User level view of os

System Software for Program Execution:
User-level View of Operating Systems

COMP 229 (Section PP) Week 6

Prof. Richard Zanibbi
Concordia University
February 13, 2006

Last Week...
Supports Programming,
Efficient System Use
System Software for Program Execution (pp. 1-18)
Systems and application software (again)
Resource Abstraction Applies System Software
Resource Sharing to a Problem

Operating System Strategies (pp. 18-39) including
A brief history of computing and operating systems BIOS
(batch, timesharing, personal, SCC’s, embedded, networks)
The modern OS as integration/evolution of previous systems

Goals:
1. Learn the function and operation of operating systems
2. Learn how to exploit OS design during programming (for
program execution) “Hacker notes” 2

Resource Abstractions
Files
Represent data on storage devices (e.g. disk)
Represent data structured as a list

Abstract Machines
Represent resources for executing programs
Program running on abstract machine: a “process”

Good Abstractions for Programmers:
Easy to understand
Provide necessary control (not “too abstract”)
COMP 229-PP (Week 6, Feb. 8) 3

Resource Sharing
Resource Sharing
Time-multiplexed: share time on resource (“take turns”, e.g. processor)
Space-multiplexed: share space on resource (e.g. memory)

Transparent Resource Sharing
Sharing “invisible” to programmer (done by OS)
e.g. simulating multiple abstract machines (multiprogramming)

Explicit Resource Sharing
Programmer requests resource sharing
e.g. sharing memory between programs

Resource Isolation
OS permits one process (abstract machine) access to a resource at a time

Security and Consistency Issues
Prevent problematic or malicious access to resources
4

Example

(Multiprogramming “best case”
performance, Fig. 1.8)


Operating System Strategies
OS Strategy
General characteristics of the programmer’s abstract machine

Early Days (approx. pre-1960)
One program at a time, basically no OS

Strategies
1. Batch systems: executed non-interactively, multiprogramming
2. Timesharing systems: multiple users interacting with the system;
extended multiprogramming, security
3. Personal Computers, and Workstations: single-user machines;
multiprogrammed, minimize wait time rather than max. hardware
use (many had no file system (!); provided by CP/M, MS-DOS)
4. Embedded systems: guaranteed response times (“real time”)
5. Small, Communicating Computers (SCC’s): new resource
management, power management, device management
6. Network Technology: handle resources, information on networks


This week...
User-Level View of Operating Systems
The Programmer’s Abstract Machine (pp. 42-47)
Sequential Computation (“single thread”)
Multithreaded Computation

Process Resources (pp. 47-52)
esp. files (under UNIX, Windows)

More on Processes and Threads (pp. 52-58)

Writing Concurrent Programs (pp. 58-72)
Executing Computations
Executing Commands in UNIX and Windows

Objects (pp. 72-74)
7

The Programmer’s
Abstract Machine
(pp. 42-52)

Sequential Computation
Algorithm
A process or set of rules for calculation or problem-solving, esp. with a
computer (CA Oxford Dictionary)

Sequential algorithms
Describe a sequence of operations
Fundamental abstraction for computation

Examples
Sorting and Search Algorithms
(e.g. Vol.3,“The Art of Computer Programming” (Donald Knuth))

Algorithms May be Defined Using
– Natural Language (e.g. English or French)
– Pseudocode (“informal” operation sequence)
– Mathematics
• may not provide an explicit operation sequence
– As a Source Program in a Programming Language (e.g. C/C++, Java)
• explicit sequence of operations (unambiguous) 9

System Calls
Binary Program
– Machine-readable translation of source program (e.g.
“load module”, or executable program (.exe) )
– Represents explicit machine instructions used to
implement an algorithm (e.g. op. sequence)

System Calls in Binary Programs
– Procedure calls to the system call interface invoke OS
features (e.g. stop a program, read data from a device)
– System calls allow the programmer to take advantage
of resource abstractions and sharing built into the OS


What is POSIX?

Portable Operating System Interface
Developed to standardize UNIX system call
interfaces
POSIX.1: published 1988
Now used by a large number of UNIXes,
including Linux
• Linus Torvalds, Univ. Helsinki, Finland, first
announced Linux on the Minix newsgroup in 1991


Sequential Computation and
Multiprogramming
Running a Program
1. Translate source to binary program (and possibly link)
2. Provide program input data, parameters
3. Direct OS to start executing instructions from the transfer address
(“main entry point”) (includes OS calling program loader)
4. Program halts when last statement is executed or system call for halt
(exit()) is made

Execution Engine (part of a process)
Used by OS to implement multiprogramming
Status of abstract machine running the process
Copy of the runtime stack for the process

Runtime Stack
Used to implement “scopes” in languages (among other things)
Contains local variables, return addresses, etc.

