s4: server: Fix crash in NTVFS server caused by ordering of destructor calls.
In the NTVFS server we have the following talloc heirarchy:
event_ctx
|
---------------------------------------------------- .. other children
| | |
msg_dgm_ref srv_conn msg_dgm_ref
^ |
| NTVFS structures
| |
| XXXXXX
| |
| |
--------------------- pointer to msg_dgm_ref
Some of the structures under NTVFS (marked XXXXX) can have
pointers to imessaging contexts which internally have pointers
to msg_dgm_ref structurs allocated off event_ctx.
The original code calls:
model_ops->terminate(event_ctx, srv_conn->lp_ctx, reason);
talloc_free(srv_conn);
But model_ops->terminate() calls talloc_free(event_ctx) and
then calls exit(). In this case srv_conn is never explicitly
freed, but only freed as a talloc child of the event_ctx.
Depending on the ordering of the linked list of talloc children
under event_ctx(which can be reordered via talloc_free/reinit
of msg_dgm_ref) a pointer to msg_dgm_ref under srv_conn can
be left pointing to memory that was already freed. This pointer
is then used in the destructor for a file object called when
srv_conn is freed.
Re-ordering this to explicitly call TALLOC_FREE(srv_conn) first
and then model_ops->terminate() fixes this problem.
Signed-off-by: Jeremy Allison <jra@samba.org>
Reviewed-by: Ralph Böhme <slow@samba.org>
Autobuild-User(master): Jeremy Allison <jra@samba.org>
Autobuild-Date(master): Sun Apr 2 05:18:39 CEST 2017 on sn-devel-144