down, this is done using the new
functions dns_view_weakattach() / dns_view_weakdetach() which
guarantee that the view will not be freed but still allow it
to be shut down.
The zones themselves now only have a single reference count, with
similar "weak" semantics. Managed zones must now be shut down
explicitly by calling dns_zone_shutdown(). To shut down all
zones in a zone table, call dns_zt_shutdown().
The zone manager is now reference counted, weakly. To shut down the
zone manager, you must explicitly call dns_zonemgr_shutdown().