Test

Contents At a Glance
I Advanced UNIX Programming with Linux

Advanced Linux Programming

1 Getting Started

3

2 Writing Good GNU/Linux Software 17 3 Processes 45 4 Threads 61 5 Interprocess Communication 95 II Mastering Linux 6 Devices 129 7 The /proc File System 147

8 Linux System Calls 167 9 Inline Assembly Code 189 10 Security 197 11 A Sample GNU/Linux Application 219 III Appendixes A Other Development Tools 259 B Low-Level I/O 281 C Table of Signals 301 D Online Resources 303 E Open Publication License Version 1.0 305 F GNU General Public License 309

Advanced Linux Programming
Mark Mitchell, Jeffrey Oldham, and Alex Samuel

www.newriders.com 201 West 103rd Street, Indianapolis, Indiana 46290 An Imprint of Pearson Education Boston • Indianapolis • London • Munich • New York • San Francisco

Advanced Linux Programming
Copyright © 2001 by New Riders Publishing FIRST EDITION: June, 2001 All rights reserved. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or by any information storage and retrieval system, without written permission from the publisher, except for the inclusion of brief quotations in a review. International Standard Book Number: 0-7357-1043-0 Library of Congress Catalog Card Number: 00-105343 05 04 03 02 01 7 6 5 4 3 2 1 Interpretation of the printing code:The rightmost doubledigit number is the year of the book’s printing; the rightmost single-digit number is the number of the book’s printing. For example, the printing code 01-1 shows that the ﬁrst printing of the book occurred in 2001. Composed in Bembo and MCPdigital by New Riders Publishing. Printed in the United States of America.

Publisher David Dwyer Associate Publisher Al Valvano Executive Editor Stephanie Wall Managing Editor Gina Brown Acquisitions Editor Ann Quinn Development Editor Laura Loveall Product Marketing Manager Stephanie Layton Publicity Manager Susan Petro Project Editor Caroline Wise Copy Editor Krista Hansing Senior Indexer Cheryl Lenser Manufacturing Coordinator Jim Conway Book Designer Louisa Klucznik Cover Designer Brainstorm Design, Inc. Cover Production Aren Howell Proofreader Debra Neel Composition Amy Parker

Trademarks
All terms mentioned in this book that are known to be trademarks or service marks have been appropriately capitalized. New Riders Publishing cannot attest to the accuracy of this information. Use of a term in this book should not be regarded as affecting the validity of any trademark or service mark. PostScript is a trademark of Adobe Systems, Inc. Linux is a trademark of Linus Torvalds.

Warning and Disclaimer
This book is designed to provide information about Advanced Linux Programming. Every effort has been made to make this book as complete and as accurate as possible, but no warranty or ﬁtness is implied. The information is provided on an as-is basis.The authors and New Riders Publishing shall have neither liability nor responsibility to any person or entity with respect to any loss or damages arising from the information contained in this book or from the use of the discs or programs that may accompany it.

Table of Contents
I Advanced UNIX Programming with Linux 1
1 Getting Started 3
1.1 Editing with Emacs 4 1.2 Compiling with GCC 6 1.3 Automating the Process with GNU Make 9 1.4 Debugging with GNU Debugger (GDB) 11 1.5 Finding More Information 13

2 Writing Good GNU/Linux Software 17
2.1 Interaction With the Execution Environment 17 2.2 Coding Defensively 30 2.3 Writing and Using Libraries 36

3 Processes
3.1 3.2 3.3 3.4

45

Looking at Processes 45 Creating Processes 48 Signals 52 Process Termination 55

4 Threads
4.1 4.2 4.3 4.4 4.5 4.6

61

Thread Creation 62 Thread Cancellation 69 Thread-Speciﬁc Data 72 Synchronization and Critical Sections 77 GNU/Linux Thread Implementation 92 Processes Vs.Threads 94

Contents

vii

5 Interprocess Communication
5.1 5.2 5.3 5.4 5.5 Shared Memory 96 Processes Semaphores 101 Mapped Memory 105 Pipes 110 Sockets 116

95

II Mastering Linux 127
6 Devices
6.1 6.2 6.3 6.4 6.5 6.6 6.7

129

Device Types 130 Device Numbers 130 Device Entries 131 Hardware Devices 133 Special Devices 136 PTYs 142 ioctl 144

7 The /proc File System
7.1 7.2 7.3 7.4 7.5 7.6

147

Extracting Information from /proc 148 Process Entries 150 Hardware Information 158 Kernel Information 160 Drives, Mounts, and File Systems 161 System Statistics 165

8 Linux System Calls

167

8.1 Using strace 168 8.2 access:Testing File Permissions 169 8.3 fcntl: Locks and Other File Operations 171 8.4 fsync and fdatasync: Flushing Disk Buffers 173 8.5 getrlimit and setrlimit: Resource Limits 174 8.6 getrusage: Process Statistics 175 8.7 gettimeofday:Wall-Clock Time 176

viii

Contents

8.8 The mlock Family: Locking Physical Memory 177 8.9 mprotect: Setting Memory Permissions 179 8.10 nanosleep: High-Precision Sleeping 181 8.11 readlink: Reading Symbolic Links 182 8.12 sendﬁle: Fast Data Transfers 183 8.13 setitimer: Setting Interval Timers 185 8.14 sysinfo: Obtaining System Statistics 186 8.15 uname 187

9 Inline Assembly Code
9.1 9.2 9.3 9.4 9.5 9.6

189

When to Use Assembly Code 190 Simple Inline Assembly 191 Extended Assembly Syntax 192 Example 194 Optimization Issues 196 Maintenance and Portability Issues 196

10 Security

197

10.1 Users and Groups 198 10.2 Process User IDs and Process Group IDs 199 10.3 File System Permissions 200 10.4 Real and Effective IDs 205 10.5 Authenticating Users 208 10.6 More Security Holes 211

11 A Sample GNU/Linux Application 219
11.1 11.2 11.3 11.4 11.5 Overview 219 Implementation 221 Modules 239 Using the Server 252 Finishing Up 255

Contents

ix

III Appendixes

257
259

A Other Development Tools

A.1 Static Program Analysis 259 A.2 Finding Dynamic Memory Errors 261 A.3 Proﬁling 269

B Low-Level I/O
B.1 B.2 B.3 B.4

281

Reading and Writing Data 282 stat 291 Vector Reads and Writes 293 Relation to Standard C Library I/O Functions 295 B.5 Other File Operations 296 B.6 Reading Directory Contents 296

C Table of Signals D Online Resources

301 303

D.1 General Information 303 D.2 Information About GNU/Linux Software 304 D.3 Other Sites 304

E Open Publication License Version 1.0 305
I. Requirement on Both Unmodiﬁed and Modiﬁed Versions 305 II. Copyright 306 III. Scope of License 306 IV. Requirements on Modiﬁed Works 306 V. Good-Practice Recommendations 306 VI. License Options 307 Open Publication Policy Appendix 307

x

Contents

F GNU General Public License

309

Preamble 309 Terms and Conditions for Copying, Distribution and Modiﬁcation 310 End of Terms and Conditions 315 How to Apply These Terms to Your New Programs 315

Index

317

Table of Program Listings
1.1 1.2 1.3 2.1 2.2 2.3 main.c (C source ﬁle), 6 reciprocal.cpp (C++ source ﬁle), 6 reciprocal.hpp (header ﬁle), 7 arglist.c (argc and argv parameters), 18 getopt_long.c (getopt_long function), 21 print_env.c (printing execution environment), 26 2.4 client.c (network client program), 26 2.5 temp_ﬁle.c (mkstemp function), 28 2.6 readﬁle.c (resource allocation during error checking), 35 2.7 test.c (library contents), 37 2.8 app.c (program with library functions), 37 2.9 tifftest.c (libtiff library), 40 3.1 print-pid.c (printing process IDs), 46 3.2 system.c (system function), 48 3.3 fork.c (fork function), 49 3.4 fork-exec.c (fork and exec functions), 51 3.5 sigusr1.c (signal handlers), 54 3.6 zombie.c (zombie processes), 58 3.7 sigchld.c (cleaning up child processes), 60 4.1 thread-create.c (creating threads), 63 4.2 thread-create2 (creating two threads), 64 4.3 thread-create2.c (revised main function), 65 4.4 primes.c (prime number computation in a thread), 67 4.5 detached.c (creating detached threads), 69 4.6 critical-section.c (critical sections), 71 4.7 tsd.c (thread-speciﬁc data), 73 4.8 cleanup.c (cleanup handlers), 75 4.9 cxx-exit.cpp (C++ thread cleanup), 76 4.10 job-queue1.c (thread race conditions), 78 4.11 job-queue2.c (mutexes), 80 4.12 job-queue3.c (semaphores), 84 4.13 spin-condvar.c (condition variables), 87

Program Listings

xiii

4.14 condvar.c (condition variables), 90 4.15 thread-pid (printing thread process IDs), 92 5.1 shm.c (shared memory), 99 5.2 sem_all_deall.c (semaphore allocation and deallocation), 102 5.3 sem_init.c (semaphore initialization), 102 5.4 sem_pv.c (semaphore wait and post operations), 104 5.5 mmap-write.c (mapped memory), 106 5.6 mmap-read.c (mapped memory), 107 5.7 pipe.c (parent-child process communication), 111 5.8 dup2.c (output redirection), 113 5.9 popen.c (popen command), 114 5.10 socket-server.c (local sockets), 120 5.11 socket-client.c (local sockets), 121 5.12 socket-inet.c (Internet-domain sockets), 124 6.1 random_number.c (random number generation), 138 6.2 cdrom-eject.c (ioctl example), 144 7.1 clock-speed.c (cpu clock speed from /proc/cpuinfo), 149 7.2 get-pid.c (process ID from /proc/self), 151 7.3 print-arg-list.c (printing process argument lists), 153 7.4 print-environment.c (process environment), 154 7.5 get-exe-path.c (program executable path), 155 7.6 open-and-spin.c (opening ﬁles), 157 7.7 print-uptime.c (system uptime and idle time), 165 8.1 check-access.c (ﬁle access permissions), 170 8.2 lock-ﬁle.c (write locks), 171 8.3 write_journal_entry.c (data buffer ﬂushing), 173 8.4 limit-cpu.c (resource limits), 175 8.5 print-cpu-times.c (process statistics), 176

xiv

Program Listings

8.6 print-time.c (date/time printing), 177 8.7 mprotect.c (memory access), 180 8.8 better_sleep.c (high-precision sleep), 182 8.9 print-symlink.c (symbolic links), 183 8.10 copy.c (sendﬁle system call), 184 8.11 itemer.c (interal timers), 185 8.12 sysinfo.c (system statistics), 187 8.13 print-uname (version number and hardware information), 188 9.1 bit-pos-loop.c (bit position with loop), 194 9.2 bit-pos-asm.c (bit position with bsrl), 195 10.1 simpleid.c (printing user and group IDs), 200 10.2 stat-perm.c (viewing ﬁle permissions with stat system call), 202 10.3 setuid-test.c (setuid programs), 207 10.4 pam.c (PAM example), 209 10.5 temp-ﬁle.c (temporary ﬁle creation), 214 10.6 grep-dictionary.c (word search), 216 11.1 server.h (function and variable declarations), 222 11.2 common.c (utility functions), 223 11.3 module.c (loading server modules), 226 11.4 server.c (server implementation), 228 11.5 main.c (main server program), 235 11.6 time.c (show wall-clock time), 239 11.7 issue.c (GNU/Linux distribution information), 240 11.8 diskfree.c (free disk space information), 242 11.9 processes.c (summarizing running processes), 244 11.10 Makeﬁle (Makeﬁle for sample application program), 252

Program Listings

xv

A.1 hello.c (Hello World), 260 A.2 malloc-use.c (dynamic memory allocation), 267 A.3 calculator.c (main calculator program), 274 A.4 number.c (unary number implementation), 276 A.5 stack.c (unary number stack), 279 A.6 deﬁnitions.h (header ﬁle for calculator program), 280 B.1 create-ﬁle.c (create a new ﬁle), 284 B.2 timestamp.c (append a timestamp), 285 B.3 write-all.c (write all buffered data), 286 B.4 hexdump.c (print a hexadecimal ﬁle dump), 287 B.5 lseek-huge.c (creating large ﬁles), 289 B.6 read-ﬁle.c (reading ﬁles into buffers), 292 B.7 write-args.c (writev function), 294 B.8 listdir.c (printing directory listings), 297

About the Authors
Mark Mitchell received a bachelor of arts degree in computer science from Harvard in 1994 and a master of science degree from Stanford in 1999. His research interests centered on computational complexity and computer security. Mark has participated substantially in the development of the GNU Compiler Collection, and he has a strong interest in developing quality software. Jeffrey Oldham received a bachelor of arts degree in computer science from Rice University in 1991. After working at the Center for Research on Parallel Computation, he obtained a doctor of philosophy degree from Stanford in 2000. His research interests center on algorithm engineering, concentrating on ﬂow and other combinatorial algorithms. He works on GCC and scientiﬁc computing software. Alex Samuel graduated from Harvard in 1995 with a degree in physics. He worked as a software engineer at BBN before returning to study physics at Caltech and the Stanford Linear Accelerator Center. Alex administers the Software Carpentry project and works on various other projects, such as optimizations in GCC. Mark and Alex founded CodeSourcery LLC together in 1999. Jeffrey joined the company in 2000. CodeSourcery’s mission is to provide development tools for GNU/Linux and other operating systems; to make the GNU tool chain a commercial-quality, standards-conforming development tool set; and to provide general consulting and engineering services. CodeSourcery’s Web site is http://www.codesourcery.com.

xvi

About the Technical Reviewers
These reviewers contributed their considerable hands-on expertise to the entire development process for Advanced Linux Programming. As the book was being written, these dedicated professionals reviewed all the material for technical content, organization, and ﬂow.Their feedback was critical to ensuring that Advanced Linux Programming ﬁts our reader’s need for the highest quality technical information. Glenn Becker has many degrees, all in theatre. He presently works as an online producer for SCIFI.COM, the online component of the SCI FI channel, in New York City. At home he runs Debian GNU/Linux and obsesses about such topics as system administration, security, software internationalization, and XML. John Dean received a BSc(Hons) from the University of Shefﬁeld in 1974, in pure science. As an undergraduate at Shefﬁeld, John developed his interest in computing. In 1986 he received a MSc from Cranﬁeld Institute of Science and Technology in Control Engineering.While working for Roll Royce and Associates, John became involved in developing control software for computer-aided inspection equipment of nuclear steam-raising plants. Since leaving RR&A in 1978, he has worked in the petrochemical industry developing and maintaining process control software. John worked a volunteer software developer for MySQL from 1996 until May 2000, when he joined MySQL as a full-time employee. John’s area of responsibility is MySQL on MS Windows and developing a new MySQL GUI client using Trolltech’s Qt GUI application toolkit on both Windows and platforms that run X-11.

xvii

Acknowledgments
We greatly appreciate the pioneering work of Richard Stallman, without whom there would never have been the GNU Project, and of Linus Torvalds, without whom there would never have been the Linux kernel. Countless others have worked on parts of the GNU/Linux operating system, and we thank them all. We thank the faculties of Harvard and Rice for our undergraduate educations, and Caltech and Stanford for our graduate training. Without all who taught us, we would never have dared to teach others! W. Richard Stevens wrote three excellent books on UNIX programming, and we have consulted them extensively. Roland McGrath, Ulrich Drepper, and many others wrote the GNU C library and its outstanding documentation. Robert Brazile and Sam Kendall reviewed early outlines of this book and made wonderful suggestions about tone and content. Our technical editors and reviewers (especially Glenn Becker and John Dean) pointed out errors, made suggestions, and provided continuous encouragement. Of course, any errors that remain are no fault of theirs! Thanks to Ann Quinn, of New Riders, for handling all the details involved in publishing a book; Laura Loveall, also of New Riders, for not letting us fall too far behind on our deadlines; and Stephanie Wall, also of New Riders, for encouraging us to write this book in the ﬁrst place!

xviii

Tell Us What You Think
As the reader of this book, you are the most important critic and commentator.We value your opinion and want to know what we’re doing right, what we could do better, what areas you’d like to see us publish in, and any other words of wisdom you’re willing to pass our way. As the Executive Editor for the Web Development team at New Riders Publishing, I welcome your comments.You can fax, email, or write me directly to let me know what you did or didn’t like about this book—as well as what we can do to make our books stronger. Please note that I cannot help you with technical problems related to the topic of this book, and that due to the high volume of mail I receive, I might not be able to reply to every message. When you write, please be sure to include this book’s title and author, as well as your name and phone or fax number. I will carefully review your comments and share them with the author and editors who worked on the book. Fax: 317-581-4663 Email: Stephanie.Wall@newriders.com Mail: Stephanie Wall Executive Editor New Riders Publishing 201 West 103rd Street Indianapolis, IN 46290 USA

xix

Introduction
GNU/Linux has taken the world of computers by storm. At one time, personal computer users were forced to choose among proprietary operating environments and applications. Users had no way of ﬁxing or improving these programs, could not look “under the hood,” and were often forced to accept restrictive licenses. GNU/Linux and other open source systems have changed that—now PC users, administrators, and developers can choose a free operating environment complete with tools, applications, and full source code. A great deal of the success of GNU/Linux is owed to its open source nature. Because the source code for programs is publicly available, everyone can take part in development, whether by ﬁxing a small bug or by developing and distributing a complete major application.This opportunity has enticed thousands of capable developers worldwide to contribute new components and improvements to GNU/Linux, to the point that modern GNU/Linux systems rival the features of any proprietary system, and distributions include thousands of programs and applications spanning many CDROMs or DVDs. The success of GNU/Linux has also validated much of the UNIX philosophy. Many of the application programming interfaces (APIs) introduced in AT&T and BSD UNIX variants survive in Linux and form the foundation on which programs are built.The UNIX philosophy of many small command line-oriented programs working together is the organizational principle that makes GNU/Linux so powerful. Even when these programs are wrapped in easy-to-use graphical user interfaces, the underlying commands are still available for power users and automated scripts. A powerful GNU/Linux application harnesses the power of these APIs and commands in its inner workings. GNU/Linux’s APIs provide access to sophisticated features such as interprocess communication, multithreading, and high-performance networking. And many problems can be solved simply by assembling existing commands and programs using simple scripts.

GNU and Linux
Where did the name GNU/Liux come from? You’ve certainly heard of Linux before, and you may have heard of the GNU Project.You may not have heard the name GNU/Linux, although you’re probably familiar with the system it refers to. Linux is named after Linus Torvalds, the creator and original author of the kernel that runs a GNU/Linux system.The kernel is the program that performs the most basic functions of an operating system: It controls and interfaces with the computer’s hardware, handles allocation of memory and other resources, allows multiple programs to run at the same time, manages the ﬁle system, and so on.

xx

The kernel by itself doesn’t provide features that are useful to users. It can’t even provide a simple prompt for users to enter basic commands. It provides no way for users to manage or edit ﬁles, communicate with other computers, or write other programs.These tasks require the use of a wide array of other programs, including command shells, ﬁle utilities, editors, and compilers. Many of these programs, in turn, use libraries of general-purpose functions, such as the library containing standard C library functions, which are not included in the kernel. On GNU/Linux systems, many of these other programs and libraries are software developed as part of the GNU Project.1 A great deal of this software predates the Linux kernel.The aim of the GNU Project is “to develop a complete UNIX-like operating system which is free software” (from the GNU Project Web site, http://www.gnu.org). The Linux kernel and software from the GNU Project has proven to be a powerful combination. Although the combination is often called “Linux” for short, the complete system couldn’t work without GNU software, any more than it could operate without the kernel. For this reason, throughout this book we’ll refer to the complete system as GNU/Linux, except when we are speciﬁcally talking about the Linux kernel.

The GNU General Public License
The source code contained in this book is covered by the GNU General Public License (GPL), which is listed in Appendix F, “GNU General Public License.” A great deal of free software, especially GNU/Linux software, is licensed under it. For instance, the Linux kernel itself is licensed under the GPL, as are many other GNU programs and libraries you’ll ﬁnd in GNU/Linux distributions. If you use the source code in this book, be sure to read and understand the terms of the GPL. The GNU Project Web site includes an extensive discussion of the GPL (http://www.gnu.org/copyleft/) and other free software licenses.You can ﬁnd information about open source software licenses at http://www.opensource.org/ licenses/index.html.

Who Should Read This Book?
This book is intended for three types of readers: You might be a developer already experienced with programming for the GNU/Linux system, and you want to learn about some of its advanced features and capabilities.You might be interested in writing more sophisticated programs with features such as multiprocessing, multithreading, interprocess communication, and interaction with hardware devices.You might want to improve your programs by making them run faster, more reliably, and more securely, or by designing them to interact better with the rest of the GNU/Linux system. n 1. GNU is a recursive acronym: It stands for “GNU’s Not UNIX.”

xxi

n

n

You might be a developer experienced with another UNIX-like system who’s interested in developing GNU/Linux software, too.You might already be familiar with standard APIs such as those in the POSIX speciﬁcation.To develop GNU/Linux software, you need to know the peculiarities of the system, its limitations, additional capabilities, and conventions. You might be a developer making the transition from a non-UNIX environment, such as Microsoft’s Win32 platform.You might already be familiar with the general principles of writing good software, but you need to know the speciﬁc techniques that GNU/Linux programs use to interact with the system and with each other. And you want to make sure your programs ﬁt naturally into the GNU/Linux system and behave as users expect them to.

This book is not intended to be a comprehensive guide or reference to all aspects of GNU/Linux programming. Instead, we’ll take a tutorial approach, introducing the most important concepts and techniques, and giving examples of how to use them. Section 1.5, “Finding More Information,” in Chapter 1, “Getting Started,” contains references to additional documentation, where you can obtain complete details about these and other aspects of GNU/Linux programming. Because this is a book about advanced topics, we’ll assume that you are already familiar with the C programming language and that you know how to use the standard C library functions in your programs.The C language is the most widely used language for developing GNU/Linux software; most of the commands and libraries that we discuss in this book, and most of the Linux kernel itself, are written in C. The information in this book is equally applicable to C++ programs because that language is roughly a superset of C. Even if you program in another language, you’ll ﬁnd this information useful because C language APIs and conventions are the lingua franca of GNU/Linux. If you’ve programmed on another UNIX-like system platform before, chances are good that you already know your way around Linux’s low-level I/O functions (open, read, stat, and so on).These are different from the standard C library’s I/O functions (fopen, fprintf, fscanf, and so on). Both are useful in GNU/Linux programming, and we use both sets of I/O functions throughout this book. If you’re not familiar with the low-level I/O functions, jump to the end of the book and read Appendix B, “Low-Level I/O,” before you start Chapter 2, “Writing Good GNU/Linux Software.”

xxii

This book does not provide a general introduction to GNU/Linux systems. We assume that you already have a basic knowledge of how to interact with a GNU/Linux system and perform basic operations in graphical and command-line environments. If you’re new to GNU/Linux, start with one of the many excellent introductory books, such as Michael Tolber’s Inside Linux (New Riders Publishing, 2001).

Conventions
This book follows a few typographical conventions: n n

n

n

A new term is set in italics the ﬁrst time it is introduced. Program text, functions, variables, and other “computer language” are set in a ﬁxed-pitch font—for example, printf (“Hello, world!\bksl n”). Names of commands, ﬁles, and directories are also set in a ﬁxed-pitch font—for example, cd /. When we show interactions with a command shell, we use % as the shell prompt (your shell is probably conﬁgured to use a different prompt). Everything after the prompt is what you type, while other lines of text are the system’s response. For example, in this interaction
% uname Linux

n

the system prompted you with %.You entered the uname command.The system responded by printing Linux. The title of each source code listing includes a ﬁlename in parentheses. If you type in the listing, save it to a ﬁle by this name.You can also download the source code listings from the Advanced Linux Programming Web site (http://www.newriders.com or http://www.advancedlinuxprogramming.com).

We wrote this book and developed the programs listed in it using the Red Hat 6.2 distribution of GNU/Linux.This distribution incorporates release 2.2.14 of the Linux kernel, release 2.1.3 of the GNU C library, and the EGCS 1.1.2 release of the GNU C compiler.The information and programs in this book should generally be applicable to other versions and distributions of GNU/Linux as well, including 2.4 releases of the Linux kernel and 2.2 releases of the GNU C library.

xxiii

I
Advanced UNIX Programming with Linux
1 2 3 4 5 Getting Started Writing Good GNU/Linux Software Processes Threads Interprocess Communication

1
Getting Started

HIS CHAPTER SHOWS YOU HOW TO PERFORM THE BASIC steps required to create a C or C++ Linux program. In particular, this chapter shows you how to create and modify C and C++ source code, compile that code, and debug the result. If you’re already accustomed to programming under Linux, you can skip ahead to Chapter 2, “Writing Good GNU/Linux Software;” pay careful attention to Section 2.3, “Writing and Using Libraries,” for information about static versus dynamic linking that you might not already know. Throughout this book, we’ll assume that you’re familiar with the C or C++ programming languages and the most common functions in the standard C library.The source code examples in this book are in C, except when demonstrating a particular feature or complication of C++ programming.We also assume that you know how to perform basic operations in the Linux command shell, such as creating directories and copying files. Because many Linux programmers got started programming in the Windows environment, we’ll occasionally point out similarities and contrasts between Windows and Linux.

T

4

Chapter 1

Getting Started

1.1 Editing with Emacs
An editor is the program that you use to edit source code. Lots of different editors are available for Linux, but the most popular and full-featured editor is probably GNU Emacs.
About Emacs
Emacs is much more than an editor. It is an incredibly powerful program, so much so that at CodeSourcery, it is affectionately known as the One True Program, or just the OTP for short. You can read and send email from within Emacs, and you can customize and extend Emacs in ways far too numerous to discuss here. You can even browse the Web from within Emacs!

If you’re familiar with another editor, you can certainly use it instead. Nothing in the rest of this book depends on using Emacs. If you don’t already have a favorite Linux editor, then you should follow along with the mini-tutorial given here. If you like Emacs and want to learn about its advanced features, you might consider reading one of the many Emacs books available. One excellent tutorial, Learning GNU Emacs, is written by Debra Cameron, Bill Rosenblatt, and Eric S. Raymond (O’Reilly, 1996).

1.1.1 Opening a C or C++ Source File
You can start Emacs by typing emacs in your terminal window and pressing the Return key.When Emacs has been started, you can use the menus at the top to create a new source file. Click the Files menu, choose Open Files, and then type the name of the file that you want to open in the “minibuffer” at the bottom of the screen.1 If you want to create a C source file, use a filename that ends in .c or .h. If you want to create a C++ source file, use a filename that ends in .cpp, .hpp, .cxx, .hxx, .C, or .H. When the file is open, you can type as you would in any ordinary word-processing program.To save the file, choose the Save Buffer entry on the Files menu.When you’re finished using Emacs, you can choose the Exit Emacs option on the Files menu. If you don’t like to point and click, you can use keyboard shortcuts to automatically open files, save files, and exit Emacs.To open a file, type C-x C-f. (The C-x means to hold down the Control key and then press the x key.) To save a file, type C-x C-s.To exit Emacs, just type C-x C-c. If you want to get a little better acquainted with Emacs, choose the Emacs Tutorial entry on the Help menu.The tutorial provides you with lots of tips on how to use Emacs effectively.

1. If you’re not running in an X Window system, you’ll have to press F10 to access the menus.

1.1

Editing with Emacs

5

1.1.2 Automatic Formatting
If you’re accustomed to programming in an Integrated Development Environment (IDE), you’ll also be accustomed to having the editor help you format your code. Emacs can provide the same kind of functionality. If you open a C or C++ source file, Emacs automatically figures out that the file contains source code, not just ordinary text. If you hit the Tab key on a blank line, Emacs moves the cursor to an appropriately indented point. If you hit the Tab key on a line that already contains some text, Emacs indents the text. So, for example, suppose that you have typed in the following: int main () { printf (“Hello, world\n”); }

If you press the Tab key on the line with the call to code to look like this: int main () { printf (“Hello, world\n”); }

printf, Emacs

will reformat your

Notice how the line has been appropriately indented. As you use Emacs more, you’ll see how it can help you perform all kinds of complicated formatting tasks. If you’re ambitious, you can program Emacs to perform literally any kind of automatic formatting you can imagine. People have used this facility to implement Emacs modes for editing just about every kind of document, to implement games2, and to implement database front ends.

1.1.3 Syntax Highlighting
In addition to formatting your code, Emacs can make it easier to read C and C++ code by coloring different syntax elements. For example, Emacs can turn keywords one color, built-in types such as int another color, and comments another color. Using color makes it a lot easier to spot some common syntax errors. The easiest way to turn on colorization is to edit the file ~/.emacs and insert the following string:
(global-font-lock-mode t)

Save the file, exit Emacs, and restart. Now open a C or C++ source file and enjoy! You might have noticed that the string you inserted into your .emacs looks like code from the LISP programming language.That’s because it is LISP code! Much of Emacs is actually written in LISP. You can add functionality to Emacs by writing more LISP code.

2.Try running the command M-x dunnet if you want to play an old-fashioned text adventure game.

6

Chapter 1

Getting Started

1.2 Compiling with GCC
A compiler turns human-readable source code into machine-readable object code that can actually run.The compilers of choice on Linux systems are all part of the GNU Compiler Collection, usually known as GCC.3 GCC also include compilers for C, C++, Java, Objective-C, Fortran, and Chill.This book focuses mostly on C and C++ programming. Suppose that you have a project like the one in Listing 1.2 with one C++ source file (reciprocal.cpp) and one C source file (main.c) like in Listing 1.1.These two files are supposed to be compiled and then linked together to produce a program called reciprocal.4 This program will compute the reciprocal of an integer.
Listing 1.1 (main.c) C source file—main.c
#include #include “reciprocal.hpp” int main (int argc, char **argv) { int i; i = atoi (argv[1]); printf (“The reciprocal of %d is %g\n”, i, reciprocal (i)); return 0; }

Listing 1.2 (reciprocal.cpp) C++ source file—reciprocal.cpp
#include #include “reciprocal.hpp” double reciprocal (int i) { // I should be non-zero. assert (i != 0); return 1.0/i; }

3. For more information about GCC, visit

http://gcc.gnu.org.

4. In Windows, executables usually have names that end in .exe. Linux programs, on the other hand, usually have no extension. So, the Windows equivalent of this program would probably be called reciprocal.exe; the Linux version is just plain reciprocal.

1.2

Compiling with GCC

7

There’s also one header file called

reciprocal.hpp

(see Listing 1.3).

Listing 1.3 (reciprocal.hpp) Header file—reciprocal.hpp
#ifdef __cplusplus extern “C” { #endif extern double reciprocal (int i);

#ifdef __cplusplus } #endif

The first step is to turn the C and C++ source code into object code.

