- Cache Manager

This document describes the current implementation of the cache manager. 

- Description

In its current state the CM (cache manager) only includes primitives to
cache disk blocks, this is useful for filesystem metadata but it requires an
additional copy operation when reading files from cache. This will be fixed.

Each disk drive with data in the cache has an associated DCCB (Device Cache
Control Block) which details all the blocks from the device in memory. If a
filesystem requires cache services for a device it must call CbInitDccb to
initialize this structure.

Each block with data from a device has an associated CCB (Cache Control
Block) with a pointer to the physical memory used to hold the block, and
various state and locking information. When a filesystem requires a block
from the device it calls CbAcquireForRead or CbAcquireForWrite which ensure
the data for the block is uptodate (loading it from disk if necessary) and
return a pointer to the associated CCB. When a filesystem has finished with
a block it calls CbReleaseFromRead or CbReleaseFromWrite, it is important to
call these functions because the CM can only release blocks from the cache
if they have no active readers or writers. The CM also enforces cache
consistency by ensuring that while multiple threads can be reading a block
simultaneously, there is only ever one active writers. 

The CM provides no support for deadlock prevention/detection as it has no
knowledge of the layout of a a filesystem (nor should it have).

- TODO