Fig. 1.2: “Classic” processes (e.g. UNIX) 12

Multiprocessing vs. Multithreading
“Classic” Process
Sequential program “run” alone on its own abstract machine

Multiprocessing
Concurrent execution of processes
Processes contain separate resource abstractions, binary program copies

Multithreading
Thread: another name for “execution engine”
• “Classic” processes have a single thread
Multithreading: multiple threads within a single process
Threads within a process share resources, including binary program

Example: Fig. 2.6 (Multithreaded process, as found in Windows
NT/XP, Linux with POSIX threads extension (2.2+))
Example: Fig 2.3 (“accountant example”)

Implementing Threads
Example: Java
Java Virtual Machine (JVM) supports threads
Thread base class supports threads
Create subclass of Thread (e.g. MyThread): instances
define a new thread

Executing Threads
– If multiple processors are available, threads may be
executed in parallel
– Otherwise (usually) threads share the processor
(through time-multiplexing)


Process Resources

(pp. 47-52)

OS Management of
Resources Used by Processes
Resource Allocation
Threads access resources through system calls
(insures resource isolation)
1. Threads must request a resource before use
2. After request, thread is suspended until resource is allocated to it
• while suspended, other threads/processes allocated processor time
3. After a resource is allocated to a thread, all threads in its
associated process may share access to the resource (the
process “owns” the resource)

Memory and Processors
Traditionally distinguished from other resources
– Processes/threads (implicitly) request processor when ready
– When user process is started, loader requests memory from the
OS to run the program (e.g. user login -> shell -> interaction)
16

Resource Descriptors
– Data structures used to represent physical and
abstract resources managed by the OS
– Indicates if resource is available
– Includes record of which processes are
currently blocked waiting for the resource

e.g. file descriptors, process descriptors


Files and File Descriptors
Files
– Named, linear stream of bytes stored on a device
– Common (default) method for representing data storage
– Forms the basis of many other device abstractions
– OS maintains record of which files are currently open

File Lock
OS preventing processes other than currently allocated from accessing a
file (e.g. for writing)

File Descriptor
OS’s internal representation of file attributes (for file resources)
• e.g. Indicates if file is locked, or read-only
UNIX: a non-negative integer (file descriptor (inode) number)
Windows: “HANDLE”: reference to internal data structure


(Internal) OS Representation of
Byte Stream Files
Name: Test (ASCII)
Byte Value
File Pointer
On file open: 0 0100 0001 A
(default) pointer set to 0 1 0100 0010 B
Read/Write K bytes: 2 0100 0011 C
Advance pointer by K
3 0100 0100 D
Setting File Pointer: 4 0100 0101 E
lseek() (POSIX)
SetFilePointer() (Win) 5 0100 0110 F
No “Type” for Bytes: 6 0100 0111 G
19
Treated as “raw” bytes

Example: POSIX File System Calls
System Call Effect
open() Open file for read or write. OS creates internal
representation, optionally locks the file. Returns
a file descriptor (integer), or -1 (error)
close() Close file, releasing associated locks and
resources (e.g. internal representation)
read() Read bytes into a buffer. Normally blocks
(suspends) a process until completion. Returns #
bytes read, or -1 (error)
write() Write bytes from a buffer. Normally blocks
(suspends) a process until completion.
Returns # bytes written, or -1 (error)
lseek() Set the file pointer location

fcntl() (“File Control”): various options to set file
20
attributes (locks, thread blocking, ....)

Rough Equivalences
for POSIX and Win32 API
POSIX Win32
open() CreateFile()/ OpenFile()
close() CloseHandle()
read() ReadFile()
write() WriteFile()
lseek() SetFilePointer()

21

Examples

UNIX file system calls, Fig. 2.4
Windows file system calls, Fig. 2.5


Using File to Represent Other Resources:
Devices and Pipes in UNIX
Other Resource Abstractions
e.g. keyboard, display, memory, processors
– The more similar these abstractions are, the easier they are for
programmers to use

UNIX Devices
Devices are represented as files in the directory tree
– Devices: located in the /dev directory (e.g. /dev/tty)
– Devices have open(), close(), read(), write(),
seek(), fcntl() commands implemented in their drivers
(similar to file interface)

