[Inspired by comments from https://crrev.com/c/804681 review]

1) Cryptohomed has multiple D-Bus entry points that allow mounting and unmounting. It is not clear which of them are currently in use and for what use cases. At least MountEx and AsyncMountGuest are in uses (kiosks may use more).

It makes sense to reduce all mounts to a single set of mount/unmount (or rather sign-in/sign-out - see #3 below) calls, probably based on MountEx for mounting. That will eliminate a lot of dead code, greatly simplify understanding cryptohome workflows, and will allow further improvements (e.g. the one in #2 below).

2) In its current state, cryptohome doesn't have proper separation of concerns between different layers involved in mounting user homedirs.

Specifically, Mount class, besides actually mounting the directory, also  finalizes boot lockbox, deals with chaps keys, for example. soon it will also be dealing with telling TPMs what type of user is currently signed in. 
The mount destinations for various migration cases may also be better set by an upper layer. Plus, it makes sense to split dircrypto and ecryptfs versions, and call one or both of them depending on flags from the upper layer.

3) It is not clear if mounting is always synonymous to user sign-in. Some actions, like adding PKCS#11 token are now tied to mounting, but conceptually are a part of the user sign-in. Mounting is also used as the only means to verify user credentials locally. So, they are tightly coupled.

Unless some decoupling makes sense and is planned, we may want to better express the intent in the cryptohomed interface and call these actions sign-ins/sign-outs rather than mounting/unmounting (if not in the D-Bus method names then at least in comments and other documentation).