poll_funcs_tevent: Fix a valgrind error
The valgrind error happened in poll_funcs_tevent_handle_destructor in
if (handle->ctx->refcount == 0)
handle->ctx was already gone at the time this destructor
was called. It happened because during messaging_init the
messaging_dgm subsystem was free'ed. The unix_msg context and the
poll_funcs_tevent_context are children of messaging_dgm_context. How
was poll_funcs_tevent_handle_destructor still called? While working
on the new notify subsystem I've added some messaging_read_send
tevent_reqs, which register themselves with the dgm_context via
messaging_dgm_register_tevent_context. They were not gone yet. When
later these were also run down due to another talloc_free somewhere else,
this destructor referenced dead memory.
This code now protects the poll_funcs_tevent_handle against the
poll_funcs_tevent_context going away first with the loop
for (h = ctx->handles; h != NULL; h = h->next) {
h->ctx = NULL;
}
in poll_funcs_tevent_context_destructor together with
if (handle->ctx == NULL) {
return 0;
}
in poll_funcs_tevent_handle_destructor.
A side-effect of this code is that messaging_read_send request won't be
satisfied anymore after a reinit_after_fork kicked in. But I think this is the
right thing anyway: Every process should register its own message handlers
explicitly.
Signed-off-by: Volker Lendecke <vl@samba.org>
Reviewed-by: Jeremy Allison <jra@samba.org>