1.2.1 Compiling a Single Source File
The name of the C compiler is gcc. To compile a C source file, you use the -c option. So, for example, entering this at the command prompt compiles the main.c source file:
% gcc -c main.c

The resulting object file is named main.o. The C++ compiler is called g++. Its operation is very similar to reciprocal.cpp is accomplished by entering the following:
% g++ -c reciprocal.cpp

gcc; compiling

The -c option tells g++ to compile the program to an object file only; without it, g++ will attempt to link the program to produce an executable. After you’ve typed this command, you’ll have an object file called reciprocal.o. You’ll probably need a couple other options to build any reasonably large program. The -I option is used to tell GCC where to search for header files. By default, GCC looks in the current directory and in the directories where headers for the standard libraries are installed. If you need to include header files from somewhere else, you’ll need the -I option. For example, suppose that your project has one directory called src, for source files, and another called include.You would compile reciprocal.cpp like this to indicate that g++ should use the ../include directory in addition to find reciprocal.hpp:
% g++ -c -I ../include reciprocal.cpp

8

Chapter 1

Getting Started

Sometimes you’ll want to define macros on the command line. For example, in production code, you don’t want the overhead of the assertion check present in reciprocal.cpp; that’s only there to help you debug the program.You turn off the check by defining the macro NDEBUG. You could add an explicit #define to reciprocal.cpp, but that would require changing the source itself. It’s easier to simply define NDEBUG on the command line, like this:
% g++ -c -D NDEBUG reciprocal.cpp

If you had wanted to define something like this:

NDEBUG

to some particular value, you could have done

% g++ -c -D NDEBUG=3 reciprocal.cpp

If you’re really building production code, you probably want to have GCC optimize the code so that it runs as quickly as possible.You can do this by using the -O2 command-line option. (GCC has several different levels of optimization; the second level is appropriate for most programs.) For example, the following compiles reciprocal.cpp with optimization turned on:
% g++ -c -O2 reciprocal.cpp

Note that compiling with optimization can make your program more difficult to debug with a debugger (see Section 1.4, “Debugging with GDB”). Also, in certain instances, compiling with optimization can uncover bugs in your program that did not manifest themselves previously. You can pass lots of other options to gcc and g++.The best way to get a complete list is to view the online documentation.You can do this by typing the following at your command prompt:
% info gcc

1.2.2 Linking Object Files
Now that you’ve compiled main.c and utilities.cpp, you’ll want to link them.You should always use g++ to link a program that contains C++ code, even if it also contains C code. If your program contains only C code, you should use gcc instead. Because this program contains both C and C++, you should use g++, like this:
% g++ -o reciprocal main.o reciprocal.o

The -o option gives the name of the file to generate as output from the link step. Now you can run reciprocal like this:
% ./reciprocal 7 The reciprocal of 7 is 0.142857

As you can see, g++ has automatically linked in the standard C runtime library containing the implementation of printf. If you had needed to link in another library (such as a graphical user interface toolkit), you would have specified the library with

1.3

Automating the Process with GNU Make

9

the -l option. In Linux, library names almost always start with lib. For example, the Pluggable Authentication Module (PAM) library is called libpam.a.To link in libpam.a, you use a command like this:
% g++ -o reciprocal main.o reciprocal.o -lpam

The compiler automatically adds the lib prefix and the .a suffix. As with header files, the linker looks for libraries in some standard places, including the /lib and /usr/lib directories that contain the standard system libraries. If you want the linker to search other directories as well, you should use the -L option, which is the parallel of the -I option discussed earlier. You can use this line to instruct the linker to look for libraries in the /usr/local/lib/pam directory before looking in the usual places:
% g++ -o reciprocal main.o reciprocal.o -L/usr/local/lib/pam -lpam

Although you don’t have to use the -I option to get the preprocessor to search the current directory, you do have to use the -L option to get the linker to search the current directory. In particular, you could use the following to instruct the linker to find the test library in the current directory:
% gcc -o app app.o -L. -ltest

1.3 Automating the Process with GNU Make
If you’re accustomed to programming for the Windows operating system, you’re probably accustomed to working with an Integrated Development Environment (IDE).You add sources files to your project, and then the IDE builds your project automatically. Although IDEs are available for Linux, this book doesn’t discuss them. Instead, this book shows you how to use GNU Make to automatically recompile your code, which is what most Linux programmers actually do. The basic idea behind make is simple.You tell make what targets you want to build and then give rules explaining how to build them.You also specify dependencies that indicate when a particular target should be rebuilt. In our sample reciprocal project, there are three obvious targets: reciprocal.o, main.o, and the reciprocal itself. You already have rules in mind for building these targets in the form of the command lines given previously.The dependencies require a little bit of thought. Clearly, reciprocal depends on reciprocal.o and main.o because you can’t link the complete program until you have built each of the object files.The object files should be rebuilt whenever the corresponding source files change.There’s one more twist in that a change to reciprocal.hpp also should cause both of the object files to be rebuilt because both source files include that header file. In addition to the obvious targets, there should always be a clean target.This target removes all the generated object files and programs so that you can start fresh.The rule for this target uses the rm command to remove the files.

10

Chapter 1

Getting Started

You can convey all that information to make by putting the information in a file named Makefile. Here’s what Makefile contains: reciprocal: main.o reciprocal.o g++ $(CFLAGS) -o reciprocal main.o reciprocal.o main.o: main.c reciprocal.hpp gcc $(CFLAGS) -c main.c reciprocal.o: reciprocal.cpp reciprocal.hpp g++ $(CFLAGS) -c reciprocal.cpp clean: rm -f *.o reciprocal

You can see that targets are listed on the left, followed by a colon and then any dependencies.The rule to build that target is on the next line. (Ignore the $(CFLAGS) bit for the moment.) The line with the rule on it must start with a Tab character, or make will get confused. If you edit your Makefile in Emacs, Emacs will help you with the formatting. If you remove the object files that you’ve already built, and just type
% make

on the command-line, you’ll see the following:
% make gcc -c main.c g++ -c reciprocal.cpp g++ -o reciprocal main.o reciprocal.o

You can see that make has automatically built the object files and then linked them. If you now change main.c in some trivial way and type make again, you’ll see the following:
% make gcc -c main.c g++ -o reciprocal main.o reciprocal.o

You can see that make knew to rebuild main.o and to re-link the program, but it didn’t bother to recompile reciprocal.cpp because none of the dependencies for reciprocal.o had changed. The $(CFLAGS) is a make variable.You can define this variable either in the Makefile itself or on the command line. GNU make will substitute the value of the variable when it executes the rule. So, for example, to recompile with optimization enabled, you would do this:
% make clean rm -f *.o reciprocal % make CFLAGS=-O2 gcc -O2 -c main.c g++ -O2 -c reciprocal.cpp g++ -O2 -o reciprocal main.o reciprocal.o

1.4

Debugging with GNU Debugger (GDB)

11

Note that the -O2 flag was inserted in place of $(CFLAGS) in the rules. In this section, you’ve seen only the most basic capabilities of make.You can find out more by typing this:
% info make

In that manual, you’ll find information about how to make maintaining a Makefile easier, how to reduce the number of rules that you need to write, and how to automatically compute dependencies.You can also find more information in GNU, Autoconf, Automake, and Libtool by Gary V.Vaughan, Ben Elliston,Tom Tromey, and Ian Lance Taylor (New Riders Publishing, 2000).

1.4 Debugging with GNU Debugger (GDB)
The debugger is the program that you use to figure out why your program isn’t behaving the way you think it should.You’ll be doing this a lot.5 The GNU Debugger (GDB) is the debugger used by most Linux programmers.You can use GDB to step through your code, set breakpoints, and examine the value of local variables.

1.4.1 Compiling with Debugging Information
To use GDB, you’ll have to compile with debugging information enabled. Do this by adding the -g switch on the compilation command line. If you’re using a Makefile as described previously, you can just set CFLAGS equal to -g when you run make, as shown here:
% make gcc -g g++ -g g++ -g CFLAGS=-g -c main.c -c reciprocal.cpp -o reciprocal main.o reciprocal.o

When you compile with -g, the compiler includes extra information in the object files and executables.The debugger uses this information to figure out which addresses correspond to which lines in which source files, how to print out local variables, and so forth.

1.4.2 Running GDB
You can start up When gdb gdb

by typing:

% gdb reciprocal

starts up, you should see the GDB prompt:

(gdb)

5. …unless your programs always work the first time.

12

Chapter 1

Getting Started

The first step is to run your program inside the debugger. Just enter the command run and any program arguments.Try running the program without any arguments, like this:
(gdb) run Starting program: reciprocal Program received signal SIGSEGV, Segmentation fault. __strtol_internal (nptr=0x0, endptr=0x0, base=10, group=0) at strtol.c:287 287 strtol.c: No such file or directory. (gdb)

The problem is that there is no error-checking code in main.The program expects one argument, but in this case the program was run with no arguments.The SIGSEGV message indicates a program crash. GDB knows that the actual crash happened in a function called __strtol_internal.That function is in the standard library, and the source isn’t installed, which explains the “No such file or directory” message.You can see the stack by using the where command:
(gdb) where #0 __strtol_internal (nptr=0x0, endptr=0x0, base=10, group=0) at strtol.c:287 #1 0x40096fb6 in atoi (nptr=0x0) at ../stdlib/stdlib.h:251 #2 0x804863e in main (argc=1, argv=0xbffff5e4) at main.c:8

You can see from this display that main called the atoi function with a NULL pointer, which is the source of the trouble. You can go up two levels in the stack until you reach main by using the up command:
(gdb) up 2 #2 0x804863e in main (argc=1, argv=0xbffff5e4) at main.c:8 8 i = atoi (argv[1]);

Note that gdb is capable of finding the source for main.c, and it shows the line where the erroneous function call occurred.You can view the value of variables using the print command:
(gdb) print argv[1] $2 = 0x0

That confirms that the problem is indeed a NULL pointer passed into You can set a breakpoint by using the break command:
(gdb) break main Breakpoint 1 at 0x804862e: file main.c, line 8.

atoi.

1.5

Finding More Information

13

This command sets a breakpoint on the first line of program with an argument, like this:
(gdb) run 7 Starting program: reciprocal 7

main.6

Now try rerunning the

Breakpoint 1, main (argc=2, argv=0xbffff5e4) at main.c:8 8 i = atoi (argv[1]);

You can see that the debugger has stopped at the breakpoint. You can step over the call to atoi using the next command:
(gdb) next 9 printf (“The reciprocal of %d is %g\n”, i, reciprocal (i));

If you want to see what’s going on inside
(gdb) step reciprocal (i=7) at reciprocal.cpp:6 6 assert (i != 0);

reciprocal, use

the

step

command like this:

You’re now in the body of the reciprocal function. You might find it more convenient to run gdb from within Emacs rather than using gdb directly from the command line. Use the command M-x gdb to start up gdb in an Emacs window. If you are stopped at a breakpoint, Emacs automatically pulls up the appropriate source file. It’s easier to figure out what’s going on when you’re looking at the whole file rather than just one line of text.

1.5 Finding More Information
Nearly every Linux distribution comes with a great deal of useful documentation.You could learn most of what we’ll talk about in this book by reading documentation in your Linux distribution (although it would probably take you much longer).The documentation isn’t always well-organized, though, so the tricky part is finding what you need. Documentation is also sometimes out-of-date, so take everything that you read with a grain of salt. If the system doesn’t behave the way a man page (manual pages) says it should, for instance, it may be that the man page is outdated. To help you navigate, here are the most useful sources of information about advanced Linux programming.

6. Some people have commented that saying break main is a little bit funny because usually you want to do this only when main is already broken.

14

Chapter 1

Getting Started

1.5.1 Man Pages
Linux distributions include man pages for most standard commands, system calls, and standard library functions.The man pages are divided into numbered sections; for programmers, the most important are these: (1) User commands (2) System calls (3) Standard library functions (8) System/administrative commands The numbers denote man page sections. Linux’s man pages come installed on your system; use the man command to access them.To look up a man page, simply invoke man name, where name is a command or function name. In a few cases, the same name occurs in more than one section; you can specify the section explicitly by placing the section number before the name. For example, if you type the following, you’ll get the man page for the sleep command (in section 1 of the Linux man pages):
% man sleep

To see the man page for the
% man 3 sleep

sleep

library function, use this command:

Each man page includes a one-line summary of the command or function.The whatis name command displays all man pages (in all sections) for a command or function matching name. If you’re not sure which command or function you want, you can perform a keyword search on the summary lines, using man -k keyword. Man pages include a lot of very useful information and should be the first place you turn for help.The man page for a command describes command-line options and arguments, input and output, error codes, configuration, and the like.The man page for a system call or library function describes parameters and return values, lists error codes and side effects, and specifies which include file to use if you call the function.

1.5.2

Info

The Info documentation system contains more detailed documentation for many core components of the GNU/Linux system, plus several other programs. Info pages are hypertext documents, similar to Web pages.To launch the text-based Info browser, just type info in a shell window.You’ll be presented with a menu of Info documents installed on your system. (Press Control+H to display the keys for navigating an Info document.) Among the most useful Info documents are these: gcc—The gcc compiler libc—The GNU C library, including many system calls gdb—The GNU debugger n n n

1.5

Finding More Information

15

n

emacs—The

n

Emacs text editor info—The Info system itself

Almost all the standard Linux programming tools (including ld, the linker; as, the assembler; and gprof, the profiler) come with useful Info pages.You can jump directly to a particular Info document by specifying the page name on the command line:
% info libc

If you do most of your programming in Emacs, you can access the built-in Info browser by typing M-x info or C-h i.

1.5.3 Header Files
You can learn a lot about the system functions that are available and how to use them by looking at the system header files.These reside in /usr/include and /usr/include/sys. If you are getting compile errors from using a system call, for instance, take a look in the corresponding header file to verify that the function’s signature is the same as what’s listed in the man page. On Linux systems, a lot of the nitty-gritty details of how the system calls work are reflected in header files in the directories /usr/include/bits, /usr/include/asm, and /usr/include/linux. For instance, the numerical values of signals (described in Section 3.3, “Signals,” in Chapter 3, “Processes”) are defined in /usr/include/bits/signum.h. These header files make good reading for inquiring minds. Don’t include them directly in your programs, though; always use the header files in /usr/include or as mentioned in the man page for the function you’re using.

1.5.4

Source Code

This is Open Source, right? The final arbiter of how the system works is the system source code itself, and luckily for Linux programmers, that source code is freely available. Chances are, your Linux distribution includes full source code for the entire system and all programs included with it; if not, you’re entitled under the terms of the GNU General Public License to request it from the distributor. (The source code might not be installed on your disk, though. See your distribution’s documentation for instructions on installing it.) The source code for the Linux kernel itself is usually stored under /usr/src/linux. If this book leaves you thirsting for details of how processes, shared memory, and system devices work, you can always learn straight from the source code. Most of the system functions described in this book are implemented in the GNU C library; check your distribution’s documentation for the location of the C library source code.

2
Writing Good GNU/Linux Software

T

HIS CHAPTER COVERS SOME BASIC TECHNIQUES THAT MOST GNU/Linux programmers use. By following the guidelines presented, you’ll be able to write programs that work well within the GNU/Linux environment and meet GNU/Linux users’ expectations of how programs should operate.

2.1 Interaction With the Execution Environment
When you first studied C or C++, you learned that the special main function is the primary entry point for a program.When the operating system executes your program, it automatically provides certain facilities that help the program communicate with the operating system and the user.You probably learned about the two parameters to main, usually called argc and argv, which receive inputs to your program. You learned about the stdout and stdin (or the cout and cin streams in C++) that provide console input and output.These features are provided by the C and C++ languages, and they interact with the GNU/Linux system in certain ways. GNU/ Linux provides other ways for interacting with the operating environment, too.

18

Chapter 2

Writing Good GNU/Linux Software

2.1.1 The Argument List
You run a program from a shell prompt by typing the name of the program. Optionally, you can supply additional information to the program by typing one or more words after the program name, separated by spaces.These are called command-line arguments. (You can also include an argument that contains a space, by enclosing the argument in quotes.) More generally, this is referred to as the program’s argument list because it need not originate from a shell command line. In Chapter 3, “Processes,” you’ll see another way of invoking a program, in which a program can specify the argument list of another program directly. When a program is invoked from the shell, the argument list contains the entire command line, including the name of the program and any command-line arguments that may have been provided. Suppose, for example, that you invoke the ls command in your shell to display the contents of the root directory and corresponding file sizes with this command line:
% ls -s /

The argument list that the ls program receives has three elements.The first one is the name of the program itself, as specified on the command line, namely ls.The second and third elements of the argument list are the two command-line arguments, -s and /. The main function of your program can access the argument list via the argc and argv parameters to main (if you don’t use them, you may simply omit them).The first parameter, argc, is an integer that is set to the number of items in the argument list. The second parameter, argv, is an array of character pointers.The size of the array is argc, and the array elements point to the elements of the argument list, as NULterminated character strings. Using command-line arguments is as easy as examining the contents of argc and argv. If you’re not interested in the name of the program itself, don’t forget to skip the first element. Listing 2.1 demonstrates how to use argc and argv.
Listing 2.1 (arglist.c) Using argc and argv
#include int main (int argc, char* argv[]) { printf (“The name of this program is ‘%s’.\n”, argv[0]); printf (“This program was invoked with %d arguments.\n”, argc - 1); /* Were any command-line arguments specified? if (argc > 1) { /* Yes, print them. */ int i; printf (“The arguments are:\n”); for (i = 1; i < argc; ++i) */

2.1

Interaction With the Execution Environment

19

printf (“ } return 0; }

%s\n”, argv[i]);

2.1.2 GNU/Linux Command-Line Conventions
Almost all GNU/Linux programs obey some conventions about how command-line arguments are interpreted.The arguments that programs expect fall into two categories: options (or flags) and other arguments. Options modify how the program behaves, while other arguments provide inputs (for instance, the names of input files). Options come in two forms: n n

Short options consist of a single hyphen and a single character (usually a lowercase or uppercase letter). Short options are quicker to type. Long options consist of two hyphens, followed by a name made of lowercase and uppercase letters and hyphens. Long options are easier to remember and easier to read (in shell scripts, for instance).

Usually, a program provides both a short form and a long form for most options it supports, the former for brevity and the latter for clarity. For example, most programs understand the options -h and --help, and treat them identically. Normally, when a program is invoked from the shell, any desired options follow the program name immediately. Some options expect an argument immediately following. Many programs, for example, interpret the option --output foo to specify that output of the program should be placed in a file named foo. After the options, there may follow other command-line arguments, typically input files or input data. For example, the command ls -s / displays the contents of the root directory.The -s option modifies the default behavior of ls by instructing it to display the size (in kilobytes) of each entry.The / argument tells ls which directory to list.The --size option is synonymous with -s, so the same command could have been invoked as ls --size /. The GNU Coding Standards list the names of some commonly used command-line options. If you plan to provide any options similar to these, it’s a good idea to use the names specified in the coding standards.Your program will behave more like other programs and will be easier for users to learn.You can view the GNU Coding Standards’ guidelines for command-line options by invoking the following from a shell prompt on most GNU/Linux systems:
% info “(standards)User Interfaces”

20

Chapter 2

Writing Good GNU/Linux Software

2.1.3 Using getopt_long
Parsing command-line options is a tedious chore. Luckily, the GNU C library provides a function that you can use in C and C++ programs to make this job somewhat easier (although still a bit annoying).This function, getopt_long, understands both short and long options. If you use this function, include the header file . Suppose, for example, that you are writing a program that is to accept the three options shown in Table 2.1.
Table 2.1
-h -o filename -v

Example Program Options Long Form
--help --output filename --verbose

Short Form

Purpose Display usage summary and exit Specify output filename Print verbose messages

In addition, the program is to accept zero or more additional command-line arguments, which are the names of input files. To use getopt_long, you must provide two data structures.The first is a character string containing the valid short options, each a single letter. An option that requires an argument is followed by a colon. For your program, the string ho:v indicates that the valid options are -h, -o, and -v, with the second of these options followed by an argument. To specify the available long options, you construct an array of struct option elements. Each element corresponds to one long option and has four fields. In normal circumstances, the first field is the name of the long option (as a character string, without the two hyphens); the second is 1 if the option takes an argument, or 0 otherwise; the third is NULL; and the fourth is a character constant specifying the short option synonym for that long option.The last element of the array should be all zeros.You could construct the array like this: const struct option long_options[] = { { “help”, 0, NULL, ‘h’ }, { “output”, 1, NULL, ‘o’ }, { “verbose”, 0, NULL, ‘v’ }, { NULL, 0, NULL, 0 } };

You invoke the getopt_long function, passing it the argc and argv arguments to main, the character string describing short options, and the array of struct option elements describing the long options. Each time you call getopt_long, it parses a single option, returning the shortoption letter for that option, or –1 if no more options are found. Typically, you’ll call getopt_long in a loop, to process all the options the user has specified, and you’ll handle the specific options in a switch statement. n n

2.1

Interaction With the Execution Environment

21

n

n

n

If getopt_long encounters an invalid option (an option that you didn’t specify as a valid short or long option), it prints an error message and returns the character ? (a question mark). Most programs will exit in response to this, possibly after displaying usage information. When handling an option that takes an argument, the global variable optarg points to the text of that argument. After getopt_long has finished parsing all the options, the global variable optind contains the index (into argv) of the first nonoption argument. getopt_long Listing 2.2 shows an example of how you might use arguments.
Listing 2.2 (getopt_long.c) Using getopt_long
#include #include #include /* The name of this program. const char* program_name; */

to process your

/* Prints usage information for this program to STREAM (typically stdout or stderr), and exit the program with EXIT_CODE. Does not return. */ void print_usage (FILE* stream, int exit_code) { fprintf (stream, “Usage: %s options [ inputfile ... ]\n”, program_name); fprintf (stream, “ -h --help Display this usage information.\n” “ -o --output filename Write output to file.\n” “ -v --verbose Print verbose messages.\n”); exit (exit_code); } /* Main program entry point. ARGC contains number of argument list elements; ARGV is an array of pointers to them. */ int main (int argc, char* argv[]) { int next_option; /* A string listing valid short options letters. const char* const short_options = “ho:v”; /* An array describing valid long options. */ const struct option long_options[] = { { “help”, 0, NULL, ‘h’ }, { “output”, 1, NULL, ‘o’ }, { “verbose”, 0, NULL, ‘v’ }, */

continues

22

Chapter 2

Writing Good GNU/Linux Software

Listing 2.2 Continued
{ NULL, }; /* The name of the file to receive program output, or NULL for standard output. */ const char* output_filename = NULL; /* Whether to display verbose messages. */ int verbose = 0; /* Remember the name of the program, to incorporate in messages. The name is stored in argv[0]. */ program_name = argv[0]; do { next_option = getopt_long (argc, argv, short_options, long_options, NULL); switch (next_option) { case ‘h’: /* -h or --help */ /* User has requested usage information. Print it to standard output, and exit with exit code zero (normal termination). */ print_usage (stdout, 0); case ‘o’: /* -o or --output */ /* This option takes an argument, the name of the output file. output_filename = optarg; break; case ‘v’: /* -v or verbose = 1; break; --verbose */ 0, NULL, 0 } /* Required at end of array. */

*/

case ‘?’: /* The user specified an invalid option. */ /* Print usage information to standard error, and exit with exit code one (indicating abnormal termination). */ print_usage (stderr, 1); case -1: break; /* Done with options. */

default: /* Something else: unexpected. abort (); } } while (next_option != -1);

*/

/* Done with options. OPTIND points to first nonoption argument. For demonstration purposes, print them if the verbose option was specified. */

2.1

Interaction With the Execution Environment

23

if (verbose) { int i; for (i = optind; i < argc; ++i) printf (“Argument: %s\n”, argv[i]); } /* The main program goes here. return 0; } */

Using getopt_long may seem like a lot of work, but writing code to parse the command-line options yourself would take even longer.The getopt_long function is very sophisticated and allows great flexibility in specifying what kind of options to accept. However, it’s a good idea to stay away from the more advanced features and stick with the basic option structure described.

2.1.4 Standard I/O
The standard C library provides standard input and output streams (stdin and stdout, respectively).These are used by scanf, printf, and other library functions. In the UNIX tradition, use of standard input and output is customary for GNU/Linux programs.This allows the chaining of multiple programs using shell pipes and input and output redirection. (See the man page for your shell to learn its syntax.) The C library also provides stderr, the standard error stream. Programs should print warning and error messages to standard error instead of standard output.This allows users to separate normal output and error messages, for instance, by redirecting standard output to a file while allowing standard error to print on the console.The fprintf function can be used to print to stderr, for example: fprintf (stderr, (“Error: ...”));

These three streams are also accessible with the underlying UNIX I/O commands (read, write, and so on) via file descriptors.These are file descriptors 0 for stdin, 1 for stdout, and 2 for stderr. When invoking a program, it is sometimes useful to redirect both standard output and standard error to a file or pipe.The syntax for doing this varies among shells; for Bourne-style shells (including bash, the default shell on most GNU/Linux distributions), the syntax is this:
% program > output_file.txt 2>&1 % program 2>&1 | filter

The 2>&1 syntax indicates that file descriptor 2 (stderr) should be merged into file descriptor 1 (stdout). Note that 2>&1 must follow a file redirection (the first example) but must precede a pipe redirection (the second example).

24

Chapter 2

Writing Good GNU/Linux Software

Note that stdout is buffered. Data written to stdout is not sent to the console (or other device, if it’s redirected) until the buffer fills, the program exits normally, or stdout is closed.You can explicitly flush the buffer by calling the following: fflush (stdout);

In contrast, stderr is not buffered; data written to

stderr

goes directly to the console.1

This can produce some surprising results. For example, this loop does not print one period every second; instead, the periods are buffered, and a bunch of them are printed together when the buffer fills. while (1) { printf (“.”); sleep (1); }

In this loop, however, the periods do appear once a second: while (1) { fprintf (stderr, “.”); sleep (1); }

2.1.5 Program Exit Codes
When a program ends, it indicates its status with an exit code.The exit code is a small integer; by convention, an exit code of zero denotes successful execution, while nonzero exit codes indicate that an error occurred. Some programs use different nonzero exit code values to distinguish specific errors. With most shells, it’s possible to obtain the exit code of the most recently executed program using the special $? variable. Here’s an example in which the ls command is invoked twice and its exit code is printed after each invocation. In the first case, ls executes correctly and returns the exit code zero. In the second case, ls encounters an error (because the filename specified on the command line does not exist) and thus returns a nonzero exit code.
% ls / bin coda etc lib misc nfs proc boot dev home lost+found mnt opt root % echo $? 0 % ls bogusfile ls: bogusfile: No such file or directory % echo $? 1 sbin tmp usr var

1. In C++, the same distinction holds for cout and cerr, respectively. Note that the endl token flushes a stream in addition to printing a newline character; if you don’t want to flush the stream (for performance reasons, for example), use a newline constant, ‘\n’, instead.

2.1

Interaction With the Execution Environment

25

A C or C++ program specifies its exit code by returning that value from the main function.There are other methods of providing exit codes, and special exit codes are assigned to programs that terminate abnormally (by a signal).These are discussed further in Chapter 3.

2.1.6 The Environment
GNU/Linux provides each running program with an environment.The environment is a collection of variable/value pairs. Both environment variable names and their values are character strings. By convention, environment variable names are spelled in all capital letters. You’re probably familiar with several common environment variables already. For instance: n n

n

n

contains your username. HOME contains the path to your home directory. PATH contains a colon-separated list of directories through which Linux searches for commands you invoke. DISPLAY contains the name and display number of the X Window server on which windows from graphical X Window programs will appear.
USER

Your shell, like any other program, has an environment. Shells provide methods for examining and modifying the environment directly.To print the current environment in your shell, invoke the printenv program.Various shells have different built-in syntax for using environment variables; the following is the syntax for Bourne-style shells. The shell automatically creates a shell variable for each environment variable that it finds, so you can access environment variable values using the $varname syntax. For instance: n % echo $USER samuel % echo $HOME /home/samuel n You can use the export command to export a shell variable into the environment. For example, to set the EDITOR environment variable, you would use this:
% EDITOR=emacs % export EDITOR

Or, for short:
% export EDITOR=emacs

26

Chapter 2

Writing Good GNU/Linux Software

In a program, you access an environment variable with the getenv function in .That function takes a variable name and returns the corresponding value as a character string, or NULL if that variable is not defined in the environment.To set or clear environment variables, use the setenv and unsetenv functions, respectively. Enumerating all the variables in the environment is a little trickier.To do this, you must access a special global variable named environ, which is defined in the GNU C library.This variable, of type char**, is a NULL-terminated array of pointers to character strings. Each string contains one environment variable, in the form VARIABLE=value. The program in Listing 2.3, for instance, simply prints the entire environment by looping through the environ array.
Listing 2.3 ( print-env.c) Printing the Execution Environment
#include /* The ENVIRON variable contains the environment. extern char** environ; int main () { char** var; for (var = environ; *var != NULL; ++var) printf (“%s\n”, *var); return 0; } */

Don’t modify environ yourself; use the setenv and unsetenv functions instead. Usually, when a new program is started, it inherits a copy of the environment of the program that invoked it (the shell program, if it was invoked interactively). So, for instance, programs that you run from the shell may examine the values of environment variables that you set in the shell. Environment variables are commonly used to communicate configuration information to programs. Suppose, for example, that you are writing a program that connects to an Internet server to obtain some information.You could write the program so that the server name is specified on the command line. However, suppose that the server name is not something that users will change very often.You can use a special environment variable—say SERVER_NAME—to specify the server name; if that variable doesn’t exist, a default value is used. Part of your program might look as shown in Listing 2.4.
Listing 2.4 (client.c) Part of a Network Client Program
#include #include int main () {

2.1

Interaction With the Execution Environment

27

char* server_name = getenv (“SERVER_NAME”); if (server_name == NULL) /* The SERVER_NAME environment variable was not set. default. */ server_name = “server.my-company.com”; printf (“accessing server %s\n”, server_name); /* Access the server here... */ return 0; }

Use the

Suppose that this program is named client. Assuming that you haven’t set the SERVER_NAME variable, the default value for the server name is used:
% client accessing server server.my-company.com

But it’s easy to specify a different server:
% export SERVER_NAME=backup-server.elsewhere.net % client accessing server backup-server.elsewhere.net

2.1.7 Using Temporary Files
Sometimes a program needs to make a temporary file, to store large data for a while or to pass data to another program. On GNU/Linux systems, temporary files are stored in the /tmp directory.When using temporary files, you should be aware of the following pitfalls: More than one instance of your program may be run simultaneously (by the same user or by different users).The instances should use different temporary filenames so that they don’t collide. The file permissions of the temporary file should be set in such a way that unauthorized users cannot alter the program’s execution by modifying or replacing the temporary file. n n n

Temporary filenames should be generated in a way that cannot be predicted externally; otherwise, an attacker can exploit the delay between testing whether a given name is already in use and opening a new temporary file.

GNU/Linux provides functions, mkstemp and tmpfile, that take care of these issues for you (in addition to several functions that don’t).Which you use depends on whether you plan to hand the temporary file to another program, and whether you want to use UNIX I/O (open, write, and so on) or the C library’s stream I/O functions (fopen, fprintf, and so on).

28

Chapter 2

Writing Good GNU/Linux Software

Using mkstemp The mkstemp function creates a unique temporary filename from a filename template, creates the file with permissions so that only the current user can access it, and opens the file for read/write.The filename template is a character string ending with “XXXXXX” (six capital X’s); mkstemp replaces the X’s with characters so that the filename is unique.The return value is a file descriptor; use the write family of functions to write to the temporary file. Temporary files created with mkstemp are not deleted automatically. It’s up to you to remove the temporary file when it’s no longer needed. (Programmers should be very careful to clean up temporary files; otherwise, the /tmp file system will fill up eventually, rendering the system inoperable.) If the temporary file is for internal use only and won’t be handed to another program, it’s a good idea to call unlink on the temporary file immediately.The unlink function removes the directory entry corresponding to a file, but because files in a file system are reference-counted, the file itself is not removed until there are no open file descriptors for that file, either.This way, your program may continue to use the temporary file, and the file goes away automatically as soon as you close the file descriptor. Because Linux closes file descriptors when a program ends, the temporary file will be removed even if your program terminates abnormally. The pair of functions in Listing 2.5 demonstrates mkstemp. Used together, these functions make it easy to write a memory buffer to a temporary file (so that memory can be freed or reused) and then read it back later.
Listing 2.5 (temp_file.c) Using mkstemp
#include #include /* A handle for a temporary file created with write_temp_file. this implementation, it’s just a file descriptor. */ typedef int temp_file_handle; In

/* Writes LENGTH bytes from BUFFER into a temporary file. The temporary file is immediately unlinked. Returns a handle to the temporary file. */ temp_file_handle write_temp_file (char* buffer, size_t length) { /* Create the filename and file. The XXXXXX will be replaced with characters that make the filename unique. */ char temp_filename[] = “/tmp/temp_file.XXXXXX”; int fd = mkstemp (temp_filename); /* Unlink the file immediately, so that it will be removed when the file descriptor is closed. */ unlink (temp_filename); /* Write the number of bytes to the file first. */ write (fd, &length, sizeof (length));

2.1

Interaction With the Execution Environment

29

/* Now write the data itself. */ write (fd, buffer, length); /* Use the file descriptor as the handle for the temporary file. return fd; }

*/

/* Reads the contents of a temporary file TEMP_FILE created with write_temp_file. The return value is a newly allocated buffer of those contents, which the caller must deallocate with free. *LENGTH is set to the size of the contents, in bytes. The temporary file is removed. */ char* read_temp_file (temp_file_handle temp_file, size_t* length) { char* buffer; /* The TEMP_FILE handle is a file descriptor to the temporary file. */ int fd = temp_file; /* Rewind to the beginning of the file. */ lseek (fd, 0, SEEK_SET); /* Read the size of the data in the temporary file. */ read (fd, length, sizeof (*length)); /* Allocate a buffer and read the data. */ buffer = (char*) malloc (*length); read (fd, buffer, *length); /* Close the file descriptor, which will cause the temporary file to go away. */ close (fd); return buffer; }

Using tmpfile If you are using the C library I/O functions and don’t need to pass the temporary file to another program, you can use the tmpfile function.This creates and opens a temporary file, and returns a file pointer to it.The temporary file is already unlinked, as in the previous example, so it is deleted automatically when the file pointer is closed (with fclose) or when the program terminates. GNU/Linux provides several other functions for generating temporary files and temporary filenames, including mktemp, tmpnam, and tempnam. Don’t use these functions, though, because they suffer from the reliability and security problems already mentioned.

30

Chapter 2

Writing Good GNU/Linux Software

2.2 Coding Defensively
Writing programs that run correctly under “normal” use is hard; writing programs that behave gracefully in failure situations is harder.This section demonstrates some coding techniques for finding bugs early and for detecting and recovering from problems in a running program. The code samples presented later in this book deliberately skip extensive error checking and recovery code because this would obscure the basic functionality being presented. However, the final example in Chapter 11, “A Sample GNU/Linux Application,” comes back to demonstrating how to use these techniques to write robust programs.

