Administrivia

Class web page: http://cs140.scs.stanford.edu/

CS140 Operating Systems

Instructors: David Mazi` res, Andrea Bittau (alt) e CAs: Juan Batiz-Benet, Matt Sparks, Kiyoshi Shikuma, Ali Yahya Stanford University

- All assignments, handouts, lecture notes on-line

Textbook: Operating System Concepts, 8th Edition, by Silberschatz, Galvin, and Gagne
- This is the ofcial textbook, but mostly for background - Class will not rely heavily on textbook - Old versions okay, or might get away without textbook

Goal is to make lecture slides the primary reference

- Almost everything I talk about will be on slides - PDF slides contain links to further reading about topics - Please download slides from class web page

1/33

2/33

Administrivia 2
Staff mailing list: [email protected]
- Please mail staff list rather than individuals for help

Lecture videos
Lectures will be televised for SCPD students
- Can also watch if you miss a lecture, or to review - But resist temptation to miss a bunch of lectures and watch them all at once

Newsgroup: su.class.cs140 main discussion forum Key dates:

- Lectures: TTh 4:15-5:30, Gates B03 - Section: Some Fridays, time/location TBD - Midterm: Tuesday, Feb 8, 4:155:30pm - Final: Wednesday, March 16, 12:15pm3:15pm

SCPD students welcome to attend lecture in person

- 4:15pm lecture time conveniently at end of day - Many parking spaces dont require permit after 4pm

Other notes for SCPD students:

- Please attend exams in person if possible - Feel free to use newsgroup to nd project partners

Exams open book, can bring copies of slides

- No electronic devices permitted

3/33

4/33

Course topics
Threads & Processes Concurrency & Synchronization Scheduling Virtual Memory I/O Disks, File systems, Network le systems Protection & Security Real-life operating systems Note: Lectures will often take Unix as an example
- Most current and future OSes heavily inuenced by Unix - Windows is exception; this quarter we will mostly ignore
5/33

Course goals
Introduce you to operating system concepts
- Hard to use a computer without interacting with OS - Understanding the OS makes you a more effective programmer

Cover important systems concepts in general

- Caching, concurrency, memory management, I/O, protection

Teach you to deal with larger software systems

- Programming assignments much larger than many courses - Warning: Many people will consider course very hard - In past, majority of people report 15 hours/week

Prepare you to take graduate OS classes (CS240, 240[a-z])

6/33

Programming Assignments
Implement parts of Pintos operating system
- Built for x86 hardware, you will use hardware emulator

Grading
No incompletes
- Talk to me ASAP if you run into real problems

One setup homework (lab 0) due Thursday Four implementation projects:

- Threads - Multiprogramming - Virtual memory - File system

50% of grade based on exams using this quantity: max (midterm > 0 ? nal : 0, (midterm + nal) /2) 50% of grade from projects
- For each project, 50% of score based on passing test cases - Remaining 50% based on design and style

Most peoples projects pass most test cases

- Please, please, please turn in working code, or no credit here

Lab 1 distributed at end of this week

- Attend section this Friday for project 1 overview

Means design and style matter a lot

- Large software systems not just about producing working code - Need to produce code other people can understand - Thats why we have group projects
7/33 8/33

Implement projects in groups of up to 3 people

- Pick your partners today - Lecture will end early so that you can do this

Style
Must turn in a design document along with code
- We supply you with templates for each projects design doc

Assignment requirements
Do not look at other peoples solutions to projects Can read but dont copy other OSes
- E.g., Linux, OpenBSD/FreeBSD, etc.

CAs will manually inspect code for correctness

- E.g., must actually implement the design - Must handle corner cases (e.g., handle malloc failure)

Cite any code that inspired your code

- As long as you cite what you used, its not cheating - Worst case we deduct points if it undermines the assignments

Will deduct points for error-prone code w/o errors

- Dont use global variables if automatic ones sufce - Dont use deceptive names for variables

Projects due on Thursdays at start of lecture

- Free extension to midnight if you attend lecture - Or if you are SCPD and watch lecture before midnight

Code must be easy to read

- Indent code, keep lines and (when possible) functions short - Use a uniform coding style (try to match existing code) - Put comments on structure members, globals, functions - Dont leave in reams of commented-out garbage code
9/33

Ask cs140-staff for extension if you run into trouble

- Be sure to tell us: How much have you done? How much is left? When can you nish by?
10/33

What is an operating system?

Layer between applications and hardware

Why study operating systems?

Operating systems are a maturing eld
- Most people use a handful of mature OSes - Hard to get people to switch operating systems - Hard to have impact with a new OS

High-performance servers are an OS issue

- Face many of the same issues as OSes

Makes hardware useful to the programmer [Usually] Provides abstractions for applications
- Manages and hides details of hardware - Accesses hardware through low/level interfaces unavailable to applications

Resource consumption is an OS issue

- Battery life, radio spectrum, etc.