UNIX Pipes
Abstract resource allowing processes to communicate by chaining
outputs to inputs
e.g. > cat schedule.txt | grep deadline
– Inputs/outputs are represented using standard files for each
process (standard input and standard output) 23

More on Processes and Threads
(pp. 52-74)

Heavyweight vs. Lightweight Threads

Heavyweight (“Classic”) Process
Processes which may contain exactly one thread
(execution engine)

Lightweight Process (= “Thread”)
– A thread (execution engine) within a multithreaded
process
– “Light” because each thread does not require the
overhead of a separate process
Illustration/Example: Fig. 2.6

Thread Representation (Data)
in Detail
Runtime Stack
Data private to the thread (e.g. local variables)

Status
OS data structure: all properties unique to the thread
– program counter value
– whether thread is blocked waiting for a resource
– which resource is being waited for
– etc.


The Move Towards Threads
Lightweight Processes (Threads)
Became popular in late 1990’s (esp. w. Java)

Motivation: Sharing Resources
– Server managing shared file systems
– Windowing systems using threads to represent
individual windows within the physical display

Example: Fig. 2.7 (window threads)

Thread Libraries vs. True Threading
Thread Libraries (Early 1990’s)
Used to run threads within heavyweight (“classic”)
processes (e.g. Mach C thread library)
OS still implemented classic processes only
Problem: if one “thread” blocks, then all threads block

“True” Threading
OS implements and manages threads independently of
one another
If a thread in a multithreaded process blocks, the other
threads can still execute

Creating Processes and Threads
Initial Process
– The first process started when machine is turned on
– Usually the bootstrap loader loading the OS into
memory

Creating New Processes and/or Threads
Done using system calls (i.e. another resource request)
– If threads are available, separate system calls are used
for creating threads and creating processes


(Initial Process, after turning computer on)

Bootstrap Loader
Child Process of (Loads OS) Parent Process of

OS (UNIX)
Sibling Processes

Login Shell 1 Login Shell 2 Login Shell 3

Bash Shell

Email Program Web Browser

FORK(), JOIN(), and QUIT()
System Calls
(Conway, 1963; Dennis, Van Horne 1966)
Purpose
Define cooperating sequential processes executing on shared data
Like threads, processes can share resources (e.g. binary program copy)

FORK(label)
– Creates a child process that begins execution at “label”
– Parent (calling) process continues execution after child is created
– Parent, child are executed concurrently

QUIT()
Terminates and deallocates (removes) the process issuing the call

JOIN(count)
– Used to merge processes
– count is an integer variable shared by a parent and its child processes
– Locks processor until system call is finished (prevents inconsistency)

31

Effect of JOIN(count)

/* Decrement shared variable */
count = count – 1;
/* QUIT unless this is the last process */
if (count != 0) QUIT();

(example provided pp. 56-57)


Creating “Classic” (Heavyweight)
Processes (~1970)
Classic Process
Single-threaded
Separate resource representation per process
Each has a separate, private address space

Address Space
– Physical resources (mainly memory) that can be
referenced by the execution engine (single thread)
– Contains stack, status, data, and program (“text”)
– A memory protection mechanism: prevents writing
over memory allocated to other processes
33

Creating “Classic” (Heavyweight) Processes
Using fork() in POSIX/UNIX
fork()
Creates new process
Child is given copy of current address space
Parent and child address spaces are separate
• e.g. changing a variable in the child process does
not change the variable’s value for the parent

Process Communication In UNIX
Only permitted through open files

Example: Creating a new process using
fork()
1. fork() instruction is executed, causing
a system call
2. OS copies data (including address
space) from the original process to
the child process
3. The child process starts execution at
the instruction following the fork()
instruction


Creating Modern and (Lightweight)
Processes
Creating Modern Process
– Created by system call from a thread
– Child created with separate address space, resources
– A base thread must be created to execute the process
e.g. CreateProcess() in Windows API

Threads Creating Child Threads within a Process
– Again, using a system call from a thread
– Similar to Conway FORK(): child thread runs within the
process containing the parent thread
– Child thread has stack and status separate from parent
e.g. CreateThread() in Windows API
36

Example: Creating a New Thread
in a Modern Process
1. CreateThread(...) is executed
• NOTE: unlike fork(), CreateThread() has many
parameters
2. OS creates thread data within the
process address space (i.e. within the
same process)
3. OS starts execution of child thread at
the indicated starting point


Writing Concurrent Programs

(pp. 58-72)