2.2.1 Using assert
A good objective to keep in mind when coding application programs is that bugs or unexpected errors should cause the program to fail dramatically, as early as possible. This will help you find bugs earlier in the development and testing cycles. Failures that don’t exhibit themselves dramatically are often missed and don’t show up until the application is in users’ hands. One of the simplest methods to check for unexpected conditions is the standard C assert macro.The argument to this macro is a Boolean expression.The program is terminated if the expression evaluates to false, after printing an error message containing the source file and line number and the text of the expression.The assert macro is very useful for a wide variety of consistency checks internal to a program. For instance, use assert to test the validity of function arguments, to test preconditions and postconditions of function calls (and method calls, in C++), and to test for unexpected return values. Each use of assert serves not only as a runtime check of a condition, but also as documentation about the program’s operation within the source code. If your program contains an assert (condition) that says to someone reading your source code that condition should always be true at that point in the program, and if condition is not true, it’s probably a bug in the program. For performance-critical code, runtime checks such as uses of assert can impose a significant performance penalty. In these cases, you can compile your code with the NDEBUG macro defined, by using the -DNDEBUG flag on your compiler command line. With NDEBUG set, appearances of the assert macro will be preprocessed away. It’s a good idea to do this only when necessary for performance reasons, though, and only with performance-critical source files. Because it is possible to preprocess assert macros away, be careful that any expression you use with assert has no side effects. Specifically, you shouldn’t call functions inside assert expressions, assign variables, or use modifying operators such as ++.

2.2

Coding Defensively

31

Suppose, for example, that you call a function, do_something, repeatedly in a loop. The do_something function returns zero on success and nonzero on failure, but you don’t expect it ever to fail in your program.You might be tempted to write: for (i = 0; i < 100; ++i) assert (do_something () == 0);

However, you might find that this runtime check imposes too large a performance penalty and decide later to recompile with NDEBUG defined.This will remove the assert call entirely, so the expression will never be evaluated and do_something will never be called.You should write this instead: for (i = 0; i < 100; ++i) { int status = do_something (); assert (status == 0); }

Another thing to bear in mind is that you should not use assert to test for invalid user input. Users don’t like it when applications simply crash with a cryptic error message, even in response to invalid input.You should still always check for invalid input and produce sensible error messages in response input. Use assert for internal runtime checks only. Some good places to use assert are these: Check against null pointers, for instance, as invalid function arguments.The error message generated by {assert (pointer != NULL)}, n Assertion ‘pointer != ((void *)0)’ failed.

is more informative than the error message that would result if your program dereferenced a null pointer:
Segmentation fault (core dumped) n Check conditions on function parameter values. For instance, if a function should be called only with a positive value for parameter foo, use this at the beginning of the function body: assert (foo > 0);

This will help you detect misuses of the function, and it also makes it very clear to someone reading the function’s source code that there is a restriction on the parameter’s value. Don’t hold back; use assert liberally throughout your programs.

32

Chapter 2

Writing Good GNU/Linux Software

2.2.2 System Call Failures
Most of us were originally taught how to write programs that execute to completion along a well-defined path.We divide the program into tasks and subtasks, and each function completes a task by invoking other functions to perform corresponding subtasks. Given appropriate inputs, we expect a function to produce the correct output and side effects. The realities of computer hardware and software intrude into this idealized dream. Computers have limited resources; hardware fails; many programs execute at the same time; users and programmers make mistakes. It’s often at the boundary between the application and the operating system that these realities exhibit themselves.Therefore, when using system calls to access system resources, to perform I/O, or for other purposes, it’s important to understand not only what happens when the call succeeds, but also how and when the call can fail. System calls can fail in many ways. For example: n n

n

n

n

The system can run out of resources (or the program can exceed the resource limits enforced by the system of a single program). For example, the program might try to allocate too much memory, to write too much to a disk, or to open too many files at the same time. Linux may block a certain system call when a program attempts to perform an operation for which it does not have permission. For example, a program might attempt to write to a file marked read-only, to access the memory of another process, or to kill another user’s program. The arguments to a system call might be invalid, either because the user provided invalid input or because of a program bug. For instance, the program might pass an invalid memory address or an invalid file descriptor to a system call. Or, a program might attempt to open a directory as an ordinary file, or might pass the name of an ordinary file to a system call that expects a directory. A system call can fail for reasons external to a program.This happens most often when a system call accesses a hardware device.The device might be faulty or might not support a particular operation, or perhaps a disk is not inserted in the drive. A system call can sometimes be interrupted by an external event, such as the delivery of a signal.This might not indicate outright failure, but it is the responsibility of the calling program to restart the system call, if desired.

In a well-written program that makes extensive use of system calls, it is often the case that more code is devoted to detecting and handling errors and other exceptional circumstances than to the main work of the program.

2.2

Coding Defensively

33

2.2.3 Error Codes from System Calls
A majority of system calls return zero if the operation succeeds, or a nonzero value if the operation fails. (Many, though, have different return value conventions; for instance, malloc returns a null pointer to indicate failure. Always read the man page carefully when using a system call.) Although this information may be enough to determine whether the program should continue execution as usual, it probably does not provide enough information for a sensible recovery from errors. Most system calls use a special variable named errno to store additional information in case of failure.2 When a call fails, the system sets errno to a value indicating what went wrong. Because all system calls use the same errno variable to store error information, you should copy the value into another variable immediately after the failed call.The value of errno will be overwritten the next time you make a system call. Error values are integers; possible values are given by preprocessor macros, by convention named in all capitals and starting with “E”—for example, EACCES and EINVAL. Always use these macros to refer to errno values rather than integer values. Include the header if you use errno values. GNU/Linux provides a convenient function, strerror, that returns a character string description of an errno error code, suitable for use in error messages. Include if you use strerror. GNU/Linux also provides perror, which prints the error description directly to the stderr stream. Pass to perror a character string prefix to print before the error description, which should usually include the name of the function that failed. Include if you use perror. This code fragment attempts to open a file; if the open fails, it prints an error message and exits the program. Note that the open call returns an open file descriptor if the open operation succeeds, or –1 if the operation fails. fd = open (“inputfile.txt”, O_RDONLY); if (fd == -1) { /* The open failed. Print an error message and exit. */ fprintf (stderr, “error opening file: %s\n”, strerror (errno)); exit (1); }

Depending on your program and the nature of the system call, the appropriate action in case of failure might be to print an error message, to cancel an operation, to abort the program, to try again, or even to ignore the error. It’s important, though, to include logic that handles all possible failure modes in some way or another.

2. Actually, for reasons of thread safety, errno is implemented as a macro, but it is used like a global variable.

34

Chapter 2

Writing Good GNU/Linux Software

One possible error code that you should be on the watch for, especially with I/O functions, is EINTR. Some functions, such as read, select, and sleep, can take significant time to execute.These are considered blocking functions because program execution is blocked until the call is completed. However, if the program receives a signal while blocked in one of these calls, the call will return without completing the operation. In this case, errno is set to EINTR. Usually, you’ll want to retry the system call in this case. Here’s a code fragment that uses the chown call to change the owner of a file given by path to the user by user_id. If the call fails, the program takes action depending on the value of errno. Notice that when we detect what’s probably a bug in the program, we exit using abort or assert, which cause a core file to be generated.This can be useful for post-mortem debugging. For other unrecoverable errors, such as out-ofmemory conditions, we exit using exit and a nonzero exit value instead because a core file wouldn’t be very useful. rval = chown (path, user_id, -1); if (rval != 0) { /* Save errno because it’s clobbered by the next system call. */ int error_code = errno; /* The operation didn’t succeed; chown should return -1 on error. */ assert (rval == -1); /* Check the value of errno, and take appropriate action. */ switch (error_code) { case EPERM: /* Permission denied. */ case EROFS: /* PATH is on a read-only file system. */ case ENAMETOOLONG: /* PATH is too long. */ case ENOENT: /* PATH does not exit. */ case ENOTDIR: /* A component of PATH is not a directory. */ case EACCES: /* A component of PATH is not accessible. */ /* Something’s wrong with the file. Print an error message. */ fprintf (stderr, “error changing ownership of %s: %s\n”, path, strerror (error_code)); /* Don’t end the program; perhaps give the user a chance to choose another file... */ break; case EFAULT: /* PATH contains an invalid memory address. abort ();

This is probably a bug.

*/

case ENOMEM: /* Ran out of kernel memory. */ fprintf (stderr, “%s\n”, strerror (error_code)); exit (1); default: /* Some other, unexpected, error code. We’ve tried to handle all possible error codes; if we’ve missed one, that’s a bug! */ abort (); }; }

2.2

Coding Defensively

35

You could simply have used this code, which behaves the same way if the call succeeds: rval = chown (path, user_id, -1); assert (rval == 0);

But if the call fails, this alternative makes no effort to report, handle, or recover from errors. Whether you use the first form, the second form, or something in between depends on the error detection and recovery requirements for your program.

2.2.4 Errors and Resource Allocation
Often, when a system call fails, it’s appropriate to cancel the current operation but not to terminate the program because it may be possible to recover from the error. One way to do this is to return from the current function, passing a return code to the caller indicating the error. If you decide to return from the middle of a function, it’s important to make sure that any resources successfully allocated previously in the function are first deallocated. These resources can include memory, file descriptors, file pointers, temporary files, synchronization objects, and so on. Otherwise, if your program continues running, the resources allocated before the failure occurred will be leaked. Consider, for example, a function that reads from a file into a buffer.The function might follow these steps: 1. Allocate the buffer. 2. Open the file. 3. Read from the file into the buffer. 4. Close the file. 5. Return the buffer. If the file doesn’t exist, Step 2 will fail. An appropriate course of action might be to return NULL from the function. However, if the buffer has already been allocated in Step 1, there is a risk of leaking that memory.You must remember to deallocate the buffer somewhere along any flow of control from which you don’t return. If Step 3 fails, not only must you deallocate the buffer before returning, but you also must close the file. Listing 2.6 shows an example of how you might write this function.
Listing 2.6 (readfile.c) Freeing Resources During Abnormal Conditions
#include #include #include #include #include

continues

36

Chapter 2

Writing Good GNU/Linux Software

Listing 2.6 Continued char* read_from_file (const char* filename, size_t length) { char* buffer; int fd; ssize_t bytes_read; /* Allocate the buffer. */ buffer = (char*) malloc (length); if (buffer == NULL) return NULL; /* Open the file. */ fd = open (filename, O_RDONLY); if (fd == -1) { /* open failed. Deallocate buffer before returning. */ free (buffer); return NULL; } /* Read the data. */ bytes_read = read (fd, buffer, length); if (bytes_read != length) { /* read failed. Deallocate buffer and close fd before returning. free (buffer); close (fd); return NULL; } /* Everything’s fine. Close the file and return the buffer. */ close (fd); return buffer; }

*/

Linux cleans up allocated memory, open files, and most other resources when a program terminates, so it’s not necessary to deallocate buffers and close files before calling exit.You might need to manually free other shared resources, however, such as temporary files and shared memory, which can potentially outlive a program.

2.3 Writing and Using Libraries
Virtually all programs are linked against one or more libraries. Any program that uses a C function (such as printf or malloc) will be linked against the C runtime library. If your program has a graphical user interface (GUI), it will be linked against windowing libraries. If your program uses a database, the database provider will give you libraries that you can use to access the database conveniently. In each of these cases, you must decide whether to link the library statically or dynamically. If you choose to link statically, your programs will be bigger and harder to upgrade, but probably easier to deploy. If you link dynamically, your programs will be

2.3

Writing and Using Libraries

37

smaller, easier to upgrade, but harder to deploy.This section explains how to link both statically and dynamically, examines the trade-offs in more detail, and gives some “rules of thumb” for deciding which kind of linking is better for you.

2.3.1

Archives

An archive (or static library) is simply a collection of object files stored as a single file. (An archive is roughly the equivalent of a Windows .LIB file.) When you provide an archive to the linker, the linker searches the archive for the object files it needs, extracts them, and links them into your program much as if you had provided those object files directly. You can create an archive using the ar command. Archive files traditionally use a .a extension rather than the .o extension used by ordinary object files. Here’s how you would combine test1.o and test2.o into a single libtest.a archive:
% ar cr libtest.a test1.o test2.o

The cr flags tell ar to create the archive.3 Now you can link with this archive using the -ltest option with gcc or g++, as described in Section 1.2.2, “Linking Object Files,” in Chapter 1, “Getting Started.” When the linker encounters an archive on the command line, it searches the archive for all definitions of symbols (functions or variables) that are referenced from the object files that it has already processed but not yet defined.The object files that define those symbols are extracted from the archive and included in the final executable. Because the linker searches the archive when it is encountered on the command line, it usually makes sense to put archives at the end of the command line. For example, suppose that test.c contains the code in Listing 2.7 and app.c contains the code in Listing 2.8.
Listing 2.7 (test.c) Library Contents int f () { return 3; }

Listing 2.8 (app.c) A Program That Uses Library Functions int main () { return f (); }

3.You can use other flags to remove a file from an archive or to perform other operations on the archive.These operations are rarely used but are documented on the ar man page.

38

Chapter 2

Writing Good GNU/Linux Software

Now suppose that test.o is combined with some other object files to produce the libtest.a archive.The following command line will not work:
% gcc -o app -L. -ltest app.o app.o: In function ‘main’: app.o(.text+0x4): undefined reference to ‘f’ collect2: ld returned 1 exit status

The error message indicates that even though libtest.a contains a definition of f, the linker did not find it.That’s because libtest.a was searched when it was first encountered, and at that point the linker hadn’t seen any references to f. On the other hand, if we use this line, no error messages are issued:
% gcc -o app app.o -L. –ltest

The reason is that the reference to f in object file from the libtest.a archive.

app.o

causes the linker to include the

test.o

2.3.2 Shared Libraries
A shared library (also known as a shared object, or as a dynamically linked library) is similar to a archive in that it is a grouping of object files. However, there are many important differences.The most fundamental difference is that when a shared library is linked into a program, the final executable does not actually contain the code that is present in the shared library. Instead, the executable merely contains a reference to the shared library. If several programs on the system are linked against the same shared library, they will all reference the library, but none will actually be included.Thus, the library is “shared” among all the programs that link with it. A second important difference is that a shared library is not merely a collection of object files, out of which the linker chooses those that are needed to satisfy undefined references. Instead, the object files that compose the shared library are combined into a single object file so that a program that links against a shared library always includes all of the code in the library, rather than just those portions that are needed. To create a shared library, you must compile the objects that will make up the library using the -fPIC option to the compiler, like this:
% gcc -c -fPIC test1.c

The -fPIC option tells the compiler that you are going to be using shared object.
Position-Independent Code (PIC)

test.o

as part of a

PIC stands for position-independent code. The functions in a shared library may be loaded at different addresses in different programs, so the code in the shared object must not depend on the address (or position) at which it is loaded. This consideration has no impact on you, as the programmer, except that you must remember to use the -fPIC flag when compiling code that will be used in a shared library.

2.3

Writing and Using Libraries

39

Then you combine the object files into a shared library, like this:
% gcc -shared -fPIC -o libtest.so test1.o test2.o

The -shared option tells the linker to produce a shared library rather than an ordinary executable. Shared libraries use the extension .so, which stands for shared object. Like static archives, the name always begins with lib to indicate that the file is a library. Linking with a shared library is just like linking with a static archive. For example, the following line will link with libtest.so if it is in the current directory, or one of the standard library search directories on the system:
% gcc -o app app.o -L. –ltest

Suppose that both libtest.a and libtest.so are available.Then the linker must choose one of the libraries and not the other.The linker searches each directory (first those specified with -L options, and then those in the standard directories).When the linker finds a directory that contains either libtest.a or libtest.so, the linker stops search directories. If only one of the two variants is present in the directory, the linker chooses that variant. Otherwise, the linker chooses the shared library version, unless you explicitly instruct it otherwise.You can use the -static option to demand static archives. For example, the following line will use the libtest.a archive, even if the libtest.so shared library is also available:
% gcc -static -o app app.o -L. –ltest

The ldd command displays the shared libraries that are linked into an executable. These libraries need to be available when the executable is run. Note that ldd will list an additional library called ld-linux.so, which is a part of GNU/Linux’s dynamic linking mechanism. Using LD_LIBRARY_PATH When you link a program with a shared library, the linker does not put the full path to the shared library in the resulting executable. Instead, it places only the name of the shared library.When the program is actually run, the system searches for the shared library and loads it.The system searches only /lib and /usr/lib, by default. If a shared library that is linked into your program is installed outside those directories, it will not be found, and the system will refuse to run the program. One solution to this problem is to use the -Wl,-rpath option when linking the program. Suppose that you use this:
% gcc -o app app.o -L. -ltest -Wl,-rpath,/usr/local/lib

Then, when libraries.

app

is run, the system will search

/usr/local/lib

for any required shared

40

Chapter 2

Writing Good GNU/Linux Software

Another solution to this problem is to set the LD_LIBRARY_PATH environment variable when running the program. Like the PATH environment variable, LD_LIBRARY_PATH is a colon-separated list of directories. For example, if LD_LIBRARY_PATH is /usr/local/lib:/opt/lib, then /usr/local/lib and /opt/lib will be searched before the standard /lib and /usr/lib directories.You should also note that if you have LD_LIBRARY_PATH, the linker will search the directories given there in addition to the directories given with the -L option when it is building an executable.4

2.3.3 Standard Libraries
Even if you didn’t specify any libraries when you linked your program, it almost certainly uses a shared library.That’s because GCC automatically links in the standard C library, libc, for you.The standard C library math functions are not included in libc; instead, they’re in a separate library, libm, which you need to specify explicitly. For example, to compile and link a program compute.c which uses trigonometric functions such as sin and cos, you must invoke this code:
% gcc -o compute compute.c –lm

If you write a C++ program and link it using the c++ or get the standard C++ library, libstdc++, automatically.

g++

commands, you’ll also

2.3.4 Library Dependencies
One library will often depend on another library. For example, many GNU/Linux systems include libtiff, a library that contains functions for reading and writing image files in the TIFF format.This library, in turn, uses the libraries libjpeg (JPEG image routines) and libz (compression routines). Listing 2.9 shows a very small program that uses libtiff to open a TIFF image file.
Listing 2.9 (tifftest.c) Using libtiff
#include #include int main (int argc, char** argv) { TIFF* tiff; tiff = TIFFOpen (argv[1], “r”); TIFFClose (tiff); return 0; }

4.You might see a reference to LD_RUN_PATH in some online documentation. Don’t believe what you read; this variable does not actually do anything under GNU/Linux.

2.3

Writing and Using Libraries

41

Save this source file as tifftest.c.To compile this program and link with specify -ltiff on your link line:
% gcc -o tifftest tifftest.c –ltiff

libtiff,

By default, this will pick up the shared-library version of libtiff, found at /usr/lib/libtiff.so. Because libtiff uses libjpeg and libz, the shared-library versions of these two are also drawn in (a shared library can point to other shared libraries that it depends on).To verify this, use the ldd command:
% ldd tifftest libtiff.so.3 => /usr/lib/libtiff.so.3 (0x4001d000) libc.so.6 => /lib/libc.so.6 (0x40060000) libjpeg.so.62 => /usr/lib/libjpeg.so.62 (0x40155000) libz.so.1 => /usr/lib/libz.so.1 (0x40174000) /lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

Static libraries, on the other hand, cannot point to other libraries. If decide to link with the static version of libtiff by specifying -static on your command line, you will encounter unresolved symbols:
% gcc -static -o tifftest tifftest.c -ltiff /usr/bin/../lib/libtiff.a(tif_jpeg.o): In function ‘TIFFjpeg_error_exit’: tif_jpeg.o(.text+0x2a): undefined reference to ‘jpeg_abort’ /usr/bin/../lib/libtiff.a(tif_jpeg.o): In function ‘TIFFjpeg_create_compress’: tif_jpeg.o(.text+0x8d): undefined reference to ‘jpeg_std_error’ tif_jpeg.o(.text+0xcf): undefined reference to ‘jpeg_CreateCompress’ ...

To link this program statically, you must specify the other two libraries yourself:
% gcc -static -o tifftest tifftest.c -ltiff -ljpeg -lz

Occasionally, two libraries will be mutually dependent. In other words, the first archive will reference symbols defined in the second archive, and vice versa.This situation generally arises out of poor design, but it does occasionally arise. In this case, you can provide a single library multiple times on the command line.The linker will research the library each time it occurs. For example, this line will cause libfoo.a to be searched multiple times:
% gcc -o app app.o -lfoo -lbar –lfoo

So, even if libfoo.a references symbols in link successfully.

libbar.a, and

vice versa, the program will

2.3.5 Pros and Cons
Now that you know all about static archives and shared libraries, you’re probably wondering which to use.There are a few major considerations to keep in mind.

42

Chapter 2

Writing Good GNU/Linux Software

One major advantage of a shared library is that it saves space on the system where the program is installed. If you are installing 10 programs, and they all make use of the same shared library, then you save a lot of space by using a shared library. If you used a static archive instead, the archive is included in all 10 programs. So, using shared libraries saves disk space. It also reduces download times if your program is being downloaded from the Web. A related advantage to shared libraries is that users can upgrade the libraries without upgrading all the programs that depend on them. For example, suppose that you produce a shared library that manages HTTP connections. Many programs might depend on this library. If you find a bug in this library, you can upgrade the library. Instantly, all the programs that depend on the library will be fixed; you don’t have to relink all the programs the way you do with a static archive. Those advantages might make you think that you should always use shared libraries. However, substantial reasons exist to use static archives instead.The fact that an upgrade to a shared library affects all programs that depend on it can be a disadvantage. For example, if you’re developing mission-critical software, you might rather link to a static archive so that an upgrade to shared libraries on the system won’t affect your program. (Otherwise, users might upgrade the shared library, thereby breaking your program, and then call your customer support line, blaming you!) If you’re not going to be able to install your libraries in /lib or /usr/lib, you should definitely think twice about using a shared library. (You won’t be able to install your libraries in those directories if you expect users to install your software without administrator privileges.) In particular, the -Wl,-rpath trick won’t work if you don’t know where the libraries are going to end up. And asking your users to set LD_LIBRARY_PATH means an extra step for them. Because each user has to do this individually, this is a substantial additional burden. You’ll have to weigh these advantages and disadvantages for every program you distribute.

2.3.6 Dynamic Loading and Unloading
Sometimes you might want to load some code at run time without explicitly linking in that code. For example, consider an application that supports “plug-in” modules, such as a Web browser.The browser allows third-party developers to create plug-ins to provide additional functionality.The third-party developers create shared libraries and place them in a known location.The Web browser then automatically loads the code in these libraries. This functionality is available under Linux by using the dlopen function. You could open a shared library named libtest.so by calling dlopen like this: dlopen (“libtest.so”, RTLD_LAZY)

2.3

Writing and Using Libraries

43

(The second parameter is a flag that indicates how to bind symbols in the shared library.You can consult the online man pages for dlopen if you want more information, but RTLD_LAZY is usually the setting that you want.) To use dynamic loading functions, include the header file and link with the –ldl option to pick up the libdl library. The return value from this function is a void * that is used as a handle for the shared library.You can pass this value to the dlsym function to obtain the address of a function that has been loaded with the shared library. For example, if libtest.so defines a function named my_function, you could call it like this: void* handle = dlopen (“libtest.so”, RTLD_LAZY); void (*test)() = dlsym (handle, “my_function”); (*test)(); dlclose (handle);

The dlsym system call can also be used to obtain a pointer to a static variable in the shared library. Both dlopen and dlsym return NULL if they do not succeed. In that event, you can call dlerror (with no parameters) to obtain a human-readable error message describing the problem. The dlclose function unloads the shared library.Technically, dlopen actually loads the library only if it is not already loaded. If the library has already been loaded, dlopen simply increments the library reference count. Similarly, dlclose decrements the reference count and then unloads the library only if the reference count has reached zero. If you’re writing the code in your shared library in C++, you will probably want to declare those functions and variables that you plan to access elsewhere with the extern “C” linkage specifier. For instance, if the C++ function my_function is in a shared library and you want to access it with dlsym, you should declare it like this: extern “C” void foo ();

This prevents the C++ compiler from mangling the function name, which would change the function’s name from foo to a different, funny-looking name that encodes extra information about the function. A C compiler will not mangle names; it will use whichever name you give to your function or variable.

3
Processes

RUNNING INSTANCE OF A PROGRAM IS CALLED A PROCESS. If you have two terminal windows showing on your screen, then you are probably running the same terminal program twice—you have two terminal processes. Each terminal window is probably running a shell; each running shell is another process.When you invoke a command from a shell, the corresponding program is executed in a new process; the shell process resumes when that process completes. Advanced programmers often use multiple cooperating processes in a single application to enable the application to do more than one thing at once, to increase application robustness, and to make use of already-existing programs. Most of the process manipulation functions described in this chapter are similar to those on other UNIX systems. Most are declared in the header file ; check the man page for each function to be sure.

A

3.1 Looking at Processes
Even as you sit down at your computer, there are processes running. Every executing program uses one or more processes. Let’s start by taking a look at the processes already on your computer.

46

Chapter 3

Processes

3.1.1

Process IDs

Each process in a Linux system is identified by its unique process ID, sometimes referred to as pid. Process IDs are 16-bit numbers that are assigned sequentially by Linux as new processes are created. Every process also has a parent process (except the special init process, described in Section 3.4.3, “Zombie Processes”).Thus, you can think of the processes on a Linux system as arranged in a tree, with the init process at its root.The parent process ID, or ppid, is simply the process ID of the process’s parent. When referring to process IDs in a C or C++ program, always use the pid_t typedef, which is defined in . A program can obtain the process ID of the process it’s running in with the getpid() system call, and it can obtain the process ID of its parent process with the getppid() system call. For instance, the program in Listing 3.1 prints its process ID and its parent’s process ID.
Listing 3.1 ( print-pid.c) Printing the Process ID
#include #include int main { printf printf return } () (“The process ID is %d\n”, (int) getpid ()); (“The parent process ID is %d\n”, (int) getppid ()); 0;

Observe that if you invoke this program several times, a different process ID is reported because each invocation is in a new process. However, if you invoke it every time from the same shell, the parent process ID (that is, the process ID of the shell process) is the same.

3.1.2 Viewing Active Processes
The ps command displays the processes that are running on your system.The GNU/Linux version of ps has lots of options because it tries to be compatible with versions of ps on several other UNIX variants.These options control which processes are listed and what information about each is shown. By default, invoking ps displays the processes controlled by the terminal or terminal window in which ps is invoked. For example:
% ps PID TTY 21693 pts/8 21694 pts/8 TIME CMD 00:00:00 bash 00:00:00 ps

3.1

Looking at Processes

47

This invocation of ps shows two processes.The first, bash, is the shell running on this terminal.The second is the running instance of the ps program itself.The first column, labeled PID, displays the process ID of each. For a more detailed look at what’s running on your GNU/Linux system, invoke this:
% ps -e -o pid,ppid,command

The

-o pid,ppid,command

option instructs ps to display all processes running on the system.The option tells ps what information to show about each process— in this case, the process ID, the parent process ID, and the command running in this process.
-e

ps Output Formats With the -o option to the ps command, you specify the information about processes that you want in the output as a comma-separated list. For example, ps -o pid,user,start_time,command displays the process ID, the name of the user owning the process, the wall clock time at which the process started, and the command running in the process. See the man page for ps for the full list of field codes. You can use the -f (full listing), -l (long listing), or -j (jobs listing) options instead to get three different preset listing formats.

Here are the first few lines and last few lines of output from this command on my system.You may see different output, depending on what’s running on your system.
% ps -e -o pid,ppid,command PID PPID COMMAND 1 0 init [5] 2 1 [kflushd] 3 1 [kupdate] ... 21725 21693 xterm 21727 21725 bash 21728 21727 ps -e -o pid,ppid,command

Note that the parent process ID of the ps command, 21727, is the process ID of bash, the shell from which I invoked ps.The parent process ID of bash is in turn 21725, the process ID of the xterm program in which the shell is running.

3.1.3 Killing a Process
You can kill a running process with the kill command. Simply specify on the command line the process ID of the process to be killed. The kill command works by sending the process a SIGTERM, or termination, signal.1 This causes the process to terminate, unless the executing program explicitly handles or masks the SIGTERM signal. Signals are described in Section 3.3, “Signals.”
1.You can also use the kill command to send other signals to a process.This is described in Section 3.4, “Process Termination.”

48

Chapter 3

Processes

3.2 Creating Processes
Two common techniques are used for creating a new process.The first is relatively simple but should be used sparingly because it is inefficient and has considerably security risks.The second technique is more complex but provides greater flexibility, speed, and security.

3.2.1 Using system
The system function in the standard C library provides an easy way to execute a command from within a program, much as if the command had been typed into a shell. In fact, system creates a subprocess running the standard Bourne shell (/bin/sh) and hands the command to that shell for execution. For example, this program in Listing 3.2 invokes the ls command to display the contents of the root directory, as if you typed ls -l / into a shell.
Listing 3.2 (system.c) Using the system Call
#include int main () { int return_value; return_value = system (“ls -l /”); return return_value; }

The system function returns the exit status of the shell command. If the shell itself cannot be run, system returns 127; if another error occurs, system returns –1. Because the system function uses a shell to invoke your command, it’s subject to the features, limitations, and security flaws of the system’s shell.You can’t rely on the availability of any particular version of the Bourne shell. On many UNIX systems, /bin/sh is a symbolic link to another shell. For instance, on most GNU/Linux systems, /bin/sh points to bash (the Bourne-Again SHell), and different GNU/Linux distributions use different versions of bash. Invoking a program with root privilege with the system function, for instance, can have different results on different GNU/Linux systems.Therefore, it’s preferable to use the fork and exec method for creating processes.

3.2.2 Using fork and exec
The DOS and Windows API contains the spawn family of functions.These functions take as an argument the name of a program to run and create a new process instance of that program. Linux doesn’t contain a single function that does all this in one step. Instead, Linux provides one function, fork, that makes a child process that is an exact

3.2

Creating Processes

49

copy of its parent process. Linux provides another set of functions, the exec family, that causes a particular process to cease being an instance of one program and to instead become an instance of another program.To spawn a new process, you first use fork to make a copy of the current process.Then you use exec to transform one of these processes into an instance of the program you want to spawn. Calling fork When a program calls fork, a duplicate process, called the child process, is created.The parent process continues executing the program from the point that fork was called. The child process, too, executes the same program from the same place. So how do the two processes differ? First, the child process is a new process and therefore has a new process ID, distinct from its parent’s process ID. One way for a program to distinguish whether it’s in the parent process or the child process is to call getpid. However, the fork function provides different return values to the parent and child processes—one process “goes in” to the fork call, and two processes “come out,” with different return values.The return value in the parent process is the process ID of the child.The return value in the child process is zero. Because no process ever has a process ID of zero, this makes it easy for the program whether it is now running as the parent or the child process. Listing 3.3 is an example of using fork to duplicate a program’s process. Note that the first block of the if statement is executed only in the parent process, while the else clause is executed in the child process.
Listing 3.3 ( fork.c) Using fork to Duplicate a Program’s Process
#include #include #include int main () { pid_t child_pid; printf (“the main program process ID is %d\n”, (int) getpid ()); child_pid = fork (); if (child_pid != 0) { printf (“this is the parent process, with id %d\n”, (int) getpid ()); printf (“the child’s process ID is %d\n”, (int) child_pid); } else printf (“this is the child process, with id %d\n”, (int) getpid ()); return 0; }

50

Chapter 3

Processes

Using the exec Family The exec functions replace the program running in a process with another program. When a program calls an exec function, that process immediately ceases executing that program and begins executing a new program from the beginning, assuming that the exec call doesn’t encounter an error. Within the exec family, there are functions that vary slightly in their capabilities and how they are called. Functions that contain the letter p in their names (execvp and execlp) accept a program name and search for a program by that name in the current execution path; functions that don’t contain the p must be given the full path of the program to be executed. n n

n

Functions that contain the letter v in their names (execv, execvp, and execve) accept the argument list for the new program as a NULL-terminated array of pointers to strings. Functions that contain the letter l (execl, execlp, and execle) accept the argument list using the C language’s varargs mechanism. Functions that contain the letter e in their names (execve and execle) accept an additional argument, an array of environment variables.The argument should be a NULL-terminated array of pointers to character strings. Each character string should be of the form “VARIABLE=value”.

Because exec replaces the calling program with another one, it never returns unless an error occurs. The argument list passed to the program is analogous to the command-line arguments that you specify to a program when you run it from the shell.They are available through the argc and argv parameters to main. Remember, when a program is invoked from the shell, the shell sets the first element of the argument list argv[0]) to the name of the program, the second element of the argument list (argv[1]) to the first command-line argument, and so on.When you use an exec function in your programs, you, too, should pass the name of the function as the first element of the argument list. Using fork and exec Together A common pattern to run a subprogram within a program is first to fork the process and then exec the subprogram.This allows the calling program to continue execution in the parent process while the calling program is replaced by the subprogram in the child process. The program in Listing 3.4, like Listing 3.2, lists the contents of the root directory using the ls command. Unlike the previous example, though, it invokes the ls command directly, passing it the command-line arguments -l and / rather than invoking it through a shell.

