Re: Capability object revocation

[Neal H. Walfield]

There are two scenarios here: 1) implementing revoke which means
revoking the capability for all capability entries but the one used by
the caller; and 2) destroying a capability.

In the former case, you inhibit all RPCs on the capability object
(hurd_cap_obj_inhibit), mark all capability entries as dead *except*
the one on which this RPC is made and then resume RPCs on the
capability object (hurd_cap_obj_resume).  Note that marking a
capability entry as dead means not only setting entry->dead to true
but 1) removing an internal reference and if the reference count is
zero, dereferencing the underlying cap object (this should be its own
function); and 2) removing it from the CLIENT->CAPS_REVERSE list as
per client-release.c.

To destroy a capability, you'll want to do the same as before except
revoke all capability entries and instead of calling
hurd_cap_obj_resume, call hurd_cap_obj_end.  As you'll notice
hurd_cap_obj_end is not yet implemented.  It should set OBJ->STATE to
_HURD_CAP_STATE_BLACK a la _hurd_cap_client_end and
_hurd_cap_bucket_end.  This is necessary because in,
e.g. manager_demuxer, after looking up the capability entry and
verifying that it is not dead, we wait until OBJ->STATE returns to
_GREEN but don't again check that the entry is not dead.  This change
also requires updating the comment for the hurd_cap_obj structure in
cap-server.h and changing bucket-manage-mt.c:manage_demuxer and
ctx-cap-use.c to bail when OBJ->STATE is _BLACK.

Thanks,
Neal