* VFS cont, super_operations

a lot of refcount ops in linux start with get/put:
- dget/dput (dentry)
- iget/iput (inode)

* pathname lookup (namei), revisited

unlink("/a/b/c/d/foo.txt"):

Any new pathname that comes through a syscall, gets put into a struct called
a "nameidata" (ND) structure.

1. a series of ->lookup and ->permission
- if no permission: return error EPERM/EACCES
- if intermediate component (non-leaf) is NOT a directory: return ENOTDIR
- if any component doesn't exist: ENOENT

2. If any component is a symlink
- do a ->readlink and "replace" component w/ symlink's content/target
- suspend processing of orig pathname and perform a "subroutine" to
  interpret new symlink
- any path of a symlink could itself be a symlink...
- when creating a ND struct, count how many times we crossed a symlink
- if counter is greater than some threshold (often 20), then abort lookup
  and return error ENAMETOOLONG

3. If a directory is a mountpoint, need to cross that mount point

# mount -t ext4 /dev/sda1 /usr
$ ls /usr
homes/
$ ls -R /usr
(only homes)
# mount -t nfs server:/path /usr/homes
$ ls /usr
homes/
$ ls -R /usr
(lots of files, all of /usr/homes/*/*...)
$ ls /usr/homes/jdoe/foo.txt

lookup "usr" in "/", find that it belongs to ext4
- check each dentry you got back if dentry->d_flags & DCACHE_MOUNTED
lookup "homes" inside "usr", initially get the dentry+inode for ext4's "homes"
- now finding that ext4's "homes" dentry is DCACHE_MOUNTED
- if found, consult a global hash table of mount points, lookup the dentry
  that's mounted "here".  What's returned is the root dentry of the f/s
  mounted here (nfs in above example).
next lookup goes into nfs f/s, looking for "jdoe" inside the newly found
root dentry of that mounted nfs f/s.

#define DCACHE_MOUNTED			0x00010000 /* is a mountpoint */

- ->unlink at the very end

* struct file_system_type

name of f/s
flags
and ctor and dtor for struct super_block (ctor has to create struct
sb, populate it, and fill in sb->s_root -- root dentry).