3.2

Creating Processes

51

Listing 3.4 ( fork-exec.c) Using fork and exec Together
#include #include #include #include

/* Spawn a child process running a new program. PROGRAM is the name of the program to run; the path will be searched for this program. ARG_LIST is a NULL-terminated list of character strings to be passed as the program’s argument list. Returns the process ID of the spawned process. */ int spawn (char* program, char** arg_list) { pid_t child_pid; /* Duplicate this process. */ child_pid = fork (); if (child_pid != 0) /* This is the parent process. */ return child_pid; else { /* Now execute PROGRAM, searching for it in the path. */ execvp (program, arg_list); /* The execvp function returns only if an error occurs. */ fprintf (stderr, “an error occurred in execvp\n”); abort (); } } int main () { /* The argument list to pass to the “ls” command. */ char* arg_list[] = { “ls”, /* argv[0], the name of the program. */ “-l”, “/”, NULL /* The argument list must end with a NULL. }; /* Spawn a child process running the “ls” command. returned child process ID. */ spawn (“ls”, arg_list); printf (“done with main program\n”); return 0; }

*/

Ignore the

52

Chapter 3

Processes

3.2.3 Process Scheduling
Linux schedules the parent and child processes independently; there’s no guarantee of which one will run first, or how long it will run before Linux interrupts it and lets the other process (or some other process on the system) run. In particular, none, part, or all of the ls command may run in the child process before the parent completes.2 Linux promises that each process will run eventually—no process will be completely starved of execution resources. You may specify that a process is less important—and should be given a lower priority —by assigning it a higher niceness value. By default, every process has a niceness of zero. A higher niceness value means that the process is given a lesser execution priority; conversely, a process with a lower (that is, negative) niceness gets more execution time. To run a program with a nonzero niceness, use the nice command, specifying the niceness value with the -n option. For example, this is how you might invoke the command “sort input.txt > output.txt”, a long sorting operation, with a reduced priority so that it doesn’t slow down the system too much:
% nice -n 10 sort input.txt > output.txt

You can use the renice command to change the niceness of a running process from the command line. To change the niceness of a running process programmatically, use the nice function. Its argument is an increment value, which is added to the niceness value of the process that calls it. Remember that a positive value raises the niceness value and thus reduces the process’s execution priority. Note that only a process with root privilege can run a process with a negative niceness value or reduce the niceness value of a running process.This means that you may specify negative values to the nice and renice commands only when logged in as root, and only a process running as root can pass a negative value to the nice function. This prevents ordinary users from grabbing execution priority away from others using the system.

3.3 Signals
Signals are mechanisms for communicating with and manipulating processes in Linux. The topic of signals is a large one; here we discuss some of the most important signals and techniques that are used for controlling processes. A signal is a special message sent to a process. Signals are asynchronous; when a process receives a signal, it processes the signal immediately, without finishing the current function or even the current line of code.There are several dozen different signals, each with a different meaning. Each signal type is specified by its signal number, but in programs, you usually refer to a signal by its name. In Linux, these are defined in /usr/include/bits/signum.h. (You shouldn’t include this header file directly in your programs; instead, use .)
2. A method for serializing the two processes is presented in Section 3.4.1, “Waiting for Process Termination.”

3.3

Signals

53

When a process receives a signal, it may do one of several things, depending on the signal’s disposition. For each signal, there is a default disposition, which determines what happens to the process if the program does not specify some other behavior. For most signal types, a program may specify some other behavior—either to ignore the signal or to call a special signal-handler function to respond to the signal. If a signal handler is used, the currently executing program is paused, the signal handler is executed, and, when the signal handler returns, the program resumes. The Linux system sends signals to processes in response to specific conditions. For instance, SIGBUS (bus error), SIGSEGV (segmentation violation), and SIGFPE (floating point exception) may be sent to a process that attempts to perform an illegal operation.The default disposition for these signals it to terminate the process and produce a core file. A process may also send a signal to another process. One common use of this mechanism is to end another process by sending it a SIGTERM or SIGKILL signal.3 Another common use is to send a command to a running program.Two “userdefined” signals are reserved for this purpose: SIGUSR1 and SIGUSR2.The SIGHUP signal is sometimes used for this purpose as well, commonly to wake up an idling program or cause a program to reread its configuration files. The sigaction function can be used to set a signal disposition.The first parameter is the signal number.The next two parameters are pointers to sigaction structures; the first of these contains the desired disposition for that signal number, while the second receives the previous disposition.The most important field in the first or second sigaction structure is sa_handler. It can take one of three values: SIG_DFL, which specifies the default disposition for the signal. SIG_IGN, which specifies that the signal should be ignored. A pointer to a signal-handler function.The function should take one parameter, the signal number, and return void. n n n

Because signals are asynchronous, the main program may be in a very fragile state when a signal is processed and thus while a signal handler function executes. Therefore, you should avoid performing any I/O operations or calling most library and system functions from signal handlers. A signal handler should perform the minimum work necessary to respond to the signal, and then return control to the main program (or terminate the program). In most cases, this consists simply of recording the fact that a signal occurred.The main program then checks periodically whether a signal has occurred and reacts accordingly. It is possible for a signal handler to be interrupted by the delivery of another signal. While this may sound like a rare occurrence, if it does occur, it will be very difficult to diagnose and debug the problem. (This is an example of a race condition, discussed in Chapter 4, “Threads,” Section 4.4, “Synchronization and Critical Sections.”) Therefore, you should be very careful about what your program does in a signal handler.
3.What’s the difference? The SIGTERM signal asks a process to terminate; the process may ignore the request by masking or ignoring the signal.The SIGKILL signal always kills the process immediately because the process may not mask or ignore SIGKILL.

54

Chapter 3

Processes

Even assigning a value to a global variable can be dangerous because the assignment may actually be carried out in two or more machine instructions, and a second signal may occur between them, leaving the variable in a corrupted state. If you use a global variable to flag a signal from a signal-handler function, it should be of the special type sig_atomic_t. Linux guarantees that assignments to variables of this type are performed in a single instruction and therefore cannot be interrupted midway. In Linux, sig_atomic_t is an ordinary int; in fact, assignments to integer types the size of int or smaller, or to pointers, are atomic. If you want to write a program that’s portable to any standard UNIX system, though, use sig_atomic_t for these global variables. This program skeleton in Listing 3.5, for instance, uses a signal-handler function to count the number of times that the program receives SIGUSR1, one of the signals reserved for application use.
Listing 3.5 (sigusr1.c) Using a Signal Handler
#include #include #include #include #include

sig_atomic_t sigusr1_count = 0; void handler (int signal_number) { ++sigusr1_count; } int main () { struct sigaction sa; memset (&sa, 0, sizeof (sa)); sa.sa_handler = &handler; sigaction (SIGUSR1, &sa, NULL); /* Do some lengthy stuff here. /* ... */ */

printf (“SIGUSR1 was raised %d times\n”, sigusr1_count); return 0; }

3.4

Process Termination

55

3.4 Process Termination
Normally, a process terminates in one of two ways. Either the executing program calls the exit function, or the program’s main function returns. Each process has an exit code: a number that the process returns to its parent.The exit code is the argument passed to the exit function, or the value returned from main. A process may also terminate abnormally, in response to a signal. For instance, the SIGBUS, SIGSEGV, and SIGFPE signals mentioned previously cause the process to terminate. Other signals are used to terminate a process explicitly.The SIGINT signal is sent to a process when the user attempts to end it by typing Ctrl+C in its terminal.The SIGTERM signal is sent by the kill command.The default disposition for both of these is to terminate the process. By calling the abort function, a process sends itself the SIGABRT signal, which terminates the process and produces a core file.The most powerful termination signal is SIGKILL, which ends a process immediately and cannot be blocked or handled by a program. Any of these signals can be sent using the kill command by specifying an extra command-line flag; for instance, to end a troublesome process by sending it a SIGKILL, invoke the following, where pid is its process ID:
% kill -KILL pid

To send a signal from a program, use the kill function.The first parameter is the target process ID.The second parameter is the signal number; use SIGTERM to simulate the default behavior of the kill command. For instance, where child pid contains the process ID of the child process, you can use the kill function to terminate a child process from the parent by calling it like this: kill (child_pid, SIGTERM);

Include the and headers if you use the kill function. By convention, the exit code is used to indicate whether the program executed correctly. An exit code of zero indicates correct execution, while a nonzero exit code indicates that an error occurred. In the latter case, the particular value returned may give some indication of the nature of the error. It’s a good idea to stick with this convention in your programs because other components of the GNU/Linux system assume this behavior. For instance, shells assume this convention when you connect multiple programs with the && (logical and) and || (logical or) operators.Therefore, you should explicitly return zero from your main function, unless an error occurs.

56

Chapter 3

Processes

With most shells, it’s possible to obtain the exit code of the most recently executed program using the special $? variable. Here’s an example in which the ls command is invoked twice and its exit code is displayed after each invocation. In the first case, ls executes correctly and returns the exit code zero. In the second case, ls encounters an error (because the filename specified on the command line does not exist) and thus returns a nonzero exit code.
% ls / bin coda etc lib misc nfs proc boot dev home lost+found mnt opt root % echo $? 0 % ls bogusfile ls: bogusfile: No such file or directory % echo $? 1 sbin tmp usr var

Note that even though the parameter type of the exit function is int and the main function returns an int, Linux does not preserve the full 32 bits of the return code. In fact, you should use exit codes only between zero and 127. Exit codes above 128 have a special meaning—when a process is terminated by a signal, its exit code is 128 plus the signal number.

3.4.1 Waiting for Process Termination
If you typed in and ran the fork and exec example in Listing 3.4, you may have noticed that the output from the ls program often appears after the “main program” has already completed.That’s because the child process, in which ls is run, is scheduled independently of the parent process. Because Linux is a multitasking operating system, both processes appear to execute simultaneously, and you can’t predict whether the ls program will have a chance to run before or after the parent process runs. In some situations, though, it is desirable for the parent process to wait until one or more child processes have completed.This can be done with the wait family of system calls.These functions allow you to wait for a process to finish executing, and enable the parent process to retrieve information about its child’s termination.There are four different system calls in the wait family; you can choose to get a little or a lot of information about the process that exited, and you can choose whether you care about which child process terminated.

3.4.2 The wait System Calls
The simplest such function is called simply wait. It blocks the calling process until one of its child processes exits (or an error occurs). It returns a status code via an integer pointer argument, from which you can extract information about how the child process exited. For instance, the WEXITSTATUS macro extracts the child process’s exit code.

3.4

Process Termination

57

You can use the WIFEXITED macro to determine from a child process’s exit status whether that process exited normally (via the exit function or returning from main) or died from an unhandled signal. In the latter case, use the WTERMSIG macro to extract from its exit status the signal number by which it died. Here is the main function from the fork and exec example again.This time, the parent process calls wait to wait until the child process, in which the ls command executes, is finished. int main () { int child_status; /* The argument list to pass to the “ls” command. */ char* arg_list[] = { “ls”, /* argv[0], the name of the program. */ “-l”, “/”, NULL /* The argument list must end with a NULL. }; /* Spawn a child process running the “ls” command. returned child process ID. */ spawn (“ls”, arg_list);

*/

Ignore the

/* Wait for the child process to complete. */ wait (&child_status); if (WIFEXITED (child_status)) printf (“the child process exited normally, with exit code %d\n”, WEXITSTATUS (child_status)); else printf (“the child process exited abnormally\n”); return 0; }

Several similar system calls are available in Linux, which are more flexible or provide more information about the exiting child process.The waitpid function can be used to wait for a specific child process to exit instead of any child process.The wait3 function returns CPU usage statistics about the exiting child process, and the wait4 function allows you to specify additional options about which processes to wait for.

3.4.3 Zombie Processes
If a child process terminates while its parent is calling a wait function, the child process vanishes and its termination status is passed to its parent via the wait call. But what happens when a child process terminates and the parent is not calling wait? Does it simply vanish? No, because then information about its termination—such as whether it exited normally and, if so, what its exit status is—would be lost. Instead, when a child process terminates, is becomes a zombie process.

58

Chapter 3

Processes

A zombie process is a process that has terminated but has not been cleaned up yet. It is the responsibility of the parent process to clean up its zombie children.The wait functions do this, too, so it’s not necessary to track whether your child process is still executing before waiting for it. Suppose, for instance, that a program forks a child process, performs some other computations, and then calls wait. If the child process has not terminated at that point, the parent process will block in the wait call until the child process finishes. If the child process finishes before the parent process calls wait, the child process becomes a zombie.When the parent process calls wait, the zombie child’s termination status is extracted, the child process is deleted, and the wait call returns immediately. What happens if the parent does not clean up its children? They stay around in the system, as zombie processes.The program in Listing 3.6 forks a child process, which terminates immediately and then goes to sleep for a minute, without ever cleaning up the child process.
Listing 3.6 (zombie.c) Making a Zombie Process
#include #include #include int main () { pid_t child_pid; /* Create a child process. */ child_pid = fork (); if (child_pid > 0) { /* This is the parent process. Sleep for a minute. */ sleep (60); } else { /* This is the child process. Exit immediately. */ exit (0); } return 0; }

Try compiling this file to an executable named make-zombie. Run it, and while it’s still running, list the processes on the system by invoking the following command in another window:
% ps -e -o pid,ppid,stat,cmd

3.4

Process Termination

59

This lists the process ID, parent process ID, process status, and process command line. Observe that, in addition to the parent make-zombie process, there is another make-zombie process listed. It’s the child process; note that its parent process ID is the process ID of the main make-zombie process.The child process is marked as , and its status code is Z, for zombie. What happens when the main make-zombie program ends when the parent process exits, without ever calling wait? Does the zombie process stay around? No—try running ps again, and note that both of the make-zombie processes are gone.When a program exits, its children are inherited by a special process, the init program, which always runs with process ID of 1 (it’s the first process started when Linux boots).The init process automatically cleans up any zombie child processes that it inherits.

3.4.4 Cleaning Up Children Asynchronously
If you’re using a child process simply to exec another program, it’s fine to call wait immediately in the parent process, which will block until the child process completes. But often, you’ll want the parent process to continue running, as one or more children execute synchronously. How can you be sure that you clean up child processes that have completed so that you don’t leave zombie processes, which consume system resources, lying around? One approach would be for the parent process to call wait3 or wait4 periodically, to clean up zombie children. Calling wait for this purpose doesn’t work well because, if no children have terminated, the call will block until one does. However, wait3 and wait4 take an additional flag parameter, to which you can pass the flag value WNOHANG. With this flag, the function runs in nonblocking mode—it will clean up a terminated child process if there is one, or simply return if there isn’t.The return value of the call is the process ID of the terminated child in the former case, or zero in the latter case. A more elegant solution is to notify the parent process when a child terminates. There are several ways to do this using the methods discussed in Chapter 5, “Interprocess Communication,” but fortunately Linux does this for you, using signals. When a child process terminates, Linux sends the parent process the SIGCHLD signal. The default disposition of this signal is to do nothing, which is why you might not have noticed it before. Thus, an easy way to clean up child processes is by handling SIGCHLD. Of course, when cleaning up the child process, it’s important to store its termination status if this information is needed, because once the process is cleaned up using wait, that information is no longer available. Listing 3.7 is what it looks like for a program to use a SIGCHLD handler to clean up its child processes.

60

Chapter 3

Processes

Listing 3.7 (sigchld.c) Cleaning Up Children by Handling SIGCHLD
#include #include #include #include

sig_atomic_t child_exit_status; void clean_up_child_process (int signal_number) { /* Clean up the child process. */ int status; wait (&status); /* Store its exit status in a global variable. child_exit_status = status; }

*/

int main () { /* Handle SIGCHLD by calling clean_up_child_process. */ struct sigaction sigchld_action; memset (&sigchld_action, 0, sizeof (sigchld_action)); sigchld_action.sa_handler = &clean_up_child_process; sigaction (SIGCHLD, &sigchld_action, NULL); /* Now do things, including forking a child process. /* ... */ return 0; } */

Note how the signal handler stores the child process’s exit status in a global variable, from which the main program can access it. Because the variable is assigned in a signal handler, its type is sig_atomic_t.

4
Threads

T

HREADS, LIKE PROCESSES, ARE A MECHANISM TO ALLOW A PROGRAM to do more than one thing at a time. As with processes, threads appear to run concurrently; the Linux kernel schedules them asynchronously, interrupting each thread from time to time to give others a chance to execute. Conceptually, a thread exists within a process.Threads are a finer-grained unit of execution than processes.When you invoke a program, Linux creates a new process and in that process creates a single thread, which runs the program sequentially.That thread can create additional threads; all these threads run the same program in the same process, but each thread may be executing a different part of the program at any given time. We’ve seen how a program can fork a child process.The child process is initially running its parent’s program, with its parent’s virtual memory, file descriptors, and so on copied.The child process can modify its memory, close file descriptors, and the like without affecting its parent, and vice versa.When a program creates another thread, though, nothing is copied.The creating and the created thread share the same memory space, file descriptors, and other system resources as the original. If one thread changes the value of a variable, for instance, the other thread subsequently will see the modified value. Similarly, if one thread closes a file descriptor, other threads may not read

62

Chapter 4

Threads

from or write to that file descriptor. Because a process and all its threads can be executing only one program at a time, if any thread inside a process calls one of the exec functions, all the other threads are ended (the new program may, of course, create new threads). GNU/Linux implements the POSIX standard thread API (known as pthreads). All thread functions and data types are declared in the header file .The pthread functions are not included in the standard C library. Instead, they are in libpthread, so you should add -lpthread to the command line when you link your program.

4.1 Thread Creation
Each thread in a process is identified by a thread ID.When referring to thread IDs in C or C++ programs, use the type pthread_t. Upon creation, each thread executes a thread function.This is just an ordinary function and contains the code that the thread should run.When the function returns, the thread exits. On GNU/Linux, thread functions take a single parameter, of type void*, and have a void* return type.The parameter is the thread argument: GNU/Linux passes the value along to the thread without looking at it.Your program can use this parameter to pass data to a new thread. Similarly, your program can use the return value to pass data from an exiting thread back to its creator. The pthread_create function creates a new thread.You provide it with the following: 1. A pointer to a pthread_t variable, in which the thread ID of the new thread is stored. 2. A pointer to a thread attribute object.This object controls details of how the thread interacts with the rest of the program. If you pass NULL as the thread attribute, a thread will be created with the default thread attributes.Thread attributes are discussed in Section 4.1.5, “Thread Attributes.” 3. A pointer to the thread function.This is an ordinary function pointer, of this type: void* (*) (void*)

4. A thread argument value of type void*. Whatever you pass is simply passed as the argument to the thread function when the thread begins executing. A call to pthread_create returns immediately, and the original thread continues executing the instructions following the call. Meanwhile, the new thread begins executing the thread function. Linux schedules both threads asynchronously, and your program must not rely on the relative order in which instructions are executed in the two threads.

4.1

Thread Creation

63

The program in Listing 4.1 creates a thread that prints x’s continuously to standard error. After calling pthread_create, the main thread prints o’s continuously to standard error.
Listing 4.1 (thread-create.c) Create a Thread
#include #include /* Prints x’s to stderr. The parameter is unused. Does not return. */

void* print_xs (void* unused) { while (1) fputc (‘x’, stderr); return NULL; } /* The main program. */

int main () { pthread_t thread_id; /* Create a new thread. The new thread will run the print_xs function. */ pthread_create (&thread_id, NULL, &print_xs, NULL); /* Print o’s continuously to stderr. */ while (1) fputc (‘o’, stderr); return 0; }

Compile and link this program using the following code:
% cc -o thread-create thread-create.c -lpthread

Try running it to see what happens. Notice the unpredictable pattern of x’s and o’s as Linux alternately schedules the two threads. Under normal circumstances, a thread exits in one of two ways. One way, as illustrated previously, is by returning from the thread function.The return value from the thread function is taken to be the return value of the thread. Alternately, a thread can exit explicitly by calling pthread_exit.This function may be called from within the thread function or from some other function called directly or indirectly by the thread function.The argument to pthread_exit is the thread’s return value.

64

Chapter 4

Threads

4.1.1 Passing Data to Threads
The thread argument provides a convenient method of passing data to threads. Because the type of the argument is void*, though, you can’t pass a lot of data directly via the argument. Instead, use the thread argument to pass a pointer to some structure or array of data. One commonly used technique is to define a structure for each thread function, which contains the “parameters” that the thread function expects. Using the thread argument, it’s easy to reuse the same thread function for many threads. All these threads execute the same code, but on different data. The program in Listing 4.2 is similar to the previous example.This one creates two new threads, one to print x’s and the other to print o’s. Instead of printing infinitely, though, each thread prints a fixed number of characters and then exits by returning from the thread function.The same thread function, char_print, is used by both threads, but each is configured differently using struct char_print_parms.
Listing 4.2 (thread-create2) Create Two Threads
#include #include /* Parameters to print_function. */

struct char_print_parms { /* The character to print. */ char character; /* The number of times to print it. int count; };

*/

/* Prints a number of characters to stderr, as given by PARAMETERS, which is a pointer to a struct char_print_parms. */ void* char_print (void* parameters) { /* Cast the cookie pointer to the right type. */ struct char_print_parms* p = (struct char_print_parms*) parameters; int i; for (i = 0; i < p->count; ++i) fputc (p->character, stderr); return NULL; } /* The main program. */

int main () { pthread_t thread1_id;

4.1

Thread Creation

65

pthread_t thread2_id; struct char_print_parms thread1_args; struct char_print_parms thread2_args; /* Create a new thread to print 30,000 ’x’s. */ thread1_args.character = ’x’; thread1_args.count = 30000; pthread_create (&thread1_id, NULL, &char_print, &thread1_args); /* Create a new thread to print 20,000 o’s. */ thread2_args.character = ’o’; thread2_args.count = 20000; pthread_create (&thread2_id, NULL, &char_print, &thread2_args); return 0; }

But wait! The program in Listing 4.2 has a serious bug in it.The main thread (which runs the main function) creates the thread parameter structures (thread1_args and thread2_args) as local variables, and then passes pointers to these structures to the threads it creates.What’s to prevent Linux from scheduling the three threads in such a way that main finishes executing before either of the other two threads are done? Nothing! But if this happens, the memory containing the thread parameter structures will be deallocated while the other two threads are still accessing it.

4.1.2 Joining Threads
One solution is to force main to wait until the other two threads are done.What we need is a function similar to wait that waits for a thread to finish instead of a process. That function is pthread_join, which takes two arguments: the thread ID of the thread to wait for, and a pointer to a void* variable that will receive the finished thread’s return value. If you don’t care about the thread return value, pass NULL as the second argument. Listing 4.3 shows the corrected main function for the buggy example in Listing 4.2. In this version, main does not exit until both of the threads printing x’s and o’s have completed, so they are no longer using the argument structures.
Listing 4.3 Revised Main Function for thread-create2.c int main () { pthread_t thread1_id; pthread_t thread2_id; struct char_print_parms thread1_args; struct char_print_parms thread2_args;

continues

66

Chapter 4

Threads

Listing 4.3 Continued
/* Create a new thread to print 30,000 x’s. */ thread1_args.character = ’x’; thread1_args.count = 30000; pthread_create (&thread1_id, NULL, &char_print, &thread1_args); /* Create a new thread to print 20,000 o’s. */ thread2_args.character = ’o’; thread2_args.count = 20000; pthread_create (&thread2_id, NULL, &char_print, &thread2_args); /* Make sure pthread_join /* Make sure pthread_join the first thread has finished. */ (thread1_id, NULL); the second thread has finished. */ (thread2_id, NULL); */

/* Now we can safely return. return 0; }

The moral of the story: Make sure that any data you pass to a thread by reference is not deallocated, even by a different thread, until you’re sure that the thread is done with it.This is true both for local variables, which are deallocated when they go out of scope, and for heap-allocated variables, which you deallocate by calling free (or using delete in C++).

4.1.3 Thread Return Values
If the second argument you pass to pthread_join is non-null, the thread’s return value will be placed in the location pointed to by that argument.The thread return value, like the thread argument, is of type void*. If you want to pass back a single int or other small number, you can do this easily by casting the value to void* and then casting back to the appropriate type after calling pthread_join.1 The program in Listing 4.4 computes the nth prime number in a separate thread. That thread returns the desired prime number as its thread return value.The main thread, meanwhile, is free to execute other code. Note that the successive division algorithm used in compute_prime is quite inefficient; consult a book on numerical algorithims if you need to compute many prime numbers in your programs.

1. Note that this is not portable, and it’s up to you to make sure that your value can be cast safely to void* and back without losing bits.

4.1

Thread Creation

67

Listing 4.4 ( primes.c) Compute Prime Numbers in a Thread
#include #include /* Compute successive prime numbers (very inefficiently). Return the Nth prime number, where N is the value pointed to by *ARG. */ void* compute_prime (void* arg) { int candidate = 2; int n = *((int*) arg); while (1) { int factor; int is_prime = 1; /* Test primality by successive division. */ for (factor = 2; factor < candidate; ++factor) if (candidate % factor == 0) { is_prime = 0; break; } /* Is this the prime number we’re looking for? */ if (is_prime) { if (--n == 0) /* Return the desired prime number as the thread return value. return (void*) candidate; } ++candidate; } return NULL; } int main () { pthread_t thread; int which_prime = 5000; int prime; /* Start the computing thread, up to the 5,000th prime number. */ pthread_create (&thread, NULL, &compute_prime, &which_prime); /* Do some other work here... */ /* Wait for the prime number thread to complete, and get the result. pthread_join (thread, (void*) &prime); /* Print the largest prime it computed. */ printf(“The %dth prime number is %d.\n”, which_prime, prime); return 0; }

*/

*/

68

Chapter 4

Threads

4.1.4 More on Thread IDs
Occasionally, it is useful for a sequence of code to determine which thread is executing it.The pthread_self function returns the thread ID of the thread in which it is called.This thread ID may be compared with another thread ID using the pthread_equal function. These functions can be useful for determining whether a particular thread ID corresponds to the current thread. For instance, it is an error for a thread to call pthread_join to join itself. (In this case, pthread_join would return the error code EDEADLK.) To check for this beforehand, you might use code like this: if (!pthread_equal (pthread_self (), other_thread)) pthread_join (other_thread, NULL);

4.1.5 Thread Attributes
Thread attributes provide a mechanism for fine-tuning the behavior of individual threads. Recall that pthread_create accepts an argument that is a pointer to a thread attribute object. If you pass a null pointer, the default thread attributes are used to configure the new thread. However, you may create and customize a thread attribute object to specify other values for the attributes. To specify customized thread attributes, you must follow these steps: 1. Create a pthread_attr_t object.The easiest way is simply to declare an automatic variable of this type. 2. Call pthread_attr_init, passing a pointer to this object.This initializes the attributes to their default values. 3. Modify the attribute object to contain the desired attribute values. 4. Pass a pointer to the attribute object when calling pthread_create. 5. Call pthread_attr_destroy to release the attribute object.The pthread_attr_t variable itself is not deallocated; it may be reinitialized with pthread_attr_init. A single thread attribute object may be used to start several threads. It is not necessary to keep the thread attribute object around after the threads have been created. For most GNU/Linux application programming tasks, only one thread attribute is typically of interest (the other available attributes are primarily for specialty real-time programming).This attribute is the thread’s detach state. A thread may be created as a joinable thread (the default) or as a detached thread. A joinable thread, like a process, is not automatically cleaned up by GNU/Linux when it terminates. Instead, the thread’s exit state hangs around in the system (kind of like a zombie process) until another thread calls pthread_join to obtain its return value. Only then are its resources released. A detached thread, in contrast, is cleaned up automatically when it terminates. Because a detached thread is immediately cleaned up, another thread may not synchronize on its completion by using pthread_join or obtain its return value.

4.2

Thread Cancellation

69

To set the detach state in a thread attribute object, use pthread_attr_setdetachstate. The first argument is a pointer to the thread attribute object, and the second is the desired detach state. Because the joinable state is the default, it is necessary to call this only to create detached threads; pass PTHREAD_CREATE_DETACHED as the second argument. The code in Listing 4.5 creates a detached thread by setting the detach state thread attribute for the thread.
Listing 4.5 (detached.c) Skeleton Program That Creates a Detached Thread
#include void* thread_function (void* thread_arg) { /* Do work here... */ } int main () { pthread_attr_t attr; pthread_t thread; pthread_attr_init (&attr); pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED); pthread_create (&thread, &attr, &thread_function, NULL); pthread_attr_destroy (&attr); /* Do work here... */ */

/* No need to join the second thread. return 0; }

Even if a thread is created in a joinable state, it may later be turned into a detached thread.To do this, call pthread_detach. Once a thread is detached, it cannot be made joinable again.

4.2 Thread Cancellation
Under normal circumstances, a thread terminates when it exits normally, either by returning from its thread function or by calling pthread_exit. However, it is possible for a thread to request that another thread terminate.This is called canceling a thread. To cancel a thread, call pthread_cancel, passing the thread ID of the thread to be canceled. A canceled thread may later be joined; in fact, you should join a canceled thread to free up its resources, unless the thread is detached (see Section 4.1.5, “Thread Attributes”).The return value of a canceled thread is the special value given by PTHREAD_CANCELED.

70

Chapter 4

Threads

Often a thread may be in some code that must be executed in an all-or-nothing fashion. For instance, the thread may allocate some resources, use them, and then deallocate them. If the thread is canceled in the middle of this code, it may not have the opportunity to deallocate the resources, and thus the resources will be leaked.To counter this possibility, it is possible for a thread to control whether and when it can be canceled. A thread may be in one of three states with regard to thread cancellation. The thread may be asynchronously cancelable.The thread may be canceled at any point in its execution. The thread may be synchronously cancelable.The thread may be canceled, but not at just any point in its execution. Instead, cancellation requests are queued, and the thread is canceled only when it reaches specific points in its execution. A thread may be uncancelable. Attempts to cancel the thread are quietly ignored. n n n

When initially created, a thread is synchronously cancelable.

4.2.1 Synchronous and Asynchronous Threads
An asynchronously cancelable thread may be canceled at any point in its execution. A synchronously cancelable thread, in contrast, may be canceled only at particular places in its execution.These places are called cancellation points.The thread will queue a cancellation request until it reaches the next cancellation point. To make a thread asynchronously cancelable, use pthread_setcanceltype.This affects the thread that actually calls the function.The first argument should be PTHREAD_CANCEL_ASYNCHRONOUS to make the thread asynchronously cancelable, or PTHREAD_CANCEL_DEFERRED to return it to the synchronously cancelable state.The second argument, if not null, is a pointer to a variable that will receive the previous cancellation type for the thread.This call, for example, makes the calling thread asynchronously cancelable. pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, NULL);

What constitutes a cancellation point, and where should these be placed? The most direct way to create a cancellation point is to call pthread_testcancel.This does nothing except process a pending cancellation in a synchronously cancelable thread. You should call pthread_testcancel periodically during lengthy computations in a thread function, at points where the thread can be canceled without leaking any resources or producing other ill effects. Certain other functions are implicitly cancellation points as well.These are listed on the pthread_cancel man page. Note that other functions may use these functions internally and thus will indirectly be cancellation points.

4.2

Thread Cancellation

71

4.2.2 Uncancelable Critical Sections
A thread may disable cancellation of itself altogether with the pthread_setcancelstate function. Like pthread_setcanceltype, this affects the calling thread.The first argument is PTHREAD_CANCEL_DISABLE to disable cancellation, or PTHREAD_CANCEL_ENABLE to re-enable cancellation.The second argument, if not null, points to a variable that will receive the previous cancellation state.This call, for instance, disables thread cancellation in the calling thread. pthread_setcancelstate (PTHREAD_CANCEL_DISABLE, NULL);

Using pthread_setcancelstate enables you to implement critical sections. A critical section is a sequence of code that must be executed either in its entirety or not at all; in other words, if a thread begins executing the critical section, it must continue until the end of the critical section without being canceled. For example, suppose that you’re writing a routine for a banking program that transfers money from one account to another.To do this, you must add value to the balance in one account and deduct the same value from the balance of another account. If the thread running your routine happened to be canceled at just the wrong time between these two operations, the program would have spuriously increased the bank’s total deposits by failing to complete the transaction.To prevent this possibility, place the two operations in a critical section. You might implement the transfer with a function such as process_transaction, shown in Listing 4.6.This function disables thread cancellation to start a critical section before it modifies either account balance.
Listing 4.6 (critical-section.c) Protect a Bank Transaction with a Critical Section
#include #include #include /* An array of balances in accounts, indexed by account number. float* account_balances; /* Transfer DOLLARS from account FROM_ACCT to account TO_ACCT. Return 0 if the transaction succeeded, or 1 if the balance FROM_ACCT is too small. */ int process_transaction (int from_acct, int to_acct, float dollars) { int old_cancel_state; /* Check the balance in FROM_ACCT. */ if (account_balances[from_acct] < dollars) return 1; */

continues

72

Chapter 4

Threads

Listing 4.6 Continued
/* Begin critical section. */ pthread_setcancelstate (PTHREAD_CANCEL_DISABLE, &old_cancel_state); /* Move the money. */ account_balances[to_acct] += dollars; account_balances[from_acct] -= dollars; /* End critical section. */ pthread_setcancelstate (old_cancel_state, NULL); return 0; }

Note that it’s important to restore the old cancel state at the end of the critical section rather than setting it unconditionally to PTHREAD_CANCEL_ENABLE.This enables you to call the process_transaction function safely from within another critical section—in that case, your function will leave the cancel state the same way it found it.

4.2.3 When to Use Thread Cancellation
In general, it’s a good idea not to use thread cancellation to end the execution of a thread, except in unusual circumstances. During normal operation, a better strategy is to indicate to the thread that it should exit, and then to wait for the thread to exit on its own in an orderly fashion.We’ll discuss techniques for communicating with the thread later in this chapter, and in Chapter 5, “Interprocess Communication.”