Security is an OS issue
- Hard to achieve security without a solid foundation

[Often] Provides protection

- Prevents one process/user from clobbering another
11/33

New smart devices need new OSes Web browsers increasingly face OS issues
12/33

Primitive Operating Systems

Just a library of standard services [no protection]

Multitasking

Idea: Run more than one process at once

- When one process blocks (waiting for disk, network, user input, etc.) run another process - Standard interface above hardware-specic drivers, etc.

Problem: What can ill-behaved process do?

- Go into innite loop and never relinquish CPU - Scribble over other processes memory to make them fail

Simplifying assumptions
- System runs one program at a time - No bad users or programs (often bad assumption)

OS provides mechanisms to address these problems

- Preemption take CPU away from looping process - Memory protection protect processs memory from one another

Problem: Poor utilization

- . . . of hardware (e.g., CPU idle while waiting for disk) - . . . of human user (must wait for each program to nish)
13/33

14/33

Multitasking

Multi-user OSes

Idea: Run more than one process at once

- When one process blocks (waiting for disk, network, user input, etc.) run another process

Problem: What can ill-behaved process do?

- Go into innite loop and never relinquish CPU - Scribble over other processes memory to make them fail

Many OSes use protection to serve distrustful users Idea: With N users, system not N times slower
- Users demands for CPU, memory, etc. are bursty - Win by giving resources to users who actually need them

OS provides mechanisms to address these problems

- Preemption take CPU away from looping process - Memory protection protect processs memory from one another

What can go wrong?

- Users are gluttons, use too much CPU, etc. (need policies) - Total memory usage greater than in machine (must virtualize) - Super-linear slowdown with increasing demand (thrashing)
14/33 15/33

Multi-user OSes

Protection
Mechanisms that isolate bad programs and people Pre-emption:
- Give application a resource, take it away if needed elsewhere

Interposition: Many OSes use protection to serve distrustful users Idea: With N users, system not N times slower
- Users demands for CPU, memory, etc. are bursty - Win by giving resources to users who actually need them - Place OS between application and stuff - Track all pieces that application allowed to use (e.g., in table) - On every access, look in table to check that access legal

Privileged & unprivileged modes in CPUs :

- Applications unprivileged (user/unprivileged mode) - OS privileged (privileged/supervisor mode) - Protection operations can only be done in privileged mode

What can go wrong?

- Users are gluttons, use too much CPU, etc. (need policies) - Total memory usage greater than in machine (must virtualize) - Super-linear slowdown with increasing demand (thrashing)
15/33

16/33

Typical OS structure
P1
user kernel

System calls

P2

P3

P4

VM le IPC system sockets scheduler device TCP/IP device device

driver driver driver

network

console

disk

Most software runs as user-level processes (P[1-4]) OS kernel runs in privileged mode [shaded]
- Creates/deletes processes - Provides access to hardware
17/33

Applications can invoke kernel through system calls

- Special instruction transfers control to kernel - . . . which dispatches to one of few hundred syscall handlers
18/33

System calls (continued)

Goal: Do things app. cant do in unprivileged mode
- Like a library call, but into more privileged kernel code

System call example

Kernel supplies well-dened system call interface

- Applications set up syscall arguments and trap to kernel - Kernel performs operation and returns result

Higher-level functions built on syscall interface

- printf, scanf, gets, etc. all user-level code

Example: POSIX/UNIX interface

- open, close, read, write, ...

Standard library implemented in terms of syscalls

- printf in libc, has same privileges as application - calls write in kernel, which can send bits out serial port
19/33 20/33

UNIX le system calls

Applications open les (or devices) by name
- I/O happens through open les

Error returns
What if open fails? Returns -1 (invalid fd) Most system calls return -1 on failure
- Specic kind of error in global int errno

int open(char *path, int flags, /*mode*/...);

- flags: O RDONLY, O WRONLY, O RDWR - O CREAT: create the le if non-existent - O EXCL: (w. O CREAT) create if le exists already - O TRUNC: Truncate the le - O APPEND: Start writing from end of le - mode: nal argument with O CREAT

#include <sys/errno.h> for possible values

- 2 = ENOENT No such le or directory - 13 = EACCES Permission Denied

perror function prints human-readable message

- perror ("initfile"); initfile: No such file or directory

Returns le descriptorused for all I/O to le

21/33

22/33

Operations on le descriptors
int read (int fd, void *buf, int nbytes);
- Returns number of bytes read - Returns 0 bytes at end of le, or -1 on error

File descriptor numbers

File descriptors are inherited by processes
- When one process spawns another, same fds by default

