Device Filesystem in pOS
------------------------
05/08/05

The current design of the virtual filesystem in
pOS represents devices as real files on the
root-filesystem. On each filesystem access
(e.g open), the VFS-layer first reads the files
into buffer and tries to gather information on devices
represented by this file. It does this in any case even
in case it's opening a normal file only. This technique
has two major disadvantges:

- the device-file recognition is very unsure and unsafe,
  e.g anyone could write a file containing any devices
  numbers and thus create device nodes; and sometimes
  a file could, even though this is very unlikely, contain
  random data that could be misinterpreted as device
  reference-numbers and therefore write data accidentally
  to any device

- as the whole process takes place on any filesystem access
  it produces some unwanted overhead

To solve these issues, devfs is introduced as a memory-resident
virtual filesystem which creates virtual device nodes.
This idea isn't actually new, as it was already aimed in
previous concepts for VFS but there wasn't a satisfactory
comprehensive design concept for VFS and during development
of the VFS design, devfs was temporarily dropped. The
major requirement for devfs, the directory cache, dcache
is now fully featured in VFS. Thus devfs can completely
reside in memory.

Due to the mature design of the current VFS-concept the
devfs-implementation is very easy. It has the following
main features:

- devfs is a convential filesystem driver which exports
  all typical functions provided by any other filesystem
  driver and thus cannot be recognized as a special
  filesystem except that it has no mount source
- all file-type recognition in the mount/open procedures
  will be dropped as vfs now only needs to determine
  the fs-driver and pass the operation requested by
  the user-space program to it
- devfs is mounted during k_mount_init () and cannot
  be unmounted; in order for its directory entries
  to have proper data (especially the parent-node
  of every dentry "/" needs to be known for proper
  dcache-operation) the root-filesystem still
  needs to be mounted prior to any other filesystem
- all devices to be used are defined in hash-table
  in source code that could like this

  #define fd0	{{1,0},"fd0"}
  #define fd1	{{1,1},"fd1"}
  #define ser0	{{2,0},"ser0"}

  on devfs initialization the virtual
  devices are created in dcache based
  on this information; for this reason
  devfs has its own struct in the union
  section of s_dentry called s_devfs_dentry