4.3 Thread-Specific Data
Unlike processes, all threads in a single program share the same address space.This means that if one thread modifies a location in memory (for instance, a global variable), the change is visible to all other threads.This allows multiple threads to operate on the same data without the use interprocess communication mechanisms (which are described in Chapter 5). Each thread has its own call stack, however.This allows each thread to execute different code and to call and return from subroutines in the usual way. As in a singlethreaded program, each invocation of a subroutine in each thread has its own set of local variables, which are stored on the stack for that thread. Sometimes, however, it is desirable to duplicate a certain variable so that each thread has a separate copy. GNU/Linux supports this by providing each thread with a thread-specific data area.The variables stored in this area are duplicated for each thread, and each thread may modify its copy of a variable without affecting other threads. Because all threads share the same memory space, thread-specific data may not be accessed using normal variable references. GNU/Linux provides special functions for setting and retrieving values from the thread-specific data area.

4.3

Thread-Specific Data

73

You may create as many thread-specific data items as you want, each of type void*. Each item is referenced by a key.To create a new key, and thus a new data item for each thread, use pthread_key_create.The first argument is a pointer to a pthread_key_t variable.That key value can be used by each thread to access its own copy of the corresponding data item.The second argument to pthread_key_t is a cleanup function. If you pass a function pointer here, GNU/Linux automatically calls that function when each thread exits, passing the thread-specific value corresponding to that key.This is particularly handy because the cleanup function is called even if the thread is canceled at some arbitrary point in its execution. If the thread-specific value is null, the thread cleanup function is not called. If you don’t need a cleanup function, you may pass null instead of a function pointer. After you’ve created a key, each thread can set its thread-specific value corresponding to that key by calling pthread_setspecific.The first argument is the key, and the second is the void* thread-specific value to store.To retrieve a thread-specific data item, call pthread_getspecific, passing the key as its argument. Suppose, for instance, that your application divides a task among multiple threads. For audit purposes, each thread is to have a separate log file, in which progress messages for that thread’s tasks are recorded.The thread-specific data area is a convenient place to store the file pointer for the log file for each individual thread. Listing 4.7 shows how you might implement this.The main function in this sample program creates a key to store the thread-specific file pointer and then stores it in thread_log_key. Because this is a global variable, it is shared by all threads.When each thread starts executing its thread function, it opens a log file and stores the file pointer under that key. Later, any of these threads may call write_to_thread_log to write a message to the thread-specific log file.That function retrieves the file pointer for the thread’s log file from thread-specific data and writes the message.
Listing 4.7 (tsd.c) Per-Thread Log Files Implemented with Thread-Specific Data
#include #include #include /* The key used to associate a log file pointer with each thread. static pthread_key_t thread_log_key; /* Write MESSAGE to the log file for the current thread. */ */

void write_to_thread_log (const char* message) { FILE* thread_log = (FILE*) pthread_getspecific (thread_log_key); fprintf (thread_log, “%s\n”, message); } /* Close the log file pointer THREAD_LOG. void close_thread_log (void* thread_log) */

continues

74

Chapter 4

Threads

Listing 4.7 Continued
{ fclose ((FILE*) thread_log); } void* thread_function (void* args) { char thread_log_filename[20]; FILE* thread_log; /* Generate the filename for this thread’s log file. */ sprintf (thread_log_filename, “thread%d.log”, (int) pthread_self ()); /* Open the log file. */ thread_log = fopen (thread_log_filename, “w”); /* Store the file pointer in thread-specific data under thread_log_key. pthread_setspecific (thread_log_key, thread_log); write_to_thread_log (“Thread starting.”); /* Do work here... */ return NULL; } int main () { int i; pthread_t threads[5]; /* Create a key to associate thread log file pointers in thread-specific data. Use close_thread_log to clean up the file pointers. */ pthread_key_create (&thread_log_key, close_thread_log); /* Create threads to do the work. */ for (i = 0; i < 5; ++i) pthread_create (&(threads[i]), NULL, thread_function, NULL); /* Wait for all threads to finish. */ for (i = 0; i < 5; ++i) pthread_join (threads[i], NULL); return 0; }

*/

Observe that thread_function does not need to close the log file.That’s because when the log file key was created, close_thread_log was specified as the cleanup function for that key.Whenever a thread exits, GNU/Linux calls that function, passing the thread-specific value for the thread log key.This function takes care of closing the log file.

4.3

Thread-specific Data

75

4.3.1 Cleanup Handlers
The cleanup functions for thread-specific data keys can be very handy for ensuring that resources are not leaked when a thread exits or is canceled. Sometimes, though, it’s useful to be able to specify cleanup functions without creating a new threadspecific data item that’s duplicated for each thread. GNU/Linux provides cleanup handlers for this purpose. A cleanup handler is simply a function that should be called when a thread exits. The handler takes a single void* parameter, and its argument value is provided when the handler is registered—this makes it easy to use the same handler function to deallocate multiple resource instances. A cleanup handler is a temporary measure, used to deallocate a resource only if the thread exits or is canceled instead of finishing execution of a particular region of code. Under normal circumstances, when the thread does not exit and is not canceled, the resource should be deallocated explicitly and the cleanup handler should be removed. To register a cleanup handler, call pthread_cleanup_push, passing a pointer to the cleanup function and the value of its void* argument.The call to pthread_cleanup_push must be balanced by a corresponding call to pthread_cleanup_pop, which unregisters the cleanup handler. As a convenience, pthread_cleanup_pop takes an int flag argument; if the flag is nonzero, the cleanup action is actually performed as it is unregistered. The program fragment in Listing 4.8 shows how you might use a cleanup handler to make sure that a dynamically allocated buffer is cleaned up if the thread terminates.
Listing 4.8 (cleanup.c) Program Fragment Demonstrating a Thread Cleanup Handler
#include #include /* Allocate a temporary buffer. */

void* allocate_buffer (size_t size) { return malloc (size); } /* Deallocate a temporary buffer. */

void deallocate_buffer (void* buffer) { free (buffer); } void do_some_work () { /* Allocate a temporary buffer.

*/

continues

76

Chapter 4

Threads

Listing 4.8 Continued void* temp_buffer = allocate_buffer (1024); /* Register a cleanup handler for this buffer, to deallocate it in case the thread exits or is cancelled. */ pthread_cleanup_push (deallocate_buffer, temp_buffer); /* Do some work here that might call pthread_exit or might be cancelled... */ /* Unregister the cleanup handler. Because we pass a nonzero value, this actually performs the cleanup by calling deallocate_buffer. */ pthread_cleanup_pop (1); }

Because the argument to pthread_cleanup_pop is nonzero in this case, the cleanup function deallocate_buffer is called automatically here and does not need to be called explicitly. In this simple case, we could have used the standard library function free directly as our cleanup handler function instead of deallocate_buffer.

4.3.2 Thread Cleanup in C++
C++ programmers are accustomed to getting cleanup “for free” by wrapping cleanup actions in object destructors.When the objects go out of scope, either because a block is executed to completion or because an exception is thrown, C++ makes sure that destructors are called for those automatic variables that have them.This provides a handy mechanism to make sure that cleanup code is called no matter how the block is exited. If a thread calls pthread_exit, though, C++ doesn’t guarantee that destructors are called for all automatic variables on the thread’s stack. A clever way to recover this functionality is to invoke pthread_exit at the top level of the thread function by throwing a special exception. The program in Listing 4.9 demonstrates this. Using this technique, a function indicates its intention to exit the thread by throwing a ThreadExitException instead of calling pthread_exit directly. Because the exception is caught in the top-level thread function, all local variables on the thread’s stack will be destroyed properly as the exception percolates up.
Listing 4.9 (cxx-exit.cpp) Implementing Safe Thread Exit with C++ Exceptions
#include class ThreadExitException { public: /* Create an exception-signaling thread exit with RETURN_VALUE. ThreadExitException (void* return_value) : thread_return_value_ (return_value)

*/

4.4

Synchronization and Critical Sections

77

{ } /* Actually exit the thread, using the return value provided in the constructor. */ void* DoThreadExit () { pthread_exit (thread_return_value_); } private: /* The return value that will be used when exiting the thread. void* thread_return_value_; }; void do_some_work () { while (1) { /* Do some useful things here...

*/

*/

if (should_exit_thread_immediately ()) throw ThreadExitException (/* thread’s return value = */ NULL); } } void* thread_function (void*) { try { do_some_work (); } catch (ThreadExitException ex) { /* Some function indicated that we should exit the thread. ex.DoThreadExit (); } return NULL; }

*/

4.4 Synchronization and Critical Sections
Programming with threads is very tricky because most threaded programs are concurrent programs. In particular, there’s no way to know when the system will schedule one thread to run and when it will run another. One thread might run for a very long time, or the system might switch among threads very quickly. On a system with multiple processors, the system might even schedule multiple threads to run at literally the same time. Debugging a threaded program is difficult because you cannot always easily reproduce the behavior that caused the problem.You might run the program once and have everything work fine; the next time you run it, it might crash.There’s no way to make the system schedule the threads exactly the same way it did before.

78

Chapter 4

Threads

The ultimate cause of most bugs involving threads is that the threads are accessing the same data. As mentioned previously, that’s one of the powerful aspects of threads, but it can also be dangerous. If one thread is only partway through updating a data structure when another thread accesses the same data structure, chaos is likely to ensue. Often, buggy threaded programs contain a code that will work only if one thread gets scheduled more often—or sooner—than another thread.These bugs are called race conditions; the threads are racing one another to change the same data structure.

4.4.1 Race Conditions
Suppose that your program has a series of queued jobs that are processed by several concurrent threads.The queue of jobs is represented by a linked list of struct job objects. After each thread finishes an operation, it checks the queue to see if an additional job is available. If job_queue is non-null, the thread removes the head of the linked list and sets job_queue to the next job on the list. The thread function that processes jobs in the queue might look like Listing 4.10.
Listing 4.10 ( job-queue1.c) Thread Function to Process Jobs from the Queue
#include struct job { /* Link field for linked list. struct job* next;

*/

/* Other fields describing work to be done... */ }; /* A linked list of pending jobs. struct job* job_queue; */

/* Process queued jobs until the queue is empty. void* thread_function (void* arg) { while (job_queue != NULL) { /* Get the next available job. */ struct job* next_job = job_queue; /* Remove this job from the list. */ job_queue = job_queue->next; /* Carry out the work. */ process_job (next_job); /* Clean up. */ free (next_job); } return NULL; }

*/

4.4

Synchronization and Critical Sections

79

Now suppose that two threads happen to finish a job at about the same time, but only one job remains in the queue.The first thread checks whether job_queue is null; finding that it isn’t, the thread enters the loop and stores the pointer to the job object in next_job. At this point, Linux happens to interrupt the first thread and schedules the second.The second thread also checks job_queue and finding it non-null, also assigns the same job pointer to next_job. By unfortunate coincidence, we now have two threads executing the same job. To make matters worse, one thread will unlink the job object from the queue, leaving job_queue containing null.When the other thread evaluates job_queue->next, a segmentation fault will result. This is an example of a race condition. Under “lucky” circumstances, this particular schedule of the two threads may never occur, and the race condition may never exhibit itself. Only under different circumstances, perhaps when running on a heavily loaded system (or on an important customer’s new multiprocessor server!) may the bug exhibit itself. To eliminate race conditions, you need a way to make operations atomic. An atomic operation is indivisible and uninterruptible; once the operation starts, it will not be paused or interrupted until it completes, and no other operation will take place meanwhile. In this particular example, you want to check job_queue; if it’s not empty, remove the first job, all as a single atomic operation.

4.4.2 Mutexes
The solution to the job queue race condition problem is to let only one thread access the queue of jobs at a time. Once a thread starts looking at the queue, no other thread should be able to access it until the first thread has decided whether to process a job and, if so, has removed the job from the list. Implementing this requires support from the operating system. GNU/Linux provides mutexes, short for MUTual EXclusion locks. A mutex is a special lock that only one thread may lock at a time. If a thread locks a mutex and then a second thread also tries to lock the same mutex, the second thread is blocked, or put on hold. Only when the first thread unlocks the mutex is the second thread unblocked—allowed to resume execution. GNU/Linux guarantees that race conditions do not occur among threads attempting to lock a mutex; only one thread will ever get the lock, and all other threads will be blocked. Think of a mutex as the lock on a lavatory door.Whoever gets there first enters the lavatory and locks the door. If someone else attempts to enter the lavatory while it’s occupied, that person will find the door locked and will be forced to wait outside until the occupant emerges. To create a mutex, create a variable of type pthread_mutex_t and pass a pointer to it to pthread_mutex_init.The second argument to pthread_mutex_init is a pointer to a mutex attribute object, which specifies attributes of the mutex. As with

80

Chapter 4

Threads

pthread_create, if the attribute pointer is null, default attributes are assumed.The mutex variable should be initialized only once.This code fragment demonstrates the declaration and initialization of a mutex variable. pthread_mutex_t mutex; pthread_mutex_init (&mutex, NULL);

Another simpler way to create a mutex with default attributes is to initialize it with the special value PTHREAD_MUTEX_INITIALIZER. No additional call to pthread_mutex_init is necessary.This is particularly convenient for global variables (and, in C++, static data members).The previous code fragment could equivalently have been written like this: pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;

A thread may attempt to lock a mutex by calling pthread_mutex_lock on it. If the mutex was unlocked, it becomes locked and the function returns immediately. If the mutex was locked by another thread, pthread_mutex_lock blocks execution and returns only eventually when the mutex is unlocked by the other thread. More than one thread may be blocked on a locked mutex at one time.When the mutex is unlocked, only one of the blocked threads (chosen unpredictably) is unblocked and allowed to lock the mutex; the other threads stay blocked. A call to pthread_mutex_unlock unlocks a mutex.This function should always be called from the same thread that locked the mutex. Listing 4.11 shows another version of the job queue example. Now the queue is protected by a mutex. Before accessing the queue (either for read or write), each thread locks a mutex first. Only when the entire sequence of checking the queue and removing a job is complete is the mutex unlocked.This prevents the race condition previously described.
Listing 4.11 ( job-queue2.c) Job Queue Thread Function, Protected by a Mutex
#include #include struct job { /* Link field for linked list. struct job* next;

*/

/* Other fields describing work to be done... */ }; /* A linked list of pending jobs. struct job* job_queue; */

/* A mutex protecting job_queue. */ pthread_mutex_t job_queue_mutex = PTHREAD_MUTEX_INITIALIZER;

4.4

Synchronization and Critical Sections

81

/* Process queued jobs until the queue is empty. void* thread_function (void* arg) { while (1) { struct job* next_job;

*/

/* Lock the mutex on the job queue. */ pthread_mutex_lock (&job_queue_mutex); /* Now it’s safe to check if the queue is empty. */ if (job_queue == NULL) next_job = NULL; else { /* Get the next available job. */ next_job = job_queue; /* Remove this job from the list. */ job_queue = job_queue->next; } /* Unlock the mutex on the job queue because we’re done with the queue for now. */ pthread_mutex_unlock (&job_queue_mutex); /* Was the queue empty? if (next_job == NULL) break; If so, end the thread. */

/* Carry out the work. */ process_job (next_job); /* Clean up. */ free (next_job); } return NULL; }

shared data pointer, come between the call to and the call to pthread_mutex_unlock. A job object, stored in next_job, is accessed outside this region only after that object has been removed from the queue and is therefore inaccessible to other threads. Note that if the queue is empty (that is, job_queue is null), we don’t break out of the loop immediately because this would leave the mutex permanently locked and would prevent any other thread from accessing the job queue ever again. Instead, we remember this fact by setting next_job to null and breaking out only after unlocking the mutex. Use of the mutex to lock job_queue is not automatic; it’s up to you to add code to lock the mutex before accessing that variable and then to unlock it afterward. For example, a function to add a job to the job queue might look like this: pthread_mutex_lock void enqueue_job (struct job* new_job) { pthread_mutex_lock (&job_queue_mutex);

All accesses to

job_queue, the

82

Chapter 4

Threads

new_job->next = job_queue; job_queue = new_job; pthread_mutex_unlock (&job_queue_mutex); }

4.4.3 Mutex Deadlocks
Mutexes provide a mechanism for allowing one thread to block the execution of another.This opens up the possibility of a new class of bugs, called deadlocks. A deadlock occurs when one or more threads are stuck waiting for something that never will occur. A simple type of deadlock may occur when the same thread attempts to lock a mutex twice in a row.The behavior in this case depends on what kind of mutex is being used.Three kinds of mutexes exist: n n

n

Locking a fast mutex (the default kind) will cause a deadlock to occur. An attempt to lock the mutex blocks until the mutex is unlocked. But because the thread that locked the mutex is blocked on the same mutex, the lock cannot ever be released. Locking a recursive mutex does not cause a deadlock. A recursive mutex may safely be locked many times by the same thread.The mutex remembers how many times pthread_mutex_lock was called on it by the thread that holds the lock; that thread must make the same number of calls to pthread_mutex_unlock before the mutex is actually unlocked and another thread is allowed to lock it. GNU/Linux will detect and flag a double lock on an error-checking mutex that would otherwise cause a deadlock.The second consecutive call to pthread_mutex_lock returns the failure code EDEADLK.

By default, a GNU/Linux mutex is of the fast kind.To create a mutex of one of the other two kinds, first create a mutex attribute object by declaring a pthread_mutexattr_t variable and calling pthread_mutexattr_init on a pointer to it.Then set the mutex kind by calling pthread_mutexattr_setkind_np; the first argument is a pointer to the mutex attribute object, and the second is PTHREAD_MUTEX_RECURSIVE_NP for a recursive mutex, or PTHREAD_MUTEX_ERRORCHECK_NP for an error-checking mutex. Pass a pointer to this attribute object to pthread_mutex_init to create a mutex of this kind, and then destroy the attribute object with pthread_mutexattr_destroy. This code sequence illustrates creation of an error-checking mutex, for instance: pthread_mutexattr_t attr; pthread_mutex_t mutex; pthread_mutexattr_init (&attr); pthread_mutexattr_setkind_np (&attr, PTHREAD_MUTEX_ERRORCHECK_NP); pthread_mutex_init (&mutex, &attr); pthread_mutexattr_destroy (&attr);

4.4

Synchronization and Critical Sections

83

As suggested by the “np” suffix, the recursive and error-checking mutex kinds are specific to GNU/Linux and are not portable.Therefore, it is generally not advised to use them in programs. (Error-checking mutexes can be useful when debugging, though.)

4.4.4 Nonblocking Mutex Tests
Occasionally, it is useful to test whether a mutex is locked without actually blocking on it. For instance, a thread may need to lock a mutex but may have other work to do instead of blocking if the mutex is already locked. Because pthread_mutex_lock will not return until the mutex becomes unlocked, some other function is necessary. GNU/Linux provides pthread_mutex_trylock for this purpose. If you call pthread_mutex_trylock on an unlocked mutex, you will lock the mutex as if you had called pthread_mutex_lock, and pthread_mutex_trylock will return zero. However, if the mutex is already locked by another thread, pthread_mutex_trylock will not block. Instead, it will return immediately with the error code EBUSY.The mutex lock held by the other thread is not affected.You may try again later to lock the mutex.

4.4.5 Semaphores for Threads
In the preceding example, in which several threads process jobs from a queue, the main thread function of the threads carries out the next job until no jobs are left and then exits the thread.This scheme works if all the jobs are queued in advance or if new jobs are queued at least as quickly as the threads process them. However, if the threads work too quickly, the queue of jobs will empty and the threads will exit. If new jobs are later enqueued, no threads may remain to process them.What we might like instead is a mechanism for blocking the threads when the queue empties until new jobs become available. A semaphore provides a convenient method for doing this. A semaphore is a counter that can be used to synchronize multiple threads. As with a mutex, GNU/Linux guarantees that checking or modifying the value of a semaphore can be done safely, without creating a race condition. Each semaphore has a counter value, which is a non-negative integer. A semaphore supports two basic operations: A wait operation decrements the value of the semaphore by 1. If the value is already zero, the operation blocks until the value of the semaphore becomes positive (due to the action of some other thread).When the semaphore’s value becomes positive, it is decremented by 1 and the wait operation returns. A post operation increments the value of the semaphore by 1. If the semaphore was previously zero and other threads are blocked in a wait operation on that semaphore, one of those threads is unblocked and its wait operation completes (which brings the semaphore’s value back to zero). n n

84

Chapter 4

Threads

Note that GNU/Linux provides two slightly different semaphore implementations. The one we describe here is the POSIX standard semaphore implementation. Use these semaphores when communicating among threads The other implementation, used for communication among processes, is described in Section 5.2, “Process Semaphores.” If you use semaphores, include . A semaphore is represented by a sem_t variable. Before using it, you must initialize it using the sem_init function, passing a pointer to the sem_t variable.The second parameter should be zero,2 and the third parameter is the semaphore’s initial value. If you no longer need a semaphore, it’s good to deallocate it with sem_destroy. To wait on a semaphore, use sem_wait.To post to a semaphore, use sem_post. A nonblocking wait function, sem_trywait, is also provided. It’s similar to pthread_mutex_trylock—if the wait would have blocked because the semaphore’s value was zero, the function returns immediately, with error value EAGAIN, instead of blocking. GNU/Linux also provides a function to retrieve the current value of a semaphore, sem_getvalue, which places the value in the int variable pointed to by its second argument.You should not use the semaphore value you get from this function to make a decision whether to post to or wait on the semaphore, though.To do this could lead to a race condition: Another thread could change the semaphore’s value between the call to sem_getvalue and the call to another semaphore function. Use the atomic post and wait functions instead. Returning to our job queue example, we can use a semaphore to count the number of jobs waiting in the queue. Listing 4.12 controls the queue with a semaphore. The function enqueue_job adds a new job to the queue.
Listing 4.12 ( job-queue3.c) Job Queue Controlled by a Semaphore
#include #include #include struct job { /* Link field for linked list. struct job* next;

*/

/* Other fields describing work to be done... */ }; /* A linked list of pending jobs. struct job* job_queue; */

/* A mutex protecting job_queue. */ pthread_mutex_t job_queue_mutex = PTHREAD_MUTEX_INITIALIZER;

2. A nonzero value would indicate a semaphore that can be shared across processes, which is not supported by GNU/Linux for this type of semaphore.

4.4

Synchronization and Critical Sections

85

/* A semaphore counting the number of jobs in the queue. sem_t job_queue_count; /* Perform one-time initialization of the job queue. */

*/

void initialize_job_queue () { /* The queue is initially empty. */ job_queue = NULL; /* Initialize the semaphore which counts jobs in the queue. initial value should be zero. */ sem_init (&job_queue_count, 0, 0); } /* Process queued jobs until the queue is empty. void* thread_function (void* arg) { while (1) { struct job* next_job; */

Its

/* Wait on the job queue semaphore. If its value is positive, indicating that the queue is not empty, decrement the count by 1. If the queue is empty, block until a new job is enqueued. */ sem_wait (&job_queue_count); /* Lock the mutex on the job queue. */ pthread_mutex_lock (&job_queue_mutex); /* Because of the semaphore, we know the queue is not empty. Get the next available job. */ next_job = job_queue; /* Remove this job from the list. */ job_queue = job_queue->next; /* Unlock the mutex on the job queue because we’re done with the queue for now. */ pthread_mutex_unlock (&job_queue_mutex); /* Carry out the work. */ process_job (next_job); /* Clean up. */ free (next_job); } return NULL; } /* Add a new job to the front of the job queue. */ */)

void enqueue_job (/* Pass job-specific data here... { struct job* new_job;

continues

86

Chapter 4

Threads

Listing 4.12 Continued
/* Allocate a new job object. */ new_job = (struct job*) malloc (sizeof (struct job)); /* Set the other fields of the job struct here... */ /* Lock the mutex on the job queue before accessing it. pthread_mutex_lock (&job_queue_mutex); /* Place the new job at the head of the queue. */ new_job->next = job_queue; job_queue = new_job; */

/* Post to the semaphore to indicate that another job is available. threads are blocked, waiting on the semaphore, one will become unblocked so it can process the job. */ sem_post (&job_queue_count); /* Unlock the job queue mutex. */ pthread_mutex_unlock (&job_queue_mutex); }

If

Before taking a job from the front of the queue, each thread will first wait on the semaphore. If the semaphore’s value is zero, indicating that the queue is empty, the thread will simply block until the semaphore’s value becomes positive, indicating that a job has been added to the queue. The enqueue_job function adds a job to the queue. Just like thread_function, it needs to lock the queue mutex before modifying the queue. After adding a job to the queue, it posts to the semaphore, indicating that a new job is available. In the version shown in Listing 4.12, the threads that process the jobs never exit; if no jobs are available for a while, all the threads simply block in sem_wait.

4.4.6

Condition Variables

We’ve shown how to use a mutex to protect a variable against simultaneous access by two threads and how to use semaphores to implement a shared counter. A condition variable is a third synchronization device that GNU/Linux provides; with it, you can implement more complex conditions under which threads execute. Suppose that you write a thread function that executes a loop infinitely, performing some work on each iteration.The thread loop, however, needs to be controlled by a flag:The loop runs only when the flag is set; when the flag is not set, the loop pauses. Listing 4.13 shows how you might implement this by spinning in a loop. During each iteration of the loop, the thread function checks that the flag is set. Because the flag is accessed by multiple threads, it is protected by a mutex.This implementation may be correct, but it is not efficient.The thread function will spend lots of CPU

4.4

Synchronization and Critical Sections

87

whenever the flag is not set, checking and rechecking the flag, each time locking and unlocking the mutex.What you really want is a way to put the thread to sleep when the flag is not set, until some circumstance changes that might cause the flag to become set.
Listing 4.13 (spin-condvar.c) A Simple Condition Variable Implementation
#include int thread_flag; pthread_mutex_t thread_flag_mutex; void initialize_flag () { pthread_mutex_init (&thread_flag_mutex, NULL); thread_flag = 0; } /* Calls do_work repeatedly while the thread flag is set; otherwise spins. */ void* thread_function (void* thread_arg) { while (1) { int flag_is_set; /* Protect the flag with a mutex lock. */ pthread_mutex_lock (&thread_flag_mutex); flag_is_set = thread_flag; pthread_mutex_unlock (&thread_flag_mutex); if (flag_is_set) do_work (); /* Else don’t do anything. } return NULL; } /* Sets the value of the thread flag to FLAG_VALUE. void set_thread_flag (int flag_value) { /* Protect the flag with a mutex lock. */ pthread_mutex_lock (&thread_flag_mutex); thread_flag = flag_value; pthread_mutex_unlock (&thread_flag_mutex); } */

Just loop again.

*/

88

Chapter 4

Threads

A condition variable enables you to implement a condition under which a thread executes and, inversely, the condition under which the thread is blocked. As long as every thread that potentially changes the sense of the condition uses the condition variable properly, Linux guarantees that threads blocked on the condition will be unblocked when the condition changes. As with a semaphore, a thread may wait on a condition variable. If thread A waits on a condition variable, it is blocked until some other thread, thread B, signals the same condition variable. Unlike a semaphore, a condition variable has no counter or memory; thread A must wait on the condition variable before thread B signals it. If thread B signals the condition variable before thread A waits on it, the signal is lost, and thread A blocks until some other thread signals the condition variable again. This is how you would use a condition variable to make the previous sample more efficient: The loop in thread_function checks the flag. If the flag is not set, the thread waits on the condition variable. The set_thread_flag function signals the condition variable after changing the flag value.That way, if thread_function is blocked on the condition variable, it will be unblocked and will check the condition again. n n

There’s one problem with this:There’s a race condition between checking the flag value and signaling or waiting on the condition variable. Suppose that thread_function checked the flag and found that it was not set. At that moment, the Linux scheduler paused that thread and resumed the main one. By some coincidence, the main thread is in set_thread_flag. It sets the flag and then signals the condition variable. Because no thread is waiting on the condition variable at the time (remember that thread_function was paused before it could wait on the condition variable), the signal is lost. Now, when Linux reschedules the other thread, it starts waiting on the condition variable and may end up blocked forever. To solve this problem, we need a way to lock the flag and the condition variable together with a single mutex. Fortunately, GNU/Linux provides exactly this mechanism. Each condition variable must be used in conjunction with a mutex, to prevent this sort of race condition. Using this scheme, the thread function follows these steps: 1. The loop in thread_function locks the mutex and reads the flag value. 2. If the flag is set, it unlocks the mutex and executes the work function. 3. If the flag is not set, it atomically unlocks the mutex and waits on the condition variable. The critical feature here is in step 3, in which GNU/Linux allows you to unlock the mutex and wait on the condition variable atomically, without the possibility of another thread intervening.This eliminates the possibility that another thread may change the flag value and signal the condition variable in between thread_function’s test of the flag value and wait on the condition variable.

4.4

Synchronization and Critical Sections

89

A condition variable is represented by an instance of pthread_cond_t. Remember that each condition variable should be accompanied by a mutex.These are the functions that manipulate condition variables: pthread_cond_init initializes a condition variable.The first argument is a pointer to a pthread_cond_t instance.The second argument, a pointer to a condition variable attribute object, is ignored under GNU/Linux. The mutex must be initialized separately, as described in Section 4.4.2, “Mutexes.” pthread_cond_signal signals a condition variable. A single thread that is blocked on the condition variable will be unblocked. If no other thread is blocked on the condition variable, the signal is ignored.The argument is a pointer to the pthread_cond_t instance. A similar call, pthread_cond_broadcast, unblocks all threads that are blocked on the condition variable, instead of just one. pthread_cond_wait blocks the calling thread until the condition variable is signaled.The argument is a pointer to the pthread_cond_t instance.The second argument is a pointer to the pthread_mutex_t mutex instance. When pthread_cond_wait is called, the mutex must already be locked by the calling thread.That function atomically unlocks the mutex and blocks on the condition variable.When the condition variable is signaled and the calling thread unblocks, pthread_cond_wait automatically reacquires a lock on the mutex. n n n

Whenever your program performs an action that may change the sense of the condition you’re protecting with the condition variable, it should perform these steps. (In our example, the condition is the state of the thread flag, so these steps must be taken whenever the flag is changed.) 1. Lock the mutex accompanying the condition variable. 2. Take the action that may change the sense of the condition (in our example, set the flag). 3. Signal or broadcast the condition variable, depending on the desired behavior. 4. Unlock the mutex accompanying the condition variable. Listing 4.14 shows the previous example again, now using a condition variable to protect the thread flag. Note that in thread_function, a lock on the mutex is held before checking the value of thread_flag.That lock is automatically released by pthread_cond_wait before blocking and is automatically reacquired afterward. Also note that set_thread_flag locks the mutex before setting the value of thread_flag and signaling the mutex.

90

Chapter 4

Threads

Listing 4.14 (condvar.c) Control a Thread Using a Condition Variable
#include int thread_flag; pthread_cond_t thread_flag_cv; pthread_mutex_t thread_flag_mutex; void initialize_flag () { /* Initialize the mutex and condition variable. pthread_mutex_init (&thread_flag_mutex, NULL); pthread_cond_init (&thread_flag_cv, NULL); /* Initialize the flag value. */ thread_flag = 0; }

*/

/* Calls do_work repeatedly while the thread flag is set; blocks if the flag is clear. */ void* thread_function (void* thread_arg) { /* Loop infinitely. */ while (1) { /* Lock the mutex before accessing the flag value. */ pthread_mutex_lock (&thread_flag_mutex); while (!thread_flag) /* The flag is clear. Wait for a signal on the condition variable, indicating that the flag value has changed. When the signal arrives and this thread unblocks, loop and check the flag again. */ pthread_cond_wait (&thread_flag_cv, &thread_flag_mutex); /* When we’ve gotten here, we know the flag must be set. Unlock the mutex. */ pthread_mutex_unlock (&thread_flag_mutex); /* Do some work. */ do_work (); } return NULL; } /* Sets the value of the thread flag to FLAG_VALUE. */

void set_thread_flag (int flag_value) { /* Lock the mutex before accessing the flag value. */ pthread_mutex_lock (&thread_flag_mutex); /* Set the flag value, and then signal in case thread_function is blocked, waiting for the flag to become set. However, thread_function can’t actually check the flag until the mutex is unlocked. */

4.4

Synchronization and Critical Sections

91

thread_flag = flag_value; pthread_cond_signal (&thread_flag_cv); /* Unlock the mutex. */ pthread_mutex_unlock (&thread_flag_mutex); }

The condition protected by a condition variable can be arbitrarily complex. However, before performing any operation that may change the sense of the condition, a mutex lock should be required, and the condition variable should be signaled afterward. A condition variable may also be used without a condition, simply as a mechanism for blocking a thread until another thread “wakes it up.” A semaphore may also be used for that purpose.The principal difference is that a semaphore “remembers” the wake-up call even if no thread was blocked on it at the time, while a condition variable discards the wake-up call unless some thread is actually blocked on it at the time. Also, a semaphore delivers only a single wake-up per post; with pthread_cond_broadcast, an arbitrary and unknown number of blocked threads may be awoken at the same time.

