Processes

When running a computer program, every operating system uses the concept of a process. An operating system process encapsulates the program code and any state that it must maintain to successfully perform the task the programmer wishes to accomplish. Each process runs one or more threads of execution, which is a sequence of instructions run by a process and managed by a scheduler, which can either run multiple threads simultaneously, or switch between them as needed.

Whilst all operating systems supported by LibreOffice use processes to execute code, there are really only two process models that LibreOffice implements in the OSL - the Windows process model, and the Unix POSIX process model. Whilst many process concepts are the same between these two operating systems, there are some significant differences that the OSL attempts to unify in a common abstraction.

Common functionality

Both Unix and Windows load a process from a stored image, which contains the program instructions associated with the program. Common features of processes under both operating systems are:

• Memory handling via:

  • virtual memory which maps virtual addresses in a process to physical addresses in memory.

  • mechanisms for gathering input and output between processes

  • a set of stack frames in the call stack. Each stack frame holds:

    • a function's local variables

    • a frame pointer that holds the address of the calling function so that it can be returned control once the current function returns

    • the parameters passed to the function by the calling function

  • the heap (or sometimes called the free store) which allows programs to dynamically allocate blocks of unused memory from a large pool of memory

• Resource descriptors are allocated to a process by the operating system, for such things as files, synchronization primitives and shared memory - in Windows these are called handles and in Unix they are file descriptors

• The ability to hold processor state, also known as the process' context

• Security attributes, such as the process owner and the set of allowable operations the process has permission to run

• The ability to spawn child processes

• The ability to manage one or more threads within the process

• Environment variables, which are inherited from their parent processes

A process can exist within a number of states, the main ones being:

• The initial state is the created state, where the program has been loaded into main memory but has not been processed by the operating system scheduler

• From the created state, the process switches to a waiting state, where it waits for the scheduler to make it run via what is known as a context switch

• Once the scheduler does a context switch successfully the process will start executing, which is a state more commonly known as running

• Another valid state is for a process to be waiting on another process to finish it's exclusive control of a resource that it needs. When a process is prevent from running in these circumstances, the processes is said to be blocked.

  Figure: Process States Source: Wikipedia, License: Public Domain</span>

Differences between Win32 and POSIX process models

The areas of difference between Win32 and POSIX process models are process pipes, interprocess communication, process termination and a different security model.

Process pipes

On POSIX systems, a file descriptor is used to communicate between processes. Instead of just anonymous pipes for standard input, standard output and standard error, POSIX based systems allow two processes to connect through any file descriptor.

On Win32 systems, the STARTUPINFO structure references Windows handles (the HANDLE macro) that point to standard input, standard output and standard error pipes. When a new process is created via the CreateProcess() function, you setup the STARTUPINFO structure and use the CreatePipe() function to create an anonymous pipe to connect the read end of the pipe to the write end of the pipe - each process then associates the pipe to the hStdInput, hStdOutput and/or hStdError structure fields. However, as the STARTUPINFO structure on allows standard input, standard output and standard error you cannot setup extra channels like you can in POSIX. It is also important to note that named pipes must be used for asynchronous IO as anonymous pipes are unable to use asynchronous IO.

Interprocess communication

Aside from anonymous pipes, processes on most operating systems can communicate with each other via shared memory and named pipes. However, POSIX compliant systems can also use Unix domain sockets.

IPC Method	Description
Shared Memory	Shared memory works by allowing multiple processes access to a block of memory which is accessed by all the processes. A process will read and write to this memory to communicate between each process. As more than one process is accessing the same block of memory, synchonization primitives are necessary to mitigate race conditions and things such as dirty reads.
Named Pipes	Allows processes to communicate via the filesystem, through a file that becomes a unidirectional data channel. On a Unix system, the named pipe remains till it is specifically removed, whereas on Windows systems when the last reference to the named pipe is closed the pipe is removed.
Unix Domain Sockets	Applies to POSIX compliant operating systems. Creates a socket, but uses filesystem inodes for addressing. These allow for bidirectional communication between more than two processes and supports passing file descriptors between processes.

The OSL implements bi-directional "pipes", which are indeed named pipes on Windows; on Unix, however, it is implemented as a Unix Domain Socket so is not really a true Unix named pipe.

Process termination

Windows process termination

Windows has a very different way of terminating processes to Unix systems. In Windows, a process terminates but does not terminate any child processes it has created. It deferences any kernel objects that it holds, but until all references are removed by all processes then that kernel object will not be destroyed. According to Microsoft, the following occurs:

Terminating a process has the following results:

• Any remaining threads in the process are marked for termination.

• Any resources allocated by the process are freed.

• All kernel objects are closed.

• The process code is removed from memory.

• The process exit code is set.

• The process object is signaled.

While open handles to kernel objects are closed automatically when a process terminates, the objects themselves exist until all open handles to them are closed. Therefore, an object will remain valid after a process that is using it terminates if another process has an open handle to it.

(Terminating a Process, MSDN article)

Processes on Windows will terminate under the following circumstances:

• When a thread calls on ExitProcess()

• When the last thread terminates

• Any thread calls on the TerminateProcess() with the handle of the process object

• When an end user logs of the system

• When the system is shutdown

• If a console process receives a CTRL+C or a CTRL+BREAK signal, the process calls on ExitProcess()

Unix/POSIX process termination

On Unix systems, however, it is a bit more complicated. A process will terminate when the process calls on exit(n) (or return n in C and C++), and the value n is passed to the process' parent process. However, the process will not fully remove the process from the kernel's process tables until the parent process collects the exit status, which is the exit code and the termination reason, which is normally contained in a 16-bit integer (the first byte containing the exit code, and the second byte containing a bit field with the termination reason).

To collect a termination status from a child process, the parent process must call the wait() or waitpid() system call. The wait() system call waits for the child process to finish, and blocks the parent process until it gets the exit status of at least one of the child processes (which one exits first, it does not matter). The waitpid() system call, on the other hand, allows the PID to be specified as the first parameter, also also allows the system call to be non-blocking if the third option's parameter is set.

Note: if -1 is specified as the PID argument of waitpid(), then the process waits for the first child process to terminate, whilst a PID of 0 makes it wait for the first process in the process group to terminate. If the PID value is less than 0 then this indicates the process must wait for the first process of any whose process group ID is equal to the absolute value of PID)

If a process' parent dies and the children processes remain alive, these processes will be reparented to process 1 (the init process). Any children processes that have terminated but have not yet been waited on are called zombie processes until the parent process waits on it. If a zombie process' parent is killed, then it is reparented to PID 1 and this process periodically kills these processes via a mechanism called process reaping.

OSL C API

The OSL uses a C API for managing the process lifecycle. The API covers:

• Process creation and execution

• Process termination

• Inter-Process Communication

• Threading