* Building on old distributions.
  Trinity is developed using the latest glibc/kernel, which means from time to time
  changes are introduced which may make it fail to compile on older distributions
  (especially enterprise ones).  The preferred way to fix this is to add the missing
  declarations to compat.h
  If for some reason you can't detect presence of a definition in a header file
  (or need to test prescence of a header file), add it to configure.sh. This should
  be a last resort, in most cases an #ifndef addition to compat.h will be enough.

  This means of course that trinity will be trying to run new syscalls on older kernels
  that don't implement them, but it will only call them once. After the first -ENOSYS,
  it gets marked as "Don't run this again".  It also allows us to run and test newer
  kernels on distributions with older userspace whose header files may not define those features.

* Porting to new architectures
  - Add a syscalls-{archname}.h defining the syscall table in the same order as your arch
  - Add a arch-{archname}.h to define useful things like PAGE_OFFSET
  - If your arch starts its Linux syscalls at non-zero (like MIPS/IA64 do), define SYSCALL_OFFSET
    in arch-{archname}.h
  - Add syscalls/{syscallname}.c for your arch specific syscalls.
  - Change setup_syscall_tables in tables.c to pick the correct syscall table for your arch.
  That should be it.

* Adding new sanitise routines
  syscalls:
   - First, decide if it's worth adding a sanitise routine at all, or if the generic
     sanitise routines are good enough, by setting each arguments .arg*type member
     to an appropriate ARG_*
   - In cases where the generic routines are either insufficient, or need to be overridden,
     the .sanitise member of the syscall struct can be pointed to a routine that will
     run after the generic sanitise routines.
   - The 'child' parameter of the sanitise function is to be used to index the various
     per-child arrays in the shm.
   - Likewise, after the syscall is run, you may need to free up resources allocated
     in the .sanitise.  You can do this in the .post function.
     There's an unsigned long in the shm named 'scratch' that may be used to store
     ptrs to resources etc for retrieval later in .post
   - Beware of 'over sanitising' and limiting the possibility for boundary conditions
     to be violated. A syscall that always succeeds is not the goal, you want to be
     exercising as many error paths as possible.

  ioctls:
   TODO

* Process model.
  On startup, the first process sets up the shm etc, and then forks two processes.

  1. The 'main' process (see main.c), which is responsible for..
     - initial opening of sockets/fd's.  See sockets.c/fds.c/files.c
     - forking new children See fork_children()
     - handling those children.  See handle_child()

  2. A 'watchdog' process that makes sure the child processes are making progress,
     and assorted other 'periodic' tasks.  All this lives in watchdog.c