4.4.7 Deadlocks with Two or More Threads
Deadlocks can occur when two (or more) threads are each blocked, waiting for a condition to occur that only the other one can cause. For instance, if thread A is blocked on a condition variable waiting for thread B to signal it, and thread B is blocked on a condition variable waiting for thread A to signal it, a deadlock has occurred because neither thread will ever signal the other.You should take care to avoid the possibility of such situations because they are quite difficult to detect. One common error that can cause a deadlock involves a problem in which more than one thread is trying to lock the same set of objects. For example, consider a program in which two different threads, running two different thread functions, need to lock the same two mutexes. Suppose that thread A locks mutex 1 and then mutex 2, and thread B happens to lock mutex 2 before mutex 1. In a sufficiently unfortunate scheduling scenario, Linux may schedule thread A long enough to lock mutex 1, and then schedule thread B, which promptly locks mutex 2. Now neither thread can progress because each is blocked on a mutex that the other thread holds locked. This is an example of a more general deadlock problem, which can involve not only synchronization objects such as mutexes, but also other resources, such as locks on files or devices.The problem occurs when multiple threads try to lock the same set of resources in different orders.The solution is to make sure that all threads that lock more than one resource lock them in the same order.

92

Chapter 4

Threads

4.5 GNU/Linux Thread Implementation
The implementation of POSIX threads on GNU/Linux differs from the thread implementation on many other UNIX-like systems in an important way: on GNU/Linux, threads are implemented as processes.Whenever you call pthread_create to create a new thread, Linux creates a new process that runs that thread. However, this process is not the same as a process you would create with fork; in particular, it shares the same address space and resources as the original process rather than receiving copies. The program thread-pid shown in Listing 4.15 demonstrates this.The program creates a thread; both the original thread and the new one call the getpid function and print their respective process IDs and then spin infinitely.
Listing 4.15 (thread-pid) Print Process IDs for Threads
#include #include #include void* thread_function (void* arg) { fprintf (stderr, “child thread pid is %d\n”, (int) getpid ()); /* Spin forever. */ while (1); return NULL; } int main () { pthread_t thread; fprintf (stderr, “main thread pid is %d\n”, (int) getpid ()); pthread_create (&thread, NULL, &thread_function, NULL); /* Spin forever. */ while (1); return 0; }

Run the program in the background, and then invoke ps x to display your running processes. Don’t forget to kill the thread-pid program afterward—it consumes lots of CPU doing nothing. Here’s what the output might look like:
% cc thread-pid.c -o thread-pid -lpthread % ./thread-pid & [1] 14608 main thread pid is 14608 child thread pid is 14610 % ps x PID TTY STAT TIME COMMAND 14042 pts/9 S 0:00 bash 14608 pts/9 R 0:01 ./thread-pid

4.5

GNU/Linux Thread Implementation

93

14609 pts/9 S 14610 pts/9 R 14611 pts/9 R % kill 14608 [1]+ Terminated

0:00 ./thread-pid 0:01 ./thread-pid 0:00 ps x ./thread-pid

Job Control Notification in the Shell
The lines starting with [1] are from the shell. When you run a program in the background, the shell assigns a job number to it—in this case, 1—and prints out the program’s pid. If a background job terminates, the shell reports that fact the next time you invoke a command.

Notice that there are three processes running the thread-pid program.The first of these, with pid 14608, is the main thread in the program; the third, with pid 14610, is the thread we created to execute thread_function. How about the second thread, with pid 14609? This is the “manager thread,” which is part of the internal implementation of GNU/Linux threads.The manager thread is created the first time a program calls pthread_create to create a new thread.

4.5.1 Signal Handling
Suppose that a multithreaded program receives a signal. In which thread is the signal handler invoked? The behavior of the interaction between signals and threads varies from one UNIX-like system to another. In GNU/Linux, the behavior is dictated by the fact that threads are implemented as processes. Because each thread is a separate process, and because a signal is delivered to a particular process, there is no ambiguity about which thread receives the signal.Typically, signals sent from outside the program are sent to the process corresponding to the main thread of the program. For instance, if a program forks and the child process execs a multithreaded program, the parent process will hold the process id of the main thread of the child process’s program and will use that process id to send signals to its child.This is generally a good convention to follow yourself when sending signals to a multithreaded program. Note that this aspect of GNU/Linux’s implementation of pthreads is at variance with the POSIX thread standard. Do not rely on this behavior in programs that are meant to be portable. Within a multithreaded program, it is possible for one thread to send a signal specifically to another thread. Use the pthread_kill function to do this. Its first parameter is a thread ID, and its second parameter is a signal number.

4.5.2 The clone System Call
Although GNU/Linux threads created in the same program are implemented as separate processes, they share their virtual memory space and other resources. A child process created with fork, however, gets copies of these items. How is the former type of process created?

94

Chapter 4

Threads

The Linux clone system call is a generalized form of fork and pthread_create that allows the caller to specify which resources are shared between the calling process and the newly created process. Also, clone requires you to specify the memory region for the execution stack that the new process will use. Although we mention clone here to satisfy the reader’s curiosity, that system call should not ordinarily be used in programs. Use fork to create new processes or pthread_create to create threads.

4.6 Processes Vs. Threads
For some programs that benefit from concurrency, the decision whether to use processes or threads can be difficult. Here are some guidelines to help you decide which concurrency model best suits your program: All threads in a program must run the same executable. A child process, on the other hand, may run a different executable by calling an exec function. An errant thread can harm other threads in the same process because threads share the same virtual memory space and other resources. For instance, a wild memory write through an uninitialized pointer in one thread can corrupt memory visible to another thread. An errant process, on the other hand, cannot do so because each process has a copy of the program’s memory space. Copying memory for a new process adds an additional performance overhead relative to creating a new thread. However, the copy is performed only when the memory is changed, so the penalty is minimal if the child process only reads memory. Threads should be used for programs that need fine-grained parallelism. For example, if a problem can be broken into multiple, nearly identical tasks, threads may be a good choice. Processes should be used for programs that need coarser parallelism. Sharing data among threads is trivial because threads share the same memory. (However, great care must be taken to avoid race conditions, as described previously.) Sharing data among processes requires the use of IPC mechanisms, as described in Chapter 5.This can be more cumbersome but makes multiple processes less likely to suffer from concurrency bugs. n n n n n

5
Interprocess Communication

HAPTER 3, “PROCESSES,” DISCUSSED THE CREATION OF PROCESSES and showed how one process can obtain the exit status of a child process.That’s the simplest form of communication between two processes, but it’s by no means the most powerful.The mechanisms of Chapter 3 don’t provide any way for the parent to communicate with the child except via command-line arguments and environment variables, nor any way for the child to communicate with the parent except via the child’s exit status. None of these mechanisms provides any means for communicating with the child process while it is actually running, nor do these mechanisms allow communication with a process outside the parent-child relationship. This chapter describes means for interprocess communication that circumvent these limitations.We will present various ways for communicating between parents and children, between “unrelated” processes, and even between processes on different machines. Interprocess communication (IPC) is the transfer of data among processes. For example, a Web browser may request a Web page from a Web server, which then sends HTML data.This transfer of data usually uses sockets in a telephone-like connection. In another example, you may want to print the filenames in a directory using a command such as ls | lpr.The shell creates an ls process and a separate lpr process, connecting

C

96

Chapter 5 Interprocess Communication

the two with a pipe, represented by the “|” symbol. A pipe permits one-way communication between two related processes.The ls process writes data into the pipe, and the lpr process reads data from the pipe. In this chapter, we discuss five types of interprocess communication: Shared memory permits processes to communicate by simply reading and writing to a specified memory location. Mapped memory is similar to shared memory, except that it is associated with a file in the filesystem. Pipes permit sequential communication from one process to a related process. FIFOs are similar to pipes, except that unrelated processes can communicate because the pipe is given a name in the filesystem. Sockets support communication between unrelated processes even on different computers. n n n n n

These types of IPC differ by the following criteria: Whether they restrict communication to related processes (processes with a common ancestor), to unrelated processes sharing the same filesystem, or to any computer connected to a network Whether a communicating process is limited to only write data or only read data The number of processes permitted to communicate Whether the communicating processes are synchronized by the IPC—for example, a reading process halts until data is available to read n n n n

In this chapter, we omit discussion of IPC permitting communication only a limited number of times, such as communicating via a child’s exit value.

5.1 Shared Memory
One of the simplest interprocess communication methods is using shared memory. Shared memory allows two or more processes to access the same memory as if they all called malloc and were returned pointers to the same actual memory.When one process changes the memory, all the other processes see the modification.

5.1.1 Fast Local Communication
Shared memory is the fastest form of interprocess communication because all processes share the same piece of memory. Access to this shared memory is as fast as accessing a process’s nonshared memory, and it does not require a system call or entry to the kernel. It also avoids copying data unnecessarily.

5.1

Shared Memory

97

Because the kernel does not synchronize accesses to shared memory, you must provide your own synchronization. For example, a process should not read from the memory until after data is written there, and two processes must not write to the same memory location at the same time. A common strategy to avoid these race conditions is to use semaphores, which are discussed in the next section. Our illustrative programs, though, show just a single process accessing the memory, to focus on the shared memory mechanism and to avoid cluttering the sample code with synchronization logic.

5.1.2 The Memory Model
To use a shared memory segment, one process must allocate the segment.Then each process desiring to access the segment must attach the segment. After finishing its use of the segment, each process detaches the segment. At some point, one process must deallocate the segment. Understanding the Linux memory model helps explain the allocation and attachment process. Under Linux, each process’s virtual memory is split into pages. Each process maintains a mapping from its memory addresses to these virtual memory pages, which contain the actual data. Even though each process has its own addresses, multiple processes’ mappings can point to the same page, permitting sharing of memory. Memory pages are discussed further in Section 8.8,“The mlock Family: Locking Physical Memory,” of Chapter 8,“Linux System Calls.” Allocating a new shared memory segment causes virtual memory pages to be created. Because all processes desire to access the same shared segment, only one process should allocate a new shared segment. Allocating an existing segment does not create new pages, but it does return an identifier for the existing pages.To permit a process to use the shared memory segment, a process attaches it, which adds entries mapping from its virtual memory to the segment’s shared pages.When finished with the segment, these mapping entries are removed.When no more processes want to access these shared memory segments, exactly one process must deallocate the virtual memory pages. All shared memory segments are allocated as integral multiples of the system’s page size, which is the number of bytes in a page of memory. On Linux systems, the page size is 4KB, but you should obtain this value by calling the getpagesize function.

5.1.3

Allocation

A process allocates a shared memory segment using shmget (“SHared Memory GET”). Its first parameter is an integer key that specifies which segment to create. Unrelated processes can access the same shared segment by specifying the same key value. Unfortunately, other processes may have also chosen the same fixed key, which could lead to conflict. Using the special constant IPC_PRIVATE as the key value guarantees that a brand new memory segment is created.

98

Chapter 5 Interprocess Communication

Its second parameter specifies the number of bytes in the segment. Because segments are allocated using pages, the number of actually allocated bytes is rounded up to an integral multiple of the page size. The third parameter is the bitwise or of flag values that specify options to shmget. The flag values include these: IPC_CREAT—This flag indicates that a new segment should be created.This permits creating a new segment while specifying a key value. IPC_EXCL—This flag, which is always used with IPC_CREAT, causes shmget to fail if a segment key is specified that already exists.Therefore, it arranges for the calling process to have an “exclusive” segment. If this flag is not given and the key of an existing segment is used, shmget returns the existing segment instead of creating a new one. Mode flags—This value is made of 9 bits indicating permissions granted to owner, group, and world to control access to the segment. Execution bits are ignored. An easy way to specify permissions is to use the constants defined in and documented in the section 2 stat man page.1 For example, S_IRUSR and S_IWUSR specify read and write permissions for the owner of the shared memory segment, and S_IROTH and S_IWOTH specify read and write permissions for others. n n n

For example, this invocation of shmget creates a new shared memory segment (or access to an existing one, if shm_key is already used) that’s readable and writeable to the owner but not other users. int segment_id = shmget (shm_key, getpagesize (), IPC_CREAT | S_IRUSR | S_IWUSER);

If the call succeeds, shmget returns a segment identifier. If the shared memory segment already exists, the access permissions are verified and a check is made to ensure that the segment is not marked for destruction.

5.1.4 Attachment and Detachment
To make the shared memory segment available, a process must use shmat, “SHared Memory ATtach.” Pass it the shared memory segment identifier SHMID returned by shmget.The second argument is a pointer that specifies where in your process’s address space you want to map the shared memory; if you specify NULL, Linux will choose an available address.The third argument is a flag, which can include the following: SHM_RND indicates that the address specified for the second parameter should be rounded down to a multiple of the page size. If you don’t specify this flag, you must page-align the second argument to shmat yourself. SHM_RDONLY indicates that the segment will be only read, not written. n n

1.These permission bits are the same as those used for files.They are described in Section 10.3, “File System Permissions.”

5.1

Shared Memory

99

If the call succeeds, it returns the address of the attached shared segment. Children created by calls to fork inherit attached shared segments; they can detach the shared memory segments, if desired. When you’re finished with a shared memory segment, the segment should be detached using shmdt (“SHared Memory DeTach”). Pass it the address returned by shmat. If the segment has been deallocated and this was the last process using it, it is removed. Calls to exit and any of the exec family automatically detach segments.

5.1.5 Controlling and Deallocating Shared Memory
The shmctl (“SHared Memory ConTroL”) call returns information about a shared memory segment and can modify it.The first parameter is a shared memory segment identifier. To obtain information about a shared memory segment, pass IPC_STAT as the second argument and a pointer to a struct shmid_ds. To remove a segment, pass IPC_RMID as the second argument, and pass NULL as the third argument.The segment is removed when the last process that has attached it finally detaches it. Each shared memory segment should be explicitly deallocated using shmctl when you’re finished with it, to avoid violating the systemwide limit on the total number of shared memory segments. Invoking exit and exec detaches memory segments but does not deallocate them. See the shmctl man page for a description of other operations you can perform on shared memory segments.

5.1.6 An Example Program
The program in Listing 5.1 illustrates the use of shared memory.
Listing 5.1 (shm.c) Exercise Shared Memory
#include #include #include int main () { int segment_id; char* shared_memory; struct shmid_ds shmbuffer; int segment_size; const int shared_segment_size = 0x6400; /* Allocate a shared memory segment. */ segment_id = shmget (IPC_PRIVATE, shared_segment_size, IPC_CREAT | IPC_EXCL | S_IRUSR | S_IWUSR);

continues

100 Chapter 5 Interprocess Communication

Listing 5.1 Continued
/* Attach the shared memory segment. */ shared_memory = (char*) shmat (segment_id, 0, 0); printf (“shared memory attached at address %p\n”, shared_memory); /* Determine the segment’s size. */ shmctl (segment_id, IPC_STAT, &shmbuffer); segment_size = shmbuffer.shm_segsz; printf (“segment size: %d\n”, segment_size); /* Write a string to the shared memory segment. */ sprintf (shared_memory, “Hello, world.”); /* Detach the shared memory segment. */ shmdt (shared_memory); /* Reattach the shared memory segment, at a different address. */ shared_memory = (char*) shmat (segment_id, (void*) 0x5000000, 0); printf (“shared memory reattached at address %p\n”, shared_memory); /* Print out the string from shared memory. */ printf (“%s\n”, shared_memory); /* Detach the shared memory segment. */ shmdt (shared_memory); /* Deallocate the shared memory segment. shmctl (segment_id, IPC_RMID, 0); return 0; } */

5.1.7 Debugging
The ipcs command provides information on interprocess communication facilities, including shared segments. Use the -m flag to obtain information about shared memory. For example, this code illustrates that one shared memory segment, numbered 1627649, is in use:
% ipcs -m ------ Shared Memory Segments -------key shmid owner perms bytes 0x00000000 1627649 user 640 25600

nattch 0

status

If this memory segment was erroneously left behind by a program, you can use the ipcrm command to remove it.
% ipcrm shm 1627649

5.2

Processes Semaphores 101

5.1.8 Pros and Cons
Shared memory segments permit fast bidirectional communication among any number of processes. Each user can both read and write, but a program must establish and follow some protocol for preventing race conditions such as overwriting information before it is read. Unfortunately, Linux does not strictly guarantee exclusive access even if you create a new shared segment with IPC_PRIVATE. Also, for multiple processes to use a shared segment, they must make arrangements to use the same key.

5.2 Processes Semaphores
As noted in the previous section, processes must coordinate access to shared memory. As we discussed in Section 4.4.5, “Semaphores for Threads,” in Chapter 4, “Threads,” semaphores are counters that permit synchronizing multiple threads. Linux provides a distinct alternate implementation of semaphores that can be used for synchronizing processes (called process semaphores or sometimes System V semaphores). Process semaphores are allocated, used, and deallocated like shared memory segments. Although a single semaphore is sufficient for almost all uses, process semaphores come in sets. Throughout this section, we present system calls for process semaphores, showing how to implement single binary semaphores using them.

5.2.1 Allocation and Deallocation
The calls semget and semctl allocate and deallocate semaphores, which is analogous to shmget and shmctl for shared memory. Invoke semget with a key specifying a semaphore set, the number of semaphores in the set, and permission flags as for shmget; the return value is a semaphore set identifier.You can obtain the identifier of an existing semaphore set by specifying the right key value; in this case, the number of semaphores can be zero. Semaphores continue to exist even after all processes using them have terminated. The last process to use a semaphore set must explicitly remove it to ensure that the operating system does not run out of semaphores.To do so, invoke semctl with the semaphore identifier, the number of semaphores in the set, IPC_RMID as the third argument, and any union semun value as the fourth argument (which is ignored).The effective user ID of the calling process must match that of the semaphore’s allocator (or the caller must be root). Unlike shared memory segments, removing a semaphore set causes Linux to deallocate immediately. Listing 5.2 presents functions to allocate and deallocate a binary semaphore.

102 Chapter 5 Interprocess Communication

Listing 5.2 (sem_all_deall.c) Allocating and Deallocating a Binary Semaphore
#include #include #include /* We must define union semun ourselves. union semun { int val; struct semid_ds *buf; unsigned short int *array; struct seminfo *__buf; }; /* Obtain a binary semaphore’s ID, allocating if necessary. int binary_semaphore_allocation (key_t key, int sem_flags) { return semget (key, 1, sem_flags); } /* Deallocate a binary semaphore. All users must have finished their use. Returns -1 on failure. */ int binary_semaphore_deallocate (int semid) { union semun ignored_argument; return semctl (semid, 1, IPC_RMID, ignored_argument); } */ */

5.2.2 Initializing Semaphores
Allocating and initializing semaphores are two separate operations.To initialize a semaphore, use semctl with zero as the second argument and SETALL as the third argument. For the fourth argument, you must create a union semun object and point its array field at an array of unsigned short values. Each value is used to initialize one semaphore in the set. Listing 5.3 presents a function that initializes a binary semaphore.
Listing 5.3 (sem_init.c) Initializing a Binary Semaphore
#include #include #include

5.2

Processes Semaphores 103

/* We must define union semun ourselves. union semun { int val; struct semid_ds *buf; unsigned short int *array; struct seminfo *__buf; };

*/

/* Initialize a binary semaphore with a value of 1. int binary_semaphore_initialize (int semid) { union semun argument; unsigned short values[1]; values[0] = 1; argument.array = values; return semctl (semid, 0, SETALL, argument); }

*/

5.2.3 Wait and Post Operations
Each semaphore has a non-negative value and supports wait and post operations.The semop system call implements both operations. Its first parameter specifies a semaphore set identifier. Its second parameter is an array of struct sembuf elements, which specify the operations you want to perform.The third parameter is the length of this array. The fields of struct sembuf are listed here: sem_num is the semaphore number in the semaphore set on which the operation is performed. sem_op is an integer that specifies the semaphore operation. If sem_op is a positive number, that number is added to the semaphore value immediately. If sem_op is a negative number, the absolute value of that number is subtracted from the semaphore value. If this would make the semaphore value negative, the call blocks until the semaphore value becomes as large as the absolute value of sem_op (because some other process increments it). If sem_op is zero, the operation blocks until the semaphore value becomes zero. sem_flg is a flag value. Specify IPC_NOWAIT to prevent the operation from blocking; if the operation would have blocked, the call to semop fails instead. If you specify SEM_UNDO, Linux automatically undoes the operation on the semaphore when the process exits. n n n

104 Chapter 5 Interprocess Communication

Listing 5.4 illustrates wait and post operations for a binary semaphore.
Listing 5.4 (sem_pv.c) Wait and Post Operations for a Binary Semaphore
#include #include #include /* Wait on a binary semaphore. decrement it by 1. */ Block until the semaphore value is positive, then

int binary_semaphore_wait (int semid) { struct sembuf operations[1]; /* Use the first (and only) semaphore. operations[0].sem_num = 0; /* Decrement by 1. */ operations[0].sem_op = -1; /* Permit undo’ing. */ operations[0].sem_flg = SEM_UNDO; return semop (semid, operations, 1); }

*/

/* Post to a binary semaphore: increment its value by 1. This returns immediately. */ int binary_semaphore_post (int semid) { struct sembuf operations[1]; /* Use the first (and only) semaphore. operations[0].sem_num = 0; /* Increment by 1. */ operations[0].sem_op = 1; /* Permit undo’ing. */ operations[0].sem_flg = SEM_UNDO; return semop (semid, operations, 1); }

*/

Specifying the SEM_UNDO flag permits dealing with the problem of terminating a process while it has resources allocated through a semaphore.When a process terminates, either voluntarily or involuntarily, the semaphore’s values are automatically adjusted to “undo” the process’s effects on the semaphore. For example, if a process that has decremented a semaphore is killed, the semaphore’s value is incremented.

5.3

Mapped Memory 105

5.2.4 Debugging Semaphores
Use the command ipcs -s to display information about existing semaphore sets. Use the ipcrm sem command to remove a semaphore set from the command line. For example, to remove the semaphore set with identifier 5790517, use this line:
% ipcrm sem 5790517

5.3

Mapped Memory

Mapped memory permits different processes to communicate via a shared file. Although you can think of mapped memory as using a shared memory segment with a name, you should be aware that there are technical differences. Mapped memory can be used for interprocess communication or as an easy way to access the contents of a file. Mapped memory forms an association between a file and a process’s memory. Linux splits the file into page-sized chunks and then copies them into virtual memory pages so that they can be made available in a process’s address space.Thus, the process can read the file’s contents with ordinary memory access. It can also modify the file’s contents by writing to memory.This permits fast access to files. You can think of mapped memory as allocating a buffer to hold a file’s entire contents, and then reading the file into the buffer and (if the buffer is modified) writing the buffer back out to the file afterward. Linux handles the file reading and writing operations for you. There are uses for memory-mapped files other than interprocess communication. Some of these are discussed in Section 5.3.5, “Other Uses for mmap.”

5.3.1 Mapping an Ordinary File
To map an ordinary file to a process’s memory, use the mmap (“Memory MAPped,” pronounced “em-map”) call.The first argument is the address at which you would like Linux to map the file into your process’s address space; the value NULL allows Linux to choose an available start address.The second argument is the length of the map in bytes.The third argument specifies the protection on the mapped address range.The protection consists of a bitwise “or” of PROT_READ, PROT_WRITE, and PROT_EXEC, corresponding to read, write, and execution permission, respectively.The fourth argument is a flag value that specifies additional options.The fifth argument is a file descriptor opened to the file to be mapped.The last argument is the offset from the beginning of the file from which to start the map.You can map all or part of the file into memory by choosing the starting offset and length appropriately. The flag value is a bitwise “or” of these constraints: MAP_FIXED—If you specify this flag, Linux uses the address you request to map the file rather than treating it as a hint.This address must be page-aligned. MAP_PRIVATE—Writes to the memory range should not be written back to the attached file, but to a private copy of the file. No other process sees these writes. This mode may not be used with MAP_SHARED. n n

106 Chapter 5 Interprocess Communication

n

MAP_SHARED—Writes

are immediately reflected in the underlying file rather than buffering writes. Use this mode when using mapped memory for IPC.This mode may not be used with MAP_PRIVATE.

If the call succeeds, it returns a pointer to the beginning of the memory. On failure, it returns MAP_FAILED. When you’re finished with a memory mapping, release it by using munmap. Pass it the start address and length of the mapped memory region. Linux automatically unmaps mapped regions when a process terminates.

5.3.2 Example Programs
Let’s look at two programs to illustrate using memory-mapped regions to read and write to files.The first program, Listing 5.5, generates a random number and writes it to a memory-mapped file.The second program, Listing 5.6, reads the number, prints it, and replaces it in the memory-mapped file with double the value. Both take a command-line argument of the file to map.
Listing 5.5 (mmap-write.c) Write a Random Number to a Memory-Mapped File
#include #include #include #include #include #include #include #define FILE_LENGTH 0x100 /* Return a uniformly random number in the range [low,high]. */

int random_range (unsigned const low, unsigned const high) { unsigned const range = high - low + 1; return low + (int) (((double) range) * rand () / (RAND_MAX + 1.0)); } int main (int argc, char* const argv[]) { int fd; void* file_memory; /* Seed the random number generator. srand (time (NULL)); */

/* Prepare a file large enough to hold an unsigned integer. fd = open (argv[1], O_RDWR | O_CREAT, S_IRUSR | S_IWUSR); lseek (fd, FILE_LENGTH+1, SEEK_SET);

*/

5.3

Mapped Memory 107

write (fd, “”, 1); lseek (fd, 0, SEEK_SET); /* Create the memory mapping. */ file_memory = mmap (0, FILE_LENGTH, PROT_WRITE, MAP_SHARED, fd, 0); close (fd); /* Write a random integer to memory-mapped area. */ sprintf((char*) file_memory, “%d\n”, random_range (-100, 100)); /* Release the memory (unnecessary because the program exits). */ munmap (file_memory, FILE_LENGTH); return 0; }

The mmap-write program opens the file, creating it if it did not previously exist.The third argument to open specifies that the file is opened for reading and writing. Because we do not know the file’s length, we use lseek to ensure that the file is large enough to store an integer and then move back the file position to its beginning. The program maps the file and then closes the file descriptor because it’s no longer needed.The program then writes a random integer to the mapped memory, and thus the file, and unmaps the memory.The munmap call is unnecessary because Linux would automatically unmap the file when the program terminates.
Listing 5.6 (mmap-read.c) Read an Integer from a Memory-Mapped File, and Double It
#include #include #include #include #include #include #define FILE_LENGTH 0x100 int main (int argc, char* const argv[]) { int fd; void* file_memory; int integer; /* Open the file. */ fd = open (argv[1], O_RDWR, S_IRUSR | S_IWUSR); /* Create the memory mapping. */ file_memory = mmap (0, FILE_LENGTH, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0); close (fd);

continues

108 Chapter 5 Interprocess Communication

Listing 5.6 Continued
/* Read the integer, print it out, and double it. */ scanf (file_memory, “%d”, &integer); printf (“value: %d\n”, integer); sprintf ((char*) file_memory, “%d\n”, 2 * integer); /* Release the memory (unnecessary because the program exits). munmap (file_memory, FILE_LENGTH); return 0; }

*/

The mmap-read program reads the number out of the file and then writes the doubled value to the file. First, it opens the file and maps it for reading and writing. Because we can assume that the file is large enough to store an unsigned integer, we need not use lseek, as in the previous program.The program reads and parses the value out of memory using sscanf and then formats and writes the double value using sprintf. Here’s an example of running these example programs. It maps the file /tmp/integer-file.
% ./mmap-write /tmp/integer-file % cat /tmp/integer-file 42 % ./mmap-read /tmp/integer-file value: 42 % cat /tmp/integer-file 84

Observe that the text 42 was written to the disk file without ever calling write, and was read back in again without calling read. Note that these sample programs write and read the integer as a string (using sprintf and sscanf) for demonstration purposes only—there’s no need for the contents of a memory-mapped file to be text.You can store and retrieve arbitrary binary in a memory-mapped file.

5.3.3 Shared Access to a File
Different processes can communicate using memory-mapped regions associated with the same file. Specify the MAP_SHARED flag so that any writes to these regions are immediately transferred to the underlying file and made visible to other processes. If you don’t specify this flag, Linux may buffer writes before transferring them to the file. Alternatively, you can force Linux to incorporate buffered writes into the disk file by calling msync. Its first two parameters specify a memory-mapped region, as for munmap.The third parameter can take these flag values: MS_ASYNC—The update is scheduled but not necessarily run before the call returns. MS_SYNC—The update is immediate; the call to msync blocks until it’s done. MS_SYNC and MS_ASYNC may not both be used. n n

5.3

Mapped Memory 109

n

MS_INVALIDATE—All

other file mappings are invalidated so that they can see the mem_addr updated values. For example, to flush a shared file mapped at address bytes, call this: of length mem_length msync (mem_addr, mem_length, MS_SYNC | MS_INVALIDATE);

As with shared memory segments, users of memory-mapped regions must establish and follow a protocol to avoid race conditions. For example, a semaphore can be used to prevent more than one process from accessing the mapped memory at one time. Alternatively, you can use fcntl to place a read or write lock on the file, as described in Section 8.3, “fcntl: Locks and Other File Operations,” in Chapter 8.

5.3.4 Private Mappings
Specifying MAP_PRIVATE to mmap creates a copy-on-write region. Any write to the region is reflected only in this process’s memory; other processes that map the same file won’t see the changes. Instead of writing directly to a page shared by all processes, the process writes to a private copy of this page. All subsequent reading and writing by the process use this page.

5.3.5 Other Uses for mmap
The mmap call can be used for purposes other than interprocess communications. One common use is as a replacement for read and write. For example, rather than explicitly reading a file’s contents into memory, a program might map the file into memory and scan it using memory reads. For some programs, this is more convenient and may also run faster than explicit file I/O operations. One advanced and powerful technique used by some programs is to build data structures (ordinary struct instances, for example) in a memory-mapped file. On a subsequent invocation, the program maps that file back into memory, and the data structures are restored to their previous state. Note, though, that pointers in these data structures will be invalid unless they all point within the same mapped region of memory and unless care is taken to map the file back into the same address region that it occupied originally. Another handy technique is to map the special /dev/zero file into memory.That file, which is described in Section 6.5.2, “/dev/zero,” of Chapter 6, “Devices,” behaves as if it were an infinitely long file filled with 0 bytes. A program that needs a source of 0 bytes can mmap the file /dev/zero.Writes to /dev/zero are discarded, so the mapped memory may be used for any purpose. Custom memory allocators often map /dev/zero to obtain chunks of preinitialized memory.

110 Chapter 5 Interprocess Communication

5.4 Pipes
A pipe is a communication device that permits unidirectional communication. Data written to the “write end” of the pipe is read back from the “read end.” Pipes are serial devices; the data is always read from the pipe in the same order it was written. Typically, a pipe is used to communicate between two threads in a single process or between parent and child processes. In a shell, the symbol | creates a pipe. For example, this shell command causes the shell to produce two child processes, one for ls and one for less:
% ls | less

The shell also creates a pipe connecting the standard output of the ls subprocess with the standard input of the less process.The filenames listed by ls are sent to less in exactly the same order as if they were sent directly to the terminal. A pipe’s data capacity is limited. If the writer process writes faster than the reader process consumes the data, and if the pipe cannot store more data, the writer process blocks until more capacity becomes available. If the reader tries to read but no data is available, it blocks until data becomes available.Thus, the pipe automatically synchronizes the two processes.

5.4.1 Creating Pipes
To create a pipe, invoke the pipe command. Supply an integer array of size 2.The call to pipe stores the reading file descriptor in array position 0 and the writing file descriptor in position 1. For example, consider this code: int pipe_fds[2]; int read_fd; int write_fd; pipe (pipe_fds); read_fd = pipe_fds[0]; write_fd = pipe_fds[1];

Data written to the file descriptor

read_fd

can be read back from

write_fd.

5.4.2 Communication Between Parent and Child Processes
A call to pipe creates file descriptors, which are valid only within that process and its children. A process’s file descriptors cannot be passed to unrelated processes; however, when the process calls fork, file descriptors are copied to the new child process.Thus, pipes can connect only related processes. In the program in Listing 5.7, a fork spawns a child process.The child inherits the pipe file descriptors.The parent writes a string to the pipe, and the child reads it out. The sample program converts these file descriptors into FILE* streams using fdopen. Because we use streams rather than file descriptors, we can use the higher-level standard C library I/O functions such as printf and fgets.