Descriptors 0, 1, and 2 have special meaning

- 0 standard input (stdin in ANSI C) - 1 standard output (stdout, printf in ANSI C) - 2 standard error (stderr, perror in ANSI C) - Normally all three attached to terminal

int write (int fd, void *buf, int nbytes);

- Returns number of bytes written, -1 on error

off t lseek (int fd, off t pos, int whence);

- whence: 0 start, 1 current, 2 end Returns previous le offset, or -1 on error

Example: type.c
- Prints the contents of a le to stdout

int close (int fd);

23/33

24/33

type.c
void typefile (char *filename) { int fd, nread; char buf[1024]; fd = open (filename, O_RDONLY); if (fd == -1) { perror (filename); return; } while ((nread = read (fd, buf, sizeof (buf))) > 0) write (1, buf, nread); close (fd); }

Different system contexts

A system can typically be in one of several contexts User-level running an application Kernel process context
- Running kernel code on behalf of a particular process - E.g., performing system call - Also exception (mem. fault, numeric exception, etc.) - Or executing a kernel-only process (e.g., network le server)

Kernel code not associated w. a process

- Timer interrupt (hardclock) - Device interrupt - Softirqs, Tasklets (Linux-specic terms)

Context switch code changing address spaces

25/33 26/33

Transitions between contexts

User kernel process context: syscall, page fault User/process context interrupt handler: hardware Process context user/context switch: return Process context context switch: sleep Context switch user/process context

CPU preemption
Protection mechanism to prevent monopolizing CPU E.g., kernel programs timer to interrupt every 10 ms
- Must be in supervisor mode to write appropriate I/O registers - User code cannot re-program interval timer

Kernel sets interrupt to vector back to kernel

- Regains control whenever interval timer res - Gives CPU to another process if someone else needs it - Note: must be in supervisor mode to set interrupt entry points - No way for user code to hijack interrupt handler

Result: Cannot monopolize CPU with innite loop

- At worst get 1/N of CPU with N CPU-hungry processes

27/33

28/33

Protection is not security

How can you monopolize CPU?

Protection is not security

How can you monopolize CPU? Use multiple processes Until recently, could wedge many OSes with
int main() { while(1) fork(); } - Keeps creating more processes until system out of proc. slots

Other techniques: use all memory (chill program) Typically solved with technical/social combination
- Technical solution: Limit processes per user - Social: Reboot and yell at annoying users - Social: Pass laws (often debatable whether a good idea)

29/33

29/33

Address translation
Protect mem. of one program from actions of another Denitions
- Address space: all memory locations a program can name - Virtual address: addresses in process address space - Physical address: address of real memory - Translation: map virtual to physical addresses

More memory protection

CPU allows kernel-only virtual addresses
- Kernel typically part of all address spaces, e.g., to handle system call in same address space - But must ensure apps cant touch kernel memory

CPU lets OS disable virtual addresses

- Catch and halt buggy program that makes wild accesses - Make virtual memory seem bigger than physical (e.g., bring a page in from disk only when accessed)

Translation done on every load and store

- Modern CPUs do this in hardware for speed

CPU enforced read-only virtual addresses useful

- E.g., allows sharing of code pages between processes - Plus many other optimizations

Idea: If you cant name it, you cant touch it

- Ensure one processs translations dont include any other processs memory

CPU enforced execute disable of VAs

- Makes certain code injection attacks harder
30/33 31/33

Resource allocation & performance

Multitasking permits higher resource utilization Simple example:
- Process downloading large le mostly waits for network - You play a game while downloading the le - Higher CPU utilization than if just downloading

Useful properties to exploit

Skew
- 80% of time taken by 20% of code - 10% of memory absorbs 90% of references - Basis behind cache: place 10% in fast memory, 90% in slow, usually looks like one big fast memory

Past predicts future (a.k.a. temporal locality)

- Whats the best cache entry to replace? - If past = future, then least-recently-used entry

Complexity arises with cost of switching Example: Say disk 1,000 times slower than memory
- 1 GB memory in machine - 2 Processes want to run, each use 1 GB - Can switch processes by swapping them out to disk - Faster to run one at a time than keep context switching

Note conict between fairness & throughput

- Higher throughput (fewer cache misses, etc.) to keep running same process - But fairness says should periodically preempt CPU and give it to next process
32/33 33/33

type.c

Tue Jan 04 15:48:15 2011

#include <stdio.h> #include <unistd.h> #include <fcntl.h> void typefile (char *filename) { int fd, nread; char buf[1024]; fd = open (filename, O_RDONLY); if (fd == -1) { perror (filename); return; } while ((nread = read (fd, buf, sizeof (buf))) > 0) write (1, buf, nread); close (fd); } int main (int argc, char **argv) { int argno; for (argno = 1; argno < argc; argno++) typefile (argv[argno]); exit (0); }