Multiple Single-Threaded Processes:
the UNIX Model
Unix Process Behaviour Defined By
Text segment (program instructions)
Data segment (static program data)
Stack segment (run-time stack: local variables)
*located in the address space

UNIX Executable (Binary) Programs
Compiled/Linked programs define the text, data, and stack
segments (“a.out”)

Example: Class UNIX processes (Fig. 2.10)
Note that the address space contains the Status, Stack Segment,
Text Segment, and Data Segment

Process Descriptors and
Context Switching
Process Descriptor
OS Data Structure representing process attributes, incl.:
• PID: Process Identifier (a reference to the process descriptor)
– UNIX: PID is integer index for OS-internal table of process descr’s
• User that created the process
• Parent, Sibling, and Child processes
• Resources held by process (e.g. file descriptor references)
• Address space for the process
• Threads within process
• State and Stack Location (in memory) (for classic process)

Context Switching
Refers to switching the active process and/or thread
May involve swapping a process/thread out of memory
and storing it on a device (e.g. disk), among other ops. 40

UNIX Process Commands
“ps” command
– Lists process identifiers (PID) for the user
– “ps –aux”: lists all processes in the system

int fork()
– Returns PID of child process to the parent
– Returns 0 to child process
– Child given copy of text, data, stack segments, access
to all open file resources
– Parent process descriptor also copied
– The next process run may be the child, parent, or
some other process

fork() Examples
Example 1: What do the parent and child output?
theChild = fork()
printf(“My PID is %dn”,theChild)

Example 2: Directing child process execution
childPID = fork()
if (theChild == 0) {
/* child executes here */
codeForTheChild();
exit(0); // Important!
}
/* Parent executes here */
...


Changing Program Data Dynamically:
execve()
int execve(char *path,char *argv[],char *envp[]);
program arguments environment
variables
Effect of execve()
– Replaces text, data, and stack areas using the program
loader
– After loading, stack is cleared, variables initialized
– Program is started (call does not return)

System Calls to Wait for Child Process to Stop:
wait(): block until any child process terminates
waitpid(P): block until child process with PID P terminates
43

Multiple Modern Processes
(e.g. Windows)
Process Creation (CreateProcess())
Flexible: many more parameters than in fork() (10!)
Base process can start anywhere in program code
Returns two resource descriptor references:
• one for the new process
• one for the new base thread

Thread Creation (CreateThread())
Again, flexible, this time with 6 parameters
A thread may create a sibling thread

(see text pp. 66-67 for function signatures, discussion)
44

Executing Commands in
UNIX and Windows Shells
Fatal Errors
Unrecoverable process errors
• e.g. trying to write outside a processes’ address space

System Shells
Users needs to be able to execute any program,
however buggy, but without crashing the shell

Solution:
Create a child process to execute each command
• e.g. using fork() or CreateProcess()

Objects: Another Abstraction
Objects
– Originally used to represent independent “computation units” in
simulations; similar to a small process (e.g. workflow simulation)
– Have private data and operations, along with message passing
operations (e.g. get/set operations)
– Can be used to represent a coordinated execution of distributed
computational units (as was common in simulations)

Classes
– Simula 67: introduces notion of classes to define object
behaviour (similar to programs for threads/processes)
– First widespread application: interfaces (e.g. InterViews, 1989)
– Java: OS calls into the Java Virtual Machine (JVM), which runs
on top of the native OS (a simulation in an abstract machine!)
– OS may be designed to support efficient object implementations
– Early attempts at “pure” OO operating systems: efficiency issues
47

Windows NT/2000/XP
Kernel and Objects
Primitive OS Components
Defined as objects
Implemented as C functions: no language inheritance

Threads
– Creating a thread generates a kernel Dispatcher
Object, which then has additional status fields specific
to threads added (“pseudo inheritence”)
– This approach allows Kernel operations to be applied
uniformly to a broad class of abstract data types


Summary
• Classic Processes (e.g. in “Old” UNIX)
• Multithreaded Processes (e.g. in Linux,
Windows)
• Creating Processes and Threads
• Resource Descriptors (file and process)
• Resource Management
• File (e.g. open()) and Process (e.g. fork(),
CreateThread()) System Calls
• Overview of Writing Concurrent Programs
• Objects and Operating Systems 49

Next Week...

Operating Systems Organization
pages 89-110

Process Management
pages 197-206


User level view of os

More Related Content

What's hot (19)

Similar to User level view of os (20)

More from Mohd Arif (20)

Recently uploaded (20)

User level view of os