5.4

Pipes 111

Listing 5.7 (pipe.c) Using a Pipe to Communicate with a Child Process
#include #include #include /* Write COUNT copies of MESSAGE to STREAM, pausing for a second between each. */ void writer (const char* message, int count, FILE* stream) { for (; count > 0; --count) { /* Write the message to the stream, and send it off immediately. fprintf (stream, “%s\n”, message); fflush (stream); /* Snooze a while. */ sleep (1); } } /* Read random strings from the stream as long as possible. */

*/

void reader (FILE* stream) { char buffer[1024]; /* Read until we hit the end of the stream. fgets reads until either a newline or the end-of-file. */ while (!feof (stream) && !ferror (stream) && fgets (buffer, sizeof (buffer), stream) != NULL) fputs (buffer, stdout); } int main () { int fds[2]; pid_t pid; /* Create a pipe. File descriptors for the two ends of the pipe are placed in fds. */ pipe (fds); /* Fork a child process. */ pid = fork (); if (pid == (pid_t) 0) { FILE* stream; /* This is the child process. Close our copy of the write end of the file descriptor. */ close (fds[1]); /* Convert the read file descriptor to a FILE object, and read from it. */ stream = fdopen (fds[0], “r”); reader (stream);

continues

112 Chapter 5 Interprocess Communication

Listing 5.7 Continued close (fds[0]); } else { /* This is the parent process. */ FILE* stream; /* Close our copy of the read end of the file descriptor. */ close (fds[0]); /* Convert the write file descriptor to a FILE object, and write to it. */ stream = fdopen (fds[1], “w”); writer (“Hello, world.”, 5, stream); close (fds[1]); } return 0; }

At the beginning of main, fds is declared to be an integer array with size 2.The pipe call creates a pipe and places the read and write file descriptors in that array.The program then forks a child process. After closing the read end of the pipe, the parent process starts writing strings to the pipe. After closing the write end of the pipe, the child reads strings from the pipe. Note that after writing in the writer function, the parent flushes the pipe by calling fflush. Otherwise, the string may not be sent through the pipe immediately. When you invoke the command ls | less, two forks occur: one for the ls child process and one for the less child process. Both of these processes inherit the pipe file descriptors so they can communicate using a pipe.To have unrelated processes communicate, use a FIFO instead, as discussed in Section 5.4.5, “FIFOs.”

5.4.3 Redirecting the Standard Input, Output, and Error Streams
Frequently, you’ll want to create a child process and set up one end of a pipe as its standard input or standard output. Using the dup2 call, you can equate one file descriptor with another. For example, to redirect a process’s standard input to a file descriptor fd, use this line: dup2 (fd, STDIN_FILENO);

The symbolic constant STDIN_FILENO represents the file descriptor for the standard input, which has the value 0.The call closes standard input and then reopens it as a duplicate of fd so that the two may be used interchangeably. Equated file descriptors share the same file position and the same set of file status flags.Thus, characters read from fd are not reread from standard input.

5.4

Pipes 113

The program in Listing 5.8 uses dup2 to send the output from a pipe to the sort command.2 After creating a pipe, the program forks.The parent process prints some strings to the pipe.The child process attaches the read file descriptor of the pipe to its standard input using dup2. It then executes the sort program.
Listing 5.8 (dup2.c) Redirect Output from a Pipe with dup2
#include #include #include #include

int main () { int fds[2]; pid_t pid; /* Create a pipe. File descriptors for the two ends of the pipe are placed in fds. */ pipe (fds); /* Fork a child process. */ pid = fork (); if (pid == (pid_t) 0) { /* This is the child process. Close our copy of the write end of the file descriptor. */ close (fds[1]); /* Connect the read end of the pipe to standard input. */ dup2 (fds[0], STDIN_FILENO); /* Replace the child process with the “sort” program. */ execlp (“sort”, “sort”, 0); } else { /* This is the parent process. */ FILE* stream; /* Close our copy of the read end of the file descriptor. */ close (fds[0]); /* Convert the write file descriptor to a FILE object, and write to it. */ stream = fdopen (fds[1], “w”); fprintf (stream, “This is a test.\n”); fprintf (stream, “Hello, world.\n”); fprintf (stream, “My dog has fleas.\n”); fprintf (stream, “This program is great.\n”); fprintf (stream, “One fish, two fish.\n”); fflush (stream); close (fds[1]); /* Wait for the child process to finish. */ waitpid (pid, NULL, 0); } return 0; }

2. sort reads lines of text from standard input, sorts them into alphabetical order, and prints them to standard output.

114 Chapter 5 Interprocess Communication

5.4.4

popen and pclose

A common use of pipes is to send data to or receive data from a program being run in a subprocess.The popen and pclose functions ease this paradigm by eliminating the need to invoke pipe, fork, dup2, exec, and fdopen. Compare Listing 5.9, which uses popen and pclose, to the previous example (Listing 5.8).
Listing 5.9 (popen.c) Example Using popen
#include #include int main () { FILE* stream = popen (“sort”, “w”); fprintf (stream, “This is a test.\n”); fprintf (stream, “Hello, world.\n”); fprintf (stream, “My dog has fleas.\n”); fprintf (stream, “This program is great.\n”); fprintf (stream, “One fish, two fish.\n”); return pclose (stream); }

The call to popen creates a child process executing the sort command, replacing calls to pipe, fork, dup2, and execlp.The second argument, “w”, indicates that this process wants to write to the child process.The return value from popen is one end of a pipe; the other end is connected to the child process’s standard input. After the writing finishes, pclose closes the child process’s stream, waits for the process to terminate, and returns its status value. The first argument to popen is executed as a shell command in a subprocess running /bin/sh.The shell searches the PATH environment variable in the usual way to find programs to execute. If the second argument is “r”, the function returns the child process’s standard output stream so that the parent can read the output. If the second argument is “w”, the function returns the child process’s standard input stream so that the parent can send data. If an error occurs, popen returns a null pointer. Call pclose to close a stream returned by popen. After closing the specified stream, pclose waits for the child process to terminate.

5.4.5

FIFOs

A first-in, first-out (FIFO) file is a pipe that has a name in the filesystem. Any process can open or close the FIFO; the processes on either end of the pipe need not be related to each other. FIFOs are also called named pipes.

5.4

Pipes 115

You can make a FIFO using the mkfifo command. Specify the path to the FIFO on the command line. For example, create a FIFO in /tmp/fifo by invoking this:
% mkfifo /tmp/fifo % ls -l /tmp/fifo prw-rw-rw1 samuel

users

0 Jan 16 14:04 /tmp/fifo

The first character of the output from ls is p, indicating that this file is actually a FIFO (named pipe). In one window, read from the FIFO by invoking the following:
% cat < /tmp/fifo

In a second window, write to the FIFO by invoking this:
% cat > /tmp/fifo

Then type in some lines of text. Each time you press Enter, the line of text is sent through the FIFO and appears in the first window. Close the FIFO by pressing Ctrl+D in the second window. Remove the FIFO with this line:
% rm /tmp/fifo

Creating a FIFO Create a FIFO programmatically using the mkfifo function.The first argument is the path at which to create the FIFO; the second parameter specifies the pipe’s owner, group, and world permissions, as discussed in Chapter 10, “Security,” Section 10.3, “File System Permissions.” Because a pipe must have a reader and a writer, the permissions must include both read and write permissions. If the pipe cannot be created (for instance, if a file with that name already exists), mkfifo returns –1. Include and if you call mkfifo. Accessing a FIFO Access a FIFO just like an ordinary file.To communicate through a FIFO, one program must open it for writing, and another program must open it for reading. Either low-level I/O functions (open, write, read, close, and so on, as listed in Appendix B, “Low-Level I/O”) or C library I/O functions (fopen, fprintf, fscanf, fclose, and so on) may be used. For example, to write a buffer of data to a FIFO using low-level I/O routines, you could use this code: int fd = open (fifo_path, O_WRONLY); write (fd, data, data_length); close (fd);

To read a string from the FIFO using C library I/O functions, you could use this code:
FILE* fifo = fopen (fifo_path, “r”); fscanf (fifo, “%s”, buffer); fclose (fifo);

116 Chapter 5 Interprocess Communication

A FIFO can have multiple readers or multiple writers. Bytes from each writer are written atomically up to a maximum size of PIPE_BUF (4KB on Linux). Chunks from simultaneous writers can be interleaved. Similar rules apply to simultaneous reads. Differences from Windows Named Pipes Pipes in the Win32 operating systems are very similar to Linux pipes. (Refer to the Win32 library documentation for technical details about these.) The main differences concern named pipes, which, for Win32, function more like sockets.Win32 named pipes can connect processes on separate computers connected via a network. On Linux, sockets are used for this purpose. Also,Win32 allows multiple reader-writer connections on a named pipe without interleaving data, and pipes can be used for two-way communication.3

5.5

Sockets

A socket is a bidirectional communication device that can be used to communicate with another process on the same machine or with a process running on other machines. Sockets are the only interprocess communication we’ll discuss in this chapter that permit communication between processes on different computers. Internet programs such as Telnet, rlogin, FTP, talk, and the World Wide Web use sockets. For example, you can obtain the WWW page from a Web server using the Telnet program because they both use sockets for network communications.4 To open a connection to a WWW server at www.codesourcery.com, use telnet www.codesourcery.com 80.The magic constant 80 specifies a connection to the Web server programming running www.codesourcery.com instead of some other process.Try typing GET / after the connection is established.This sends a message through the socket to the Web server, which replies by sending the home page’s HTML source and then closing the connection—for example:
% telnet www.codesourcery.com 80 Trying 206.168.99.1... Connected to merlin.codesourcery.com (206.168.99.1). Escape character is ‘^]’. GET / ...

3. Note that only Windows NT can create a named pipe;Windows 9x programs can form only client connections. 4. Usually, you’d use telnet to connect a Telnet server for remote logins. But you can also use telnet to connect to a server of a different kind and then type comments directly at it.

5.5

Sockets 117

5.5.1 Socket Concepts
When you create a socket, you must specify three parameters: communication style, namespace, and protocol. A communication style controls how the socket treats transmitted data and specifies the number of communication partners.When data is sent through a socket, it is packaged into chunks called packets.The communication style determines how these packets are handled and how they are addressed from the sender to the receiver. Connection styles guarantee delivery of all packets in the order they were sent. If packets are lost or reordered by problems in the network, the receiver automatically requests their retransmission from the sender. A connection-style socket is like a telephone call:The addresses of the sender and receiver are fixed at the beginning of the communication when the connection is established. n n

Datagram styles do not guarantee delivery or arrival order. Packets may be lost or reordered in transit due to network errors or other conditions. Each packet must be labeled with its destination and is not guaranteed to be delivered.The system guarantees only “best effort,” so packets may disappear or arrive in a different order than shipping. A datagram-style socket behaves more like postal mail.The sender specifies the receiver’s address for each individual message.

A socket namespace specifies how socket addresses are written. A socket address identifies one end of a socket connection. For example, socket addresses in the “local namespace” are ordinary filenames. In “Internet namespace,” a socket address is composed of the Internet address (also known as an Internet Protocol address or IP address) of a host attached to the network and a port number.The port number distinguishes among multiple sockets on the same host. A protocol specifies how data is transmitted. Some protocols are TCP/IP, the primary networking protocols used by the Internet; the AppleTalk network protocol; and the UNIX local communication protocol. Not all combinations of styles, namespaces, and protocols are supported.

5.5.2 System Calls
Sockets are more flexible than previously discussed communication techniques.These are the system calls involving sockets: socket—Creates a socket closes—Destroys a socket connect—Creates a connection between two sockets bind—Labels a server socket with an address listen—Configures a socket to accept conditions accept—Accepts a connection and creates a new socket for the connection Sockets are represented by file descriptors.

118 Chapter 5 Interprocess Communication

Creating and Destroying Sockets The socket and close functions create and destroy sockets, respectively.When you create a socket, specify the three socket choices: namespace, communication style, and protocol. For the namespace parameter, use constants beginning with PF_ (abbreviating “protocol families”). For example, PF_LOCAL or PF_UNIX specifies the local namespace, and PF_INET specifies Internet namespaces. For the communication style parameter, use constants beginning with SOCK_. Use SOCK_STREAM for a connection-style socket, or use SOCK_DGRAM for a datagram-style socket. The third parameter, the protocol, specifies the low-level mechanism to transmit and receive data. Each protocol is valid for a particular namespace-style combination. Because there is usually one best protocol for each such pair, specifying 0 is usually the correct protocol. If socket succeeds, it returns a file descriptor for the socket.You can read from or write to the socket using read, write, and so on, as with other file descriptors.When you are finished with a socket, call close to remove it. Calling connect To create a connection between two sockets, the client calls connect, specifying the address of a server socket to connect to. A client is the process initiating the connection, and a server is the process waiting to accept connections.The client calls connect to initiate a connection from a local socket to the server socket specified by the second argument.The third argument is the length, in bytes, of the address structure pointed to by the second argument. Socket address formats differ according to the socket namespace. Sending Information Any technique to write to a file descriptor can be used to write to a socket. See Appendix B for a discussion of Linux’s low-level I/O functions and some of the issues surrounding their use.The send function, which is specific to the socket file descriptors, provides an alternative to write with a few additional choices; see the man page for information.

5.5.3

Servers

A server’s life cycle consists of the creation of a connection-style socket, binding an address to its socket, placing a call to listen that enables connections to the socket, placing calls to accept incoming connections, and then closing the socket. Data isn’t read and written directly via the server socket; instead, each time a program accepts a new connection, Linux creates a separate socket to use in transferring data over that connection. In this section, we introduce bind, listen, and accept.

5.5

Sockets 119

An address must be bound to the server’s socket using bind if a client is to find it. Its first argument is the socket file descriptor.The second argument is a pointer to a socket address structure; the format of this depends on the socket’s address family.The third argument is the length of the address structure, in bytes.When an address is bound to a connection-style socket, it must invoke listen to indicate that it is a server. Its first argument is the socket file descriptor.The second argument specifies how many pending connections are queued. If the queue is full, additional connections will be rejected.This does not limit the total number of connections that a server can handle; it limits just the number of clients attempting to connect that have not yet been accepted. A server accepts a connection request from a client by invoking accept. The first argument is the socket file descriptor.The second argument points to a socket address structure, which is filled with the client socket’s address.The third argument is the length, in bytes, of the socket address structure.The server can use the client address to determine whether it really wants to communicate with the client.The call to accept creates a new socket for communicating with the client and returns the corresponding file descriptor.The original server socket continues to accept new client connections. To read data from a socket without removing it from the input queue, use recv. It takes the same arguments as read, plus an additional FLAGS argument. A flag of MSG_PEEK causes data to be read but not removed from the input queue.

5.5.4 Local Sockets
Sockets connecting processes on the same computer can use the local namespace represented by the synonyms PF_LOCAL and PF_UNIX.These are called local sockets or UNIX-domain sockets.Their socket addresses, specified by filenames, are used only when creating connections. The socket’s name is specified in struct sockaddr_un.You must set the sun_family field to AF_LOCAL, indicating that this is a local namespace.The sun_path field specifies the filename to use and may be, at most, 108 bytes long.The actual length of struct sockaddr_un should be computed using the SUN_LEN macro. Any filename can be used, but the process must have directory write permissions, which permit adding files to the directory.To connect to a socket, a process must have read permission for the file. Even though different computers may share the same filesystem, only processes running on the same computer can communicate with local namespace sockets. The only permissible protocol for the local namespace is 0. Because it resides in a file system, a local socket is listed as a file. For example, notice the initial s:
% ls -l /tmp/socket srwxrwx--x 1 user group 0 Nov 13 19:18 /tmp/socket

Call

unlink

to remove a local socket when you’re done with it.

120 Chapter 5 Interprocess Communication

5.5.5 An Example Using Local Namespace Sockets
We illustrate sockets with two programs.The server program, in Listing 5.10, creates a local namespace socket and listens for connections on it.When it receives a connection, it reads text messages from the connection and prints them until the connection closes. If one of these messages is “quit,” the server program removes the socket and ends.The socket-server program takes the path to the socket as its command-line argument.
Listing 5.10 (socket-server.c) Local Namespace Socket Server
#include #include #include #include #include #include

/* Read text from the socket and print it out. Continue until the socket closes. Return nonzero if the client sent a “quit” message, zero otherwise. */ int server (int client_socket) { while (1) { int length; char* text; /* First, read the length of the text message from the socket. read returns zero, the client closed the connection. */ if (read (client_socket, &length, sizeof (length)) == 0) return 0; /* Allocate a buffer to hold the text. */ text = (char*) malloc (length); /* Read the text itself, and print it. */ read (client_socket, text, length); printf (“%s\n”, text); /* Free the buffer. */ free (text); /* If the client sent the message “quit,” we’re all done. if (!strcmp (text, “quit”)) return 1; } } int main (int argc, char* const argv[]) { const char* const socket_name = argv[1]; If

*/

5.5

Sockets 121

int socket_fd; struct sockaddr_un name; int client_sent_quit_message; /* Create the socket. */ socket_fd = socket (PF_LOCAL, SOCK_STREAM, 0); /* Indicate that this is a server. */ name.sun_family = AF_LOCAL; strcpy (name.sun_path, socket_name); bind (socket_fd, &name, SUN_LEN (&name)); /* Listen for connections. */ listen (socket_fd, 5); /* Repeatedly accept connections, spinning off one server() to deal with each client. Continue until a client sends a “quit” message. do { struct sockaddr_un client_name; socklen_t client_name_len; int client_socket_fd;

*/

/* Accept a connection. */ client_socket_fd = accept (socket_fd, &client_name, &client_name_len); /* Handle the connection. */ client_sent_quit_message = server (client_socket_fd); /* Close our end of the connection. */ close (client_socket_fd); } while (!client_sent_quit_message); /* Remove the socket file. close (socket_fd); unlink (socket_name); return 0; } */

The client program, in Listing 5.11, connects to a local namespace socket and sends a message.The name path to the socket and the message are specified on the command line.
Listing 5.11 (socket-client.c) Local Namespace Socket Client
#include #include #include #include #include

continues

122 Chapter 5 Interprocess Communication

Listing 5.11 Continued
/* Write TEXT to the socket given by file descriptor SOCKET_FD. void write_text (int socket_fd, const char* text) { /* Write the number of bytes in the string, including NUL-termination. */ int length = strlen (text) + 1; write (socket_fd, &length, sizeof (length)); /* Write the string. */ write (socket_fd, text, length); } int main (int argc, char* const argv[]) { const char* const socket_name = argv[1]; const char* const message = argv[2]; int socket_fd; struct sockaddr_un name; /* Create the socket. */ socket_fd = socket (PF_LOCAL, SOCK_STREAM, 0); /* Store the server’s name in the socket address. */ name.sun_family = AF_LOCAL; strcpy (name.sun_path, socket_name); /* Connect the socket. */ connect (socket_fd, &name, SUN_LEN (&name)); /* Write the text on the command line to the socket. */ write_text (socket_fd, message); close (socket_fd); return 0; } */

Before the client sends the message text, it sends the length of that text by sending the bytes of the integer variable length. Likewise, the server reads the length of the text by reading from the socket into an integer variable.This allows the server to allocate an appropriately sized buffer to hold the message text before reading it from the socket. To try this example, start the server program in one window. Specify a path to a socket—for example, /tmp/socket.
% ./socket-server /tmp/socket

In another window, run the client a few times, specifying the same socket path plus messages to send to the client:
% ./socket-client /tmp/socket “Hello, world.” % ./socket-client /tmp/socket “This is a test.”

5.5

Sockets 123

The server program receives and prints these messages.To close the server, send the message “quit” from a client:
% ./socket-client /tmp/socket “quit”

The server program terminates.

5.5.6 Internet-Domain Sockets
UNIX-domain sockets can be used only for communication between two processes on the same computer. Internet-domain sockets, on the other hand, may be used to connect processes on different machines connected by a network. Sockets connecting processes through the Internet use the Internet namespace represented by PF_INET.The most common protocols are TCP/IP.The Internet Protocol (IP), a low-level protocol, moves packets through the Internet, splitting and rejoining the packets, if necessary. It guarantees only “best-effort” delivery, so packets may vanish or be reordered during transport. Every participating computer is specified using a unique IP number.The Transmission Control Protocol (TCP), layered on top of IP, provides reliable connection-ordered transport. It permits telephone-like connections to be established between computers and ensures that data is delivered reliably and in order.
DNS Names
Because it is easier to remember names than numbers, the Domain Name Service (DNS) associates names such as www.codesourcery.com with computers’ unique IP numbers. DNS is implemented by a worldwide hierarchy of name servers, but you don’t need to understand DNS protocols to use Internet host names in your programs.

Internet socket addresses contain two parts: a machine and a port number.This information is stored in a struct sockaddr_in variable. Set the sin_family field to AF_INET to indicate that this is an Internet namespace address.The sin_addr field stores the Internet address of the desired machine as a 32-bit integer IP number. A port number distinguishes a given machine’s different sockets. Because different machines store multibyte values in different byte orders, use htons to convert the port number to network byte order. See the man page for ip for more information. To convert human-readable hostnames, either numbers in standard dot notation (such as 10.0.0.1) or DNS names (such as www.codesourcery.com) into 32-bit IP numbers, you can use gethostbyname.This returns a pointer to the struct hostent structure; the h_addr field contains the host’s IP number. See the sample program in Listing 5.12. Listing 5.12 illustrates the use of Internet-domain sockets.The program obtains the home page from the Web server whose hostname is specified on the command line.

124 Chapter 5 Interprocess Communication

Listing 5.12 (socket-inet.c) Read from a WWW Server
#include #include #include #include #include #include #include

/* Print the contents of the home page for the server’s socket. Return an indication of success. */ void get_home_page (int socket_fd) { char buffer[10000]; ssize_t number_characters_read; /* Send the HTTP GET command for the home page. */ sprintf (buffer, “GET /\n”); write (socket_fd, buffer, strlen (buffer)); /* Read from the socket. The call to read may not return all the data at one time, so keep trying until we run out. */ while (1) { number_characters_read = read (socket_fd, buffer, 10000); if (number_characters_read == 0) return; /* Write the data to standard output. */ fwrite (buffer, sizeof (char), number_characters_read, stdout); } } int main (int argc, char* const argv[]) { int socket_fd; struct sockaddr_in name; struct hostent* hostinfo; /* Create the socket. */ socket_fd = socket (PF_INET, SOCK_STREAM, 0); /* Store the server’s name in the socket address. */ name.sin_family = AF_INET; /* Convert from strings to numbers. */ hostinfo = gethostbyname (argv[1]); if (hostinfo == NULL) return 1; else name.sin_addr = *((struct in_addr *) hostinfo->h_addr); /* Web servers use port 80. */ name.sin_port = htons (80);

5.5

Sockets 125

/* Connect to the Web server */ if (connect (socket_fd, &name, sizeof (struct sockaddr_in)) == -1) { perror (“connect”); return 1; } /* Retrieve the server’s home page. */ get_home_page (socket_fd); return 0; }

This program takes the hostname of the Web server on the command line (not a URL—that is, without the “http://”). It calls gethostbyname to translate the hostname into a numerical IP address and then connects a stream (TCP) socket to port 80 on that host.Web servers speak the Hypertext Transport Protocol (HTTP), so the program issues the HTTP GET command and the server responds by sending the text of the home page.
Standard Port Numbers
By convention, Web servers listen for connections on port 80. Most Internet network services are associated with a standard port number. For example, secure Web servers that use SSL listen for connections on port 443, and mail servers (which speak SMTP) use port 25. On GNU/Linux systems, the associations between protocol/service names and standard port numbers are listed in the file /etc/services. The first column is the protocol or service name. The second column lists the port number and the connection type: tcp for connection-oriented, or udp for datagram. If you implement custom network services using Internet-domain sockets, use port numbers greater than 1024.

For example, to retrieve the home page from the Web site invoke this:

www.codesourcery.com,

% ./socket-inet www.codesourcery.com ...

5.5.7 Socket Pairs
As we saw previously, the pipe function creates two file descriptors for the beginning and end of a pipe. Pipes are limited because the file descriptors must be used by related processes and because communication is unidirectional.The socketpair function creates two file descriptors for two connected sockets on the same computer. These file descriptors permit two-way communication between related processes.

126 Chapter 5 Interprocess Communication

Its first three parameters are the same as those of the socket call:They specify the domain, connection style, and protocol.The last parameter is a two-integer array, which is filled with the file descriptions of the two sockets, similar to pipe.When you call socketpair, you must specify PF_LOCAL as the domain.

II
Mastering Linux
6 7 8 9 10 11 Devices The /proc File System Linux System Calls Inline Assembly Code Security A Sample GNU/Linux Application

6
Devices

INUX, LIKE MOST OPERATING SYSTEMS, INTERACTS WITH HARDWARE devices via modularized software components called device drivers. A device driver hides the peculiarities of a hardware device’s communication protocols from the operating system and allows the system to interact with the device through a standardized interface. Under Linux, device drivers are part of the kernel and may be either linked statically into the kernel or loaded on demand as kernel modules. Device drivers run as part of the kernel and aren’t directly accessible to user processes. However, Linux provides a mechanism by which processes can communicate with a device driver—and through it with a hardware device—via file-like objects.These objects appear in the file system, and programs can open them, read from them, and write to them practically as if they were normal files. Using either Linux’s low-level I/O operations (see Appendix B, “Low-Level I/O”) or the standard C library’s I/O operations, your programs can communicate with hardware devices through these file-like objects. Linux also provides several file-like objects that communicate directly with the kernel rather than with device drivers.These aren’t linked to hardware devices; instead, they provide various kinds of specialized behavior that can be of use to application and system programs.

L

130 Chapter 6 Devices

Exercise Caution When Accessing Devices!
The techniques in this chapter provide direct access to device drivers running in the Linux kernel, and through them to hardware devices connected to the system. Use these techniques with care because misuse can cause impair or damage the GNU/Linux system. See especially the sidebar “Dangers of Block Devices.”

6.1 Device Types
Device files aren’t ordinary files—they do not represent regions of data on a diskbased file system. Instead, data read from or written to a device file is communicated to the corresponding device driver, and from there to the underlying device. Device files come in two flavors: A character device represents a hardware device that reads or writes a serial stream of data bytes. Serial and parallel ports, tape drives, terminal devices, and sound cards are examples of character devices. n n

A block device represents a hardware device that reads or writes data in fixed-size blocks. Unlike a character device, a block device provides random access to data stored on the device. A disk drive is an example of a block device.

Typical application programs will never use block devices.While a disk drive is represented as block devices, the contents of each disk partition typically contain a file system, and that file system is mounted into GNU/Linux’s root file system tree. Only the kernel code that implements the file system needs to access the block device directly; application programs access the disk’s contents through normal files and directories.

Dangers of Block Devices
Block devices provide direct access to disk drive data. Although most GNU/Linux systems are configured to prevent nonroot processes from accessing these devices directly, a root process can inflict severe damage by changing the contents of the disk. By writing to a disk block device, a program can modify or destroy file system control information and even a disk’s partition table and master boot record, thus rendering a drive or even the entire system unusable. Always access these devices with great care.

Applications sometimes make use of character devices, though.We’ll discuss several of them in the following sections.

6.2 Device Numbers
Linux identifies devices using two numbers: the major device number and the minor device number.The major device number specifies which driver the device corresponds to. The correspondence from major device numbers to drivers is fixed and part of the Linux kernel sources. Note that the same major device number may correspond to

6.3

Device Entries 131

two different drivers, one a character device and one a block device. Minor device numbers distinguish individual devices or components controlled by a single driver. The meaning of a minor device number depends on the device driver. For example, major device no. 3 corresponds to the primary IDE controller on the system. An IDE controller can have two devices (disk, tape, or CD-ROM drives) attached to it; the “master” device has minor device no. 0, and the “slave” device has minor device no. 64. Individual partitions on the master device (if the device supports partitions) are represented by minor device numbers 1, 2, 3, and so on. Individual partitions on the slave device are represented by minor device numbers 65, 66, 67, and so on. Major device numbers are listed in the Linux kernel sources documentation. On many GNU/Linux distributions, this documentation can be found in /usr/src/linux/Documentation/devices.txt.The special entry /proc/devices lists major device numbers corresponding to active device drivers currently loaded into the kernel. (See Chapter 7, “The /proc File System,” for more information about /proc file system entries.)

6.3 Device Entries
A device entry is in many ways the same as a regular file.You can move it using the mv command and delete it using the rm command. If you try to copy a device entry using cp, though, you’ll read bytes from the device (if the device supports reading) and write them to the destination file. If you try to overwrite a device entry, you’ll write bytes to the corresponding device instead. You can create a device entry in the file system using the mknod command (invoke man 1 mknod for the man page) or the mknod system call (invoke man 2 mknod for the man page). Creating a device entry in the file system doesn’t automatically imply that the corresponding device driver or hardware device is present or available; the device entry is merely a portal for communicating with the driver, if it’s there. Only superuser processes can create block and character devices using the mknod command or the mknod system call. To create a device using the mknod command, specify as the first argument the path at which the entry will appear in the file system. For the second argument, specify b for a block device or c for a character device. Provide the major and minor device numbers as the third and fourth arguments, respectively. For example, this command makes a character device entry named lp0 in the current directory.The device has major device no. 6 and minor device no. 0.These numbers correspond to the first parallel port on the Linux system.
% mknod ./lp0 c 6 0

132 Chapter 6 Devices

Remember that only superuser processes can create block and character devices, so you must be logged in as root to invoke this command successfully. The ls command displays device entries specially. If you invoke ls with the -l or -o options, the first character on each line of output specifies the type of the entry. Recall that - (a hyphen) designates a normal file, while d designates a directory. Similarly, b designates a block device, and c designates a character device. For the latter two, ls prints the major and minor device numbers where it would the size of an ordinary file. For example, we can display the block device that we just created:
% ls -l lp0 crw-r----1 root root 6, 0 Mar 7 17:03 lp0

In a program, you can determine whether a file system entry is a block or character device and then retrieve its device numbers using stat. See Section B.2, “stat,” in Appendix B, for instructions. To remove the entry, use rm.This doesn’t remove the device or device driver; it simply removes the device entry from the file system.
% rm ./lp0

6.3.1 The /dev Directory
By convention, a GNU/Linux system includes a directory /dev containing the full complement of character and block device entries for devices that Linux knows about. Entries in /dev have standardized names corresponding to major and minor device numbers. For example, the master device attached to the primary IDE controller, which has major and minor device numbers 3 and 0, has the standard name /dev/hda. If this device supports partitions, the first partition on it, which has minor device no. 1, has the standard name /dev/hda1.You can check that this is true on your system:
% ls -l /dev/hda /dev/hda1 brw-rw---1 root disk brw-rw---1 root disk 3, 3, 0 May 1 May 5 5 1998 /dev/hda 1998 /dev/hda1

Similarly, /dev has an entry for the parallel port character device that we used previously:
% ls -l /dev/lp0 crw-rw---1 root daemon 6, 0 May 5 1998 /dev/lp0

In most cases, you should not use mknod to create your own device entries. Use the entries in /dev instead. Non-superuser programs have no choice but to use preexisting device entries because they cannot create their own.Typically, only system administrators and developers working with specialized hardware devices will need to create device entries. Most GNU/Linux distributions include facilities to help system administrators create standard device entries with the correct names.

6.4

Hardware Devices 133

6.3.2 Accessing Devices by Opening Files
How do you use these devices? In the case of character devices, it can be quite simple: Open the device as if it were a normal file, and read from or write to it.You can even use normal file commands such as cat, or your shell’s redirection syntax, to send data to or from the device. For example, if you have a printer connected to your computer’s first parallel port, you can print files by sending them directly to /dev/lp0.1 To print the contents of document.txt, invoke the following:
% cat document.txt > /dev/lp0

You must have permission to write to the device entry for this to succeed; on many GNU/Linux systems, the permissions are set so that only root and the system’s printer daemon (lpd) can write to the file. Also, what comes out of your printer depends on how your printer interprets the contents of the data you send it. Some printers will print plain text files that are sent to them,2 while others will not. PostScript printers will render and print PostScript files that you send to them. In a program, sending data to a device is just as simple. For example, this code fragment uses low-level I/O functions to send the contents of a buffer to /dev/lp0. int fd = open (“/dev/lp0”, O_WRONLY); write (fd, buffer, buffer_length); close (fd);

6.4 Hardware Devices
Some common block devices are listed in Table 6.1. Device numbers for similar devices follow the obvious pattern (for instance, the second partition on the first SCSI drive is /dev/sda2). It’s occasionally useful to know which devices these device names correspond to when examining mounted file systems in /proc/mounts (see Section 7.5, “Drives, Mounts, and File Systems,” in Chapter 7, for more about this).
Table 6.1 Device First floppy drive Second floppy drive Primary IDE controller, master device Primary IDE controller, master device, first partition Partial Listing of Common Block Devices Name
/dev/fd0 /dev/fd1 /dev/hda /dev/hda1

Major 2 2 3 3

Minor 0 1 0 1

continues

1. Windows users will recognize that this device is similar to the magic Windows file LPR1. 2. Your printer may require explicit carriage return characters, ASCII code 14, at the end of each line, and may require a form feed character, ASCII code 12, at the end of each page.

134 Chapter 6 Devices

Table 6.1 Device

Continued Name
/dev/hdb /dev/hdb1

Major 3 3 22 22 8 8 8 8 11 11

Minor 64 65 0 64 0 1 16 17 0 1

Primary IDE controller, secondary device Primary IDE controller, secondary device, first partition Secondary IDE controller, master device Secondary IDE controller, secondary device First SCSI drive First SCSI drive, first partition Second SCSI disk Second SCSI disk, first partition First SCSI CD-ROM drive Second SCSI CD-ROM drive

/dev/hdc /dev/hdd /dev/sda /dev/sda1 /dev/sdb /dev/sdb1 /dev/scd0 /dev/scd1

Table 6.2 lists some common character devices.
Table 6.2 Device Parallel port 0 Parallel port 1 First serial port Second serial port IDE tape drive First SCSI tape drive Second SCSI tape drive System console First virtual terminal Second virtual terminal Process’s current terminal device Sound card Some Common Character Devices Name
/dev/lp0 /dev/lp1

Major or or
/dev/par0 /dev/par1

Minor 0 1 64 65 0 0 1 1 1 2 0 4

6 6 4 4 37 9 9 5 4 4 5 14

/dev/ttyS0 /dev/ttyS1 /dev/ht0 /dev/st0 /dev/st1 /dev/console /dev/tty1 /dev/tty2 /dev/tty /dev/audio

You can access certain hardware components through more than one character device; often, the different character devices provide different semantics. For example, when you use the IDE tape device /dev/ht0, Linux automatically rewinds the tape in the drive when you close the file descriptor.You can use the device /dev/nht0 to access the same tape drive, except that Linux will not automatically rewind the tape when you close the file descriptor.You sometimes might see programs using /dev/cua0 and similar devices; these are older interfaces to serial ports such as /dev/ttyS0.

6.4

Hardware Devices 135

Occasionally, you’ll want to write data directly to character devices—for example: A terminal program might access a modem directly through a serial port device. Data written to or read from the devices is transmitted via the modem to a remote computer. A tape backup program might write data directly to a tape device.The backup program could implement its own compression and error-checking format. A program can write directly to the first virtual terminal3 writing data to /dev/tty1. Terminal windows running in a graphical environment, or remote login terminal sessions, are not associated with virtual terminals; instead, they’re associated with pseudo-terminals. See Section 6.6,“PTYs,” for information about these. Sometimes a program needs to access the terminal device with which it is associated. For example, your program may need to prompt the user for a password. For security reasons, you might want to ignore redirection of standard input and output and always read the password from the terminal, no matter how the user invokes the command. One way to do this is to open /dev/tty, which always corresponds to the terminal device associated with the process that opens it. Write the prompt message to that device, and read the password from it. By ignoring standard input and output, this prevents the user from feeding your program a password from a file using shell syntax such as this: n n n n

% secure_program < my-password.txt

n

If you need to authenticate users in your program, you should learn about GNU/Linux’s PAM facility. See Section 10.5, “Authenticating Users,” in Chapter 10, “Security,” for more information. A program can play sounds through the system’s sound card by sending audio data to /dev/audio. Note that the audio data must be in Sun audio format (usually associated with the .au extension). For example, many GNU/Linux distributions come with the classic sound file /usr/share/sndconfig/sample.au. If your system includes this file, try playing it by invoking the following:
% cat /usr/share/sndconfig/sample.au > /dev/audio

If you’re planning on using sound in your program, though, you should investigate the various sound libraries and services available for GNU/Linux.The Gnome windowing environment uses the Enlightenment Sound Daemon (EsounD), at http://www.tux.org/~ricdude/EsounD.html. KDE uses aRts, at http://space.twc.de/~stefan/kde/arts-mcop-doc/. If you use one of these sound systems instead of writing directly to /dev/audio, your program will cooperate better with other programs that use the computer’s sound card.
3. On most GNU/Linux systems, you can switch to the first virtual terminal by pressing Ctrl+Alt+F1. Use Ctrl+Alt+F2 for the second virtual terminal, and so on.

136 Chapter 6 Devices

6.5 Special Devices
Linux also provides several character devices that don’t correspond to hardware devices.These entries all use the major device no. 1, which is associated with the Linux kernel’s memory device instead of a device driver.

6.5.1

/dev/null

The entry /dev/null, the null device, is very handy. It serves two purposes; you are probably familiar at least with the first one: Linux discards any data written to /dev/null. A common trick is to specify /dev/null as an output file in some context where the output is unwanted. For example, to run a command and discard its standard output (without printing it or writing it to a file), redirect standard output to /dev/null: n % verbose_command > /dev/null n Reading from /dev/null always results in an end-of-file. For instance, if you open a file descriptor to /dev/null using open and then attempt to read from the file descriptor, read will read no bytes and will return 0. If you copy from /dev/null to another file, the destination will be a zero-length file:
% cp /dev/null empty-file % ls -l empty-file -rw-rw---1 samuel samuel

0 Mar

8 00:27 empty-file

6.5.2

/dev/zero

The device entry /dev/zero behaves as if it were an infinitely long file filled with 0 bytes. As much data as you’d try to read from /dev/zero, Linux “generates” enough 0 bytes. To illustrate this, let’s run the hex dump program presented in Listing B.4 in Section B.1.4, “Reading Data,” of Appendix B.This program prints the contents of a file in hexadecimal form.
% ./hexdump /dev/zero 0x000000 : 00 00 00 00 0x000010 : 00 00 00 00 0x000020 : 00 00 00 00 0x000030 : 00 00 00 00 ... 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00

Hit Ctrl+C when you’re convinced that it will go on indefinitely. Memory mapping /dev/zero is an advanced technique for allocating memory. See Section 5.3.5, “Other Uses for mmap,” in Chapter 5, “Interprocess Communication,” for more information, and see the sidebar “Obtaining Page-Aligned Memory” in Section 8.9, “mprotect: Setting Memory Permissions,” in Chapter 8, “Linux System Calls,” for an example.

6.5

Special Devices 137

6.5.3

/dev/full

The entry /dev/full behaves as if it were a file on a file system that has no more room. A write to /dev/full fails and sets errno to ENOSPC, which ordinarily indicates that the written-to device is full. For example, you can try to write to /dev/full using the cp command:
% cp /etc/fstab /dev/full cp: /dev/full: No space left on device

The /dev/full entry is primarily useful to test how your program behaves if it runs out of disk space while writing to a file.

6.5.4 Random Number Devices
The special devices /dev/random and /dev/urandom provide access to the Linux kernel’s built-in random number–generation facility. Most software functions for generating random numbers, such as the rand function in the standard C library, actually generate pseudorandom numbers. Although these numbers satisfy some properties of random numbers, they are reproducible: If you start with the same seed value, you’ll obtain the same sequence of pseudorandom numbers every time.This behavior is inevitable because computers are intrinsically deterministic and predictable. For certain applications, though, this behavior is undesirable; for instance, it is sometimes possible to break a cryptographic algorithm if you can obtain the sequence of random numbers that it employs. To obtain better random numbers in computer programs requires an external source of randomness.The Linux kernel harnesses a particularly good source of randomness: you! By measuring the time delay between your input actions, such as keystrokes and mouse movements, Linux is capable of generating an unpredictable stream of high-quality random numbers.You can access this stream by reading from /dev/random and /dev/urandom.The data that you read is a stream of randomly generated bytes. The difference between the two devices exhibits itself when Linux exhausts its store of randomness. If you try to read a large number of bytes from /dev/random but don’t generate any input actions (you don’t type, move the mouse, or perform a similar action), Linux blocks the read operation. Only when you provide some randomness does Linux generate some more random bytes and return them to your program. For example, try displaying the contents of /dev/random using the od command.4 Each row of output shows 16 random bytes.

4.We use od here instead of the hexdump program presented in Listing B.4, even though they do pretty much the same thing, because hexdump terminates when it runs out of data, while od waits for more data to become available.The -t x1 option tells od to print file contents in hexadecimal.

138 Chapter 6 Devices

% od -t 0000000 0000020 0000040 0000060

x1 2c d3 b3 05

/dev/random 9c 7a db 2e 6d 1e a7 91 b0 8d 94 21 a3 02 cb 22

79 05 57 0a

3d 2d f3 bc

65 4d 90 c9

36 c3 61 45

c2 a6 dd dd

e3 de 26 a6

1b 54 ac 59

52 29 94 40

75 f4 c3 22

1e 46 b9 53

1a 04 3a d4

The number of lines of output that you see will vary—there may be quite a few—but the output will eventually pause when Linux exhausts its store of randomness. Now try moving your mouse or typing on the keyboard, and watch additional random numbers appear. For even better randomness, let your cat walk on the keyboard. A read from /dev/urandom, in contrast, will never block. If Linux runs out of randomness, it uses a cryptographic algorithm to generate pseudorandom bytes from the past sequence of random bytes. Although these bytes are random enough for many purposes, they don’t pass as many tests of randomness as those obtained from /dev/random. For instance, if you invoke the following, the random bytes will fly by forever, until you kill the program with Ctrl+C:
% od -t 0000000 0000020 0000040 ... x1 62 26 95 /dev/urandom 71 d6 3e af dd de 62 c0 42 78 bd 29 9c 69 49 3b 95 bc b9 6c 15 16 38 fd 7e 34 f0 ba ce c3 31 e5 2c 8d 8a dd f4 c4 3b 9b 44 2f 20 d1 54

Using random numbers from /dev/random in a program is easy, too. Listing 6.1 presents a function that generates a random number using bytes read from in /dev/random. Remember that /dev/random blocks a read until there is enough randomness available to satisfy it; you can use /dev/urandom instead if fast execution is more important and you can live with the potential lower quality of random numbers.
Listing 6.1 (random_number.c) Function to Generate a Random Number Using /dev/random
#include #include #include #include #include Obtain

/* Return a random integer between MIN and MAX, inclusive. randomness from /dev/random. */

int random_number (int min, int max) { /* Store a file descriptor opened to /dev/random in a static variable. That way, we don’t need to open the file every time this function is called. */ static int dev_random_fd = -1;

6.5

Special Devices 139

char* next_random_byte; int bytes_to_read; unsigned random_value; /* Make sure MAX is greater than MIN. assert (max > min); */

/* If this is the first time this function is called, open a file descriptor to /dev/random. */ if (dev_random_fd == -1) { dev_random_fd = open (“/dev/random”, O_RDONLY); assert (dev_random_fd != -1); } /* Read enough random bytes to fill an integer variable. */ next_random_byte = (char*) &random_value; bytes_to_read = sizeof (random_value); /* Loop until we’ve read enough bytes. Because /dev/random is filled from user-generated actions, the read may block and may only return a single random byte at a time. */ do { int bytes_read; bytes_read = read (dev_random_fd, next_random_byte, bytes_to_read); bytes_to_read -= bytes_read; next_random_byte += bytes_read; } while (bytes_to_read > 0); /* Compute a random number in the correct range. return min + (random_value % (max - min + 1)); } */

6.5.5 Loopback Devices
A loopback device enables you to simulate a block device using an ordinary disk file. Imagine a disk drive device for which data is written to and read from a file named disk-image rather than to and from the tracks and sectors of an actual physical disk drive or disk partition. (Of course, the file disk-image must reside on an actual disk, which must be larger than the simulated disk.) A loopback device enables you to use a file in this manner. Loopback devices are named /dev/loop0, /dev/loop1, and so on. Each can be used to simulate a single block device at one time. Note that only the superuser can set up a loopback device. A loopback device can be used in the same way as any other block device. In particular, you can construct a file system on the device and then mount that file system as you would mount the file system on an ordinary disk or partition. Such a file system, which resides in its entirety within an ordinary disk file, is called a virtual file system.

140 Chapter 6 Devices

To construct a virtual file system and mount it with a loopback device, follow these steps: 1. Create an empty file to hold the virtual file system.The size of the file will be the apparent size of the loopback device after it is mounted. One convenient way to construct a file of a fixed size is with the dd command. This command copies blocks (by default, 512 bytes each) from one file to another.The /dev/zero file is a convenient source of bytes to copy from. To construct a 10MB file named disk-image, invoke the following:
% dd if=/dev/zero of=/tmp/disk-image count=20480 20480+0 records in 20480+0 records out % ls -l /tmp/disk-image -rw-rw---1 root root 10485760 Mar 8 01:56 /tmp/disk-image

2. The file that you’ve just created is filled with 0 bytes. Before you mount it, you must construct a file system.This sets up the various control structures needed to organize and store files, and builds the root directory. You can build any type of file system you like in your disk image.To construct an ext2 file system (the type most commonly used for Linux disks), use the mke2fs command. Because it’s usually run on a block device, not an ordinary file, it asks for confirmation:
% mke2fs -q /tmp/disk-image mke2fs 1.18, 11-Nov-1999 for EXT2 FS 0.5b, 95/08/09 disk-image is not a block special device. Proceed anyway? (y,n) y

The -q option suppresses summary information about the newly created file system. Leave this option out if you’re curious about it. Now disk-image contains a brand-new file system, as if it were a freshly initialized 10MB disk drive.

6.5

Special Devices 141

3. Mount the file system using a loopback device.To do this, use the mount command, specifying the disk image file as the mount device. Also specify loop=loopback-device as a mount option, using the -o option to mount to tell mount which loopback device to use. For example, to mount our disk-image file system, invoke these commands. Remember, only the superuser may use a loopback device.The first command creates a directory, /tmp/virtual-fs, to use as the mount point for the virtual file system.
% mkdir /tmp/virtual-fs % mount -o loop=/dev/loop0 /tmp/disk-image /tmp/virtual-fs

Now your disk image is mounted as if it were an ordinary 10MB disk drive.
% df -h /tmp/virtual-fs Filesystem /tmp/disk-image Size 9.7M Used Avail Use% Mounted on 13k 9.2M 0% /tmp/virtual-fs

You can use it like any other disk:
% cd /tmp/virtual-fs % echo ‘Hello, world!’ > test.txt % ls -l total 13 drwxr-xr-x -rw-rw---% cat test.txt Hello, world! 2 root 1 root root root 12288 Mar 14 Mar 8 02:00 lost+found 8 02:12 test.txt

Note that

lost+found

is a directory that was automatically added by

mke2fs.5

5. If the file system is ever damaged, and some data is recovered but not associated with a file, it is placed in lost+found.

142 Chapter 6 Devices

When you’re done, unmount the virtual file system.
% cd /tmp % umount /tmp/virtual-fs

You can delete disk-image if you like, or you can mount it later to access the files on the virtual file system.You can also copy it to another computer and mount it there—the whole file system that you created on it will be intact. Instead of creating a file system from scratch, you can copy one directly from a device. For instance, you can create an image of the contents of a CD-ROM simply by copying it from the CD-ROM device. If you have an IDE CD-ROM drive, use the corresponding device name, such as /dev/hda, described previously. If you have a SCSI CD-ROM drive, the device name will be /dev/scd0 or similar.Your system may also have a symbolic link /dev/cdrom that points to the appropriate device. Consult your /etc/fstab file to determine what device corresponds to your computer’s CD-ROM drive. Simply copy that device to a file.The resulting file will be a complete disk image of the file system on the CD-ROM in the drive—for example:
% cp /dev/cdrom /tmp/cdrom-image

This may take several minutes, depending on the CD-ROM you’re copying and the speed of your drive.The resulting image file will be quite large—as large as the contents of the CD-ROM. Now you can mount this CD-ROM image without having the original CD-ROM in the drive. For example, to mount it on /mnt/cdrom, use this line:
% mount -o loop=/dev/loop0 /tmp/cdrom-image /mnt/cdrom

Because the image is on a hard disk drive, it’ll perform much faster than the actual CD-ROM disk. Note that most CD-ROMs use the file system type iso9660.

6.6

PTYs

If you run the mount command with no command-line arguments, which displays the file systems mounted on your system, you’ll notice a line that looks something like this: none on /dev/pts type devpts (rw,gid=5,mode=620)

This indicates that a special type of file system, devpts, is mounted at /dev/pts.This file system, which isn’t associated with any hardware device, is a “magic” file system that is created by the Linux kernel. It’s similar to the /proc file system; see Chapter 7 for more information about how this works.

6.6

PTYs 143

Like the /dev directory, /dev/pts contains entries corresponding to devices. But unlike /dev, which is an ordinary directory, /dev/pts is a special directory that is created dynamically by the Linux kernel.The contents of the directory vary with time and reflect the state of the running system. The entries in /dev/pts correspond to pseudo-terminals (or pseudo-TTYs, or PTYs). Linux creates a PTY for every new terminal window you open and displays a corresponding entry in /dev/pts.The PTY device acts like a terminal device—it accepts input from the keyboard and displays text output from the programs that run in it. PTYs are numbered, and the PTY number is the name of the corresponding entry in /dev/pts. You can display the terminal device associated with a process using the ps command. Specify tty as one of the fields of a custom format with the -o option.To display the process ID,TTY, and command line of each process sharing the same terminal, invoke ps -o pid,tty,cmd.

6.6.1 A PTY Demonstration
For example, you can determine the PTY associated with a given terminal window by invoking in the window this command:
% ps -o pid,tty,cmd PID TT CMD 28832 pts/4 bash 29287 pts/4 ps -o pid,tty,cmd

This particular terminal window is running in PTY 4. The PTY has a corresponding entry in /dev/pts:
% ls -l /dev/pts/4 crw--w---1 samuel tty 136, 4 Mar 8 02:56 /dev/pts/4

Note that it is a character device, and its owner is the owner of the process for which it was created. You can read from or write to the PTY device. If you read from it, you’ll hijack keyboard input that would otherwise be sent to the program running in the PTY. If you write to it, the data will appear in that window. Try opening a new terminal window, and determine its PTY number by invoking ps -o pid,tty,cmd. From another window, write some text to the PTY device. For example, if the new terminal window’s PTY number is 7, invoke this command from another window:
% echo ‘Hello, other window!’ > /dev/pts/7

The output appears in the new terminal window. If you close the new terminal window, the entry 7 in /dev/pts disappears.

144 Chapter 6 Devices

If you invoke ps to determine the TTY from a text-mode virtual terminal (press Ctrl+Alt+F1 to switch to the first virtual terminal, for instance), you’ll see that it’s running in an ordinary terminal device instead of a PTY:
% ps -o pid,tty,cmd PID TT CMD 29325 tty1 -bash 29353 tty1 ps -o pid,tty,cmd

6.7 ioctl
The ioctl system call is an all-purpose interface for controlling hardware devices.The first argument to ioctl is a file descriptor, which should be opened to the device that you want to control.The second argument is a request code that indicates the operation that you want to perform.Various request codes are available for different devices. Depending on the request code, there may be additional arguments supplying data to ioctl. Many of the available requests codes for various devices are listed in the ioctl_list man page. Using ioctl generally requires a detailed understanding of the device driver corresponding to the hardware device that you want to control. Most of these are quite specialized and are beyond the scope of this book. However, we’ll present one example to give you a taste of how ioctl is used.
Listing 6.2 (cdrom-eject.c) Eject a CD-ROM
#include #include #include #include #include #include

int main (int argc, char* argv[]) { /* Open a file descriptor to the device specified on the command line. int fd = open (argv[1], O_RDONLY); /* Eject the CD-ROM. */ ioctl (fd, CDROMEJECT); /* Close the file descriptor. */ close (fd); return 0; }

*/

6.7

ioctl 145

Listing 6.2 presents a short program that ejects the disk in a CD-ROM drive (if the drive supports this). It takes a single command-line argument, the CD-ROM drive device. It opens a file descriptor to the device and invokes ioctl with the request code CDROMEJECT.This request, defined in the header , instructs the device to eject the disk. For example, if your system has an IDE CD-ROM drive connected as the master device on the secondary IDE controller, the corresponding device is /dev/hdc.To eject the disk from the drive, invoke this line:
% ./cdrom-eject /dev/hdc

7
The /proc File System

T

RY INVOKING THE mount COMMAND WITHOUT ARGUMENTS—this displays the file systems currently mounted on your GNU/Linux computer.You’ll see one line that looks like this: none on /proc type proc (rw)

This is the special /proc file system. Notice that the first field, none, indicates that this file system isn’t associated with a hardware device such as a disk drive. Instead, /proc is a window into the running Linux kernel. Files in the /proc file system don’t correspond to actual files on a physical device. Instead, they are magic objects that behave like files but provide access to parameters, data structures, and statistics in the kernel. The “contents” of these files are not always fixed blocks of data, as ordinary file contents are. Instead, they are generated on the fly by the Linux kernel when you read from the file.You can also change the configuration of the running kernel by writing to certain files in the /proc file system. Let’s look at an example:
% ls -l /proc/version -r--r--r-1 root root 0 Jan 17 18:09 /proc/version

Note that the file size is zero; because the file’s contents are generated by the kernel, the concept of file size is not applicable. Also, if you try this command yourself, you’ll notice that the modification time on the file is the current time.

148 Chapter 7 The /proc File System

What’s in this file? The contents of /proc/version consist of a string describing the Linux kernel version number. It contains the version information that would be obtained by the uname system call, described in Chapter 8,“Linux System Calls,” in Section 8.15,“uname,” plus additional information such as the version of the compiler that was used to compile the kernel.You can read from /proc/version like you would any other file. For instance, an easy way to display its contents is with the cat command.
% cat /proc/version Linux version 2.2.14-5.0 (root@porky.devel.redhat.com) (gcc version egcs-2.91. 66 19990314/Linux (egcs-1.1.2 release)) #1 Tue Mar 7 21:07:39 EST 2000

The various entries in the /proc file system are described extensively in the page (Section 5).To view it, invoke this command:
% man 5 proc

proc

man

In this chapter, we’ll describe some of the features of the /proc file system that are most likely to be useful to application programmers, and we’ll give examples of using them. Some of the features of /proc are handy for debugging, too. If you’re interested in exactly how /proc works, take a look at the source code in the Linux kernel sources, under /usr/src/linux/fs/proc/.

7.1 Extracting Information from /proc
Most of the entries in /proc provide information formatted to be readable by humans, but the formats are simple enough to be easily parsed. For example, /proc/cpuinfo contains information about the system CPU (or CPUs, for a multiprocessor machine). The output is a table of values, one per line, with a description of the value and a colon preceding each value. For example, the output might look like this:
% cat /proc/cpuinfo processor : 0 vendor_id : GenuineIntel cpu family : 6 model : 5 model name : Pentium II (Deschutes) stepping : 2 cpu MHz : 400.913520 cache size : 512 KB fdiv_bug : no hlt_bug : no sep_bug : no f00f_bug : no coma_bug : no fpu : yes fpu_exception : yes cpuid level : 2 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr bogomips : 399.77

7.1

Extracting Information from /proc 149

We’ll describe the interpretation of some of these fields in Section 7.3.1, “CPU Information.” A simple way to extract a value from this output is to read the file into a buffer and parse it in memory using sscanf. Listing 7.1 shows an example of this.The program includes the function get_cpu_clock_speed that reads from /proc/cpuinfo into memory and extracts the first CPU’s clock speed.
Listing 7.1 (clock-speed.c) Extract CPU Clock Speed from /proc/cpuinfo
#include #include /* Returns the clock speed of the system’s CPU in MHz, as reported by /proc/cpuinfo. On a multiprocessor machine, returns the speed of the first CPU. On error returns zero. */ float get_cpu_clock_speed () { FILE* fp; char buffer[1024]; size_t bytes_read; char* match; float clock_speed; /* Read the entire contents of /proc/cpuinfo into the buffer. fp = fopen (“/proc/cpuinfo”, “r”); bytes_read = fread (buffer, 1, sizeof (buffer), fp); fclose (fp); /* Bail if read failed or if buffer isn’t big enough. */ if (bytes_read == 0 || bytes_read == sizeof (buffer)) return 0; /* NUL-terminate the text. */ buffer[bytes_read] = ‘\0’; /* Locate the line that starts with “cpu MHz”. */ match = strstr (buffer, “cpu MHz”); if (match == NULL) return 0; /* Parse the line to extract the clock speed. */ sscanf (match, “cpu MHz : %f”, &clock_speed); return clock_speed; } */

int main () { printf (“CPU clock speed: %4.0f MHz\n”, get_cpu_clock_speed ()); return 0; }

150 Chapter 7 The /proc File System

Be aware, however, that the names, semantics, and output formats of entries in the /proc file system might change in new Linux kernel revisions. If you use them in a program, you should make sure that the program’s behavior degrades gracefully if the /proc entry is missing or is formatted unexpectedly.

7.2 Process Entries
The /proc file system contains a directory entry for each process running on the GNU/Linux system.The name of each directory is the process ID of the corresponding process.1 These directories appear and disappear dynamically as processes start and terminate on the system. Each directory contains several entries providing access to information about the running process. From these process directories the /proc file system gets its name. Each process directory contains these entries: cmdline contains the argument list for the process.The cmdline entry is described in Section 7.2.2, “Process Argument List.” cwd is a symbolic link that points to the current working directory of the process (as set, for instance, with the chdir call). environ contains the process’s environment.The environ entry is described in Section 7.2.3, “Process Environment.” exe is a symbolic link that points to the executable image running in the process.The exe entry is described in Section 7.2.4, “Process Executable.” fd is a subdirectory that contains entries for the file descriptors opened by the process.These are described in Section 7.2.5, “Process File Descriptors.” maps displays information about files mapped into the process’s address. See Chapter 5, “Interprocess Communication,” Section 5.3, “Mapped Memory,” for details of how memory-mapped files work. For each mapped file, maps displays the range of addresses in the process’s address space into which the file is mapped, the permissions on these addresses, the name of the file, and other information. The maps table for each process displays the executable running in the process, any loaded shared libraries, and other files that the process has mapped in. root is a symbolic link to the root directory for this process. Usually, this is a symbolic link to /, the system root directory.The root directory for a process can be changed using the chroot call or the chroot command.2 n n n n n n n

1. On some UNIX systems, the process IDs are padded with zeros. On GNU/Linux, they are not. 2.The chroot call and command are outside the scope of this book. See the chroot man page in Section 1 for information about the command (invoke man 1 chroot), or the chroot man page in Section 2 (invoke man 2 chroot) for information about the call.

7.2

Process Entries 151

n

n

n

n

contains lots of status and statistical information about the process.These are the same data as presented in the status entry, but in raw numerical format, all on a single line.The format is difficult to read but might be more suitable for parsing by programs. If you want to use the stat entry in your programs, see the proc man page, which describes its contents, by invoking man 5 proc. statm contains information about the memory used by the process.The statm entry is described in Section 7.2.6, “Process Memory Statistics.” status contains lots of status and statistical information about the process, formatted to be comprehensible by humans. Section 7.2.7, “Process Statistics,” contains a description of the status entry. The cpu entry appears only on SMP Linux kernels. It contains a breakdown of process time (user and system) by CPU. stat Note that for security reasons, the permissions of some entries are set so that only the user who owns the process (or the superuser) can access them.

7.2.1

/proc/self

One additional entry in the /proc file system makes it easy for a program to use /proc to find information about its own process.The entry /proc/self is a symbolic link to the /proc directory corresponding to the current process.The destination of the /proc/self link depends on which process looks at it: Each process sees its own process directory as the target of the link. For example, the program in Listing 7.2 reads the target of the /proc/self link to determine its process ID. (We’re doing it this way for illustrative purposes only; calling the getpid function, described in Chapter 3, “Processes,” in Section 3.1.1, “Process IDs,” is a much easier way to do the same thing.) This program uses the readlink system call, described in Section 8.11, “readlink: Reading Symbolic Links,” to extract the target of the symbolic link.
Listing 7.2 (get-pid.c) Obtain the Process ID from /proc/self
#include #include #include /* Returns the process ID of the calling processes, as determined from the /proc/self symlink. */ pid_t get_pid_from_proc_self () { char target[32]; int pid; /* Read the target of the symbolic link. */ readlink (“/proc/self”, target, sizeof (target));

continues

152 Chapter 7 The /proc File System

Listing 7.2 Continued
/* The target is a directory named for the process ID. sscanf (target, “%d”, &pid); return (pid_t) pid; } int main () { printf (“/proc/self reports process id %d\n”, (int) get_pid_from_proc_self ()); printf (“getpid() reports process id %d\n”, (int) getpid ()); return 0; } */

7.2.2 Process Argument List
The cmdline entry contains the process argument list (see Chapter 2,“Writing Good GNU/Linux Software,” Section 2.1.1,“The Argument List”).The arguments are presented as a single character string, with arguments separated by NULs. Most string functions expect that the entire character string is terminated with a single NUL and will not handle NULs embedded within strings, so you’ll have to handle the contents specially.
NUL vs. NULL
NUL is the character with integer value 0. It’s different from NULL, which is a pointer with value 0. In C, a character string is usually terminated with a NUL character. For instance, the character string “Hello, world!” occupies 14 bytes because there is an implicit NUL after the exclamation point indicating the end of the string. NULL, on the other hand, is a pointer value that you can be sure will never correspond to a real memory address in your program. In C and C++, NUL is expressed as the character constant ‘\0’, or (char) 0. The definition of NULL differs among operating systems; on Linux, it is defined as ((void*)0) in C and simply 0 in C++.

In Section 2.1.1, we presented a program in Listing 2.1 that printed out its own argument list. Using the cmdline entries in the /proc file system, we can implement a program that prints the argument of another process. Listing 7.3 is such a program; it prints the argument list of the process with the specified process ID. Because there may be several NULs in the contents of cmdline rather than a single one at the end, we can’t determine the length of the string with strlen (which simply counts the number of characters until it encounters a NUL). Instead, we determine the length of cmdline from read, which returns the number of bytes that were read.

7.2

Process Entries 153

Listing 7.3 (print-arg-list.c) Print the Argument List of a Running Process
#include #include #include #include #include #include

/* Prints the argument list, one argument to a line, of the process given by PID. */ void print_process_arg_list (pid_t pid) { int fd; char filename[24]; char arg_list[1024]; size_t length; char* next_arg; /* Generate the name of the cmdline file for the process. */ snprintf (filename, sizeof (filename), “/proc/%d/cmdline”, (int) pid); /* Read the contents of the file. */ fd = open (filename, O_RDONLY); length = read (fd, arg_list, sizeof (arg_list)); close (fd); /* read does not NUL-terminate the buffer, so do it here. */ arg_list[length] = ‘\0’; /* Loop over arguments. Arguments are separated by NULs. */ next_arg = arg_list; while (next_arg < arg_list + length) { /* Print the argument. Each is NUL-terminated, so just treat it like an ordinary string. */ printf (“%s\n”, next_arg); /* Advance to the next argument. Since each argument is NUL-terminated, strlen counts the length of the next argument, not the entire argument list. */ next_arg += strlen (next_arg) + 1; } } int main (int argc, char* argv[]) { pid_t pid = (pid_t) atoi (argv[1]); print_process_arg_list (pid); return 0; }

154 Chapter 7 The /proc File System

For example, suppose that process 372 is the system logger daemon, syslogd.
% ps 372 PID TTY STAT TIME COMMAND 372 ? S 0:00 syslogd -m 0 % ./print-arg-list 372 syslogd -m 0

In this case, syslogd was invoked with the arguments

-m 0.

7.2.3 Process Environment
The environ entry contains a process’s environment (see Section 2.1.6, “The Environment”). As with cmdline, the individual environment variables are separated by NULs.The format of each element is the same as that used in the environ variable, namely VARIABLE=value. Listing 7.4 presents a generalization of the program in Listing 2.3 in Section 2.1.6. This version takes a process ID number on its command line and prints the environment for that process by reading it from /proc.
Listing 7.4 (print-environment.c) Display the Environment of a Process
#include #include #include #include #include #include

/* Prints the environment, one environment variable to a line, of the process given by PID. */ void print_process_environment (pid_t pid) { int fd; char filename[24]; char environment[8192]; size_t length; char* next_var; /* Generate the name of the environ file for the process. */ snprintf (filename, sizeof (filename), “/proc/%d/environ”, (int) pid); /* Read the contents of the file. */ fd = open (filename, O_RDONLY); length = read (fd, environment, sizeof (environment)); close (fd); /* read does not NUL-terminate the buffer, so do it here. */ environment[length] = ‘\0’;

7.2

Process Entries 155

/* Loop over variables. Variables are separated by NULs. */ next_var = environment; while (next_var < environment + length) { /* Print the variable. Each is NUL-terminated, so just treat it like an ordinary string. */ printf (“%s\n”, next_var); /* Advance to the next variable. Since each variable is NUL-terminated, strlen counts the length of the next variable, not the entire variable list. */ next_var += strlen (next_var) + 1; } } int main (int argc, char* argv[]) { pid_t pid = (pid_t) atoi (argv[1]); print_process_environment (pid); return 0; }

7.2.4 Process Executable
The exe entry points to the executable file being run in a process. In Section 2.1.1, we explained that typically the program executable name is passed as the first element of the argument list. Note, though, that this is purely conventional; a program may be invoked with any argument list. Using the exe entry in the /proc file system is a more reliable way to determine which executable is running. One useful technique is to extract the path containing the executable from the /proc file system. For many programs, auxiliary files are installed in directories with known paths relative to the main program executable, so it’s necessary to determine where that executable actually is.The function get_executable_path in Listing 7.5 determines the path of the executable running in the calling process by examining the symbolic link /proc/self/exe.
Listing 7.5 (get-exe-path.c) Get the Path of the Currently Running Program Executable
#include #include #include #include

/* Finds the path containing the currently running program executable. The path is placed into BUFFER, which is of length LEN. Returns the number of characters in the path, or -1 on error. */

continues

156 Chapter 7 The /proc File System

Listing 7.5 Continued size_t get_executable_path (char* buffer, size_t len) { char* path_end; /* Read the target of /proc/self/exe. */ if (readlink (“/proc/self/exe”, buffer, len) /dev/pts/4 64 Jan 30 01:02 1 -> /dev/pts/4 64 Jan 30 01:02 2 -> /dev/pts/4

(There may be other lines of output corresponding to other open file descriptors as well.) Recall that we mentioned in Section 2.1.4, “Standard I/O,” that file descriptors 0, 1, and 2 are initialized to standard input, output, and error, respectively.Thus, by writing to /proc/1261/fd/1, you can write to the device attached to stdout for the shell process—in this case, the pseudo TTY in the first window. In the second window, try writing a message to that file:
