A reference counted pooled implementation, that basically consists of a main buffer, that can be sliced off into smaller buffers, and the underlying buffer will not be freed until all the slices and the main buffer itself have also been freed. This also supports the notion of un-freeing the main buffer. Basically this allows the buffer be re-used, so if only a small slice of the buffer was used for read operations the main buffer can potentially be re-used. This prevents buffer exhaustion attacks where content is sent in many small packets, and you end up allocating a large number of buffers to hold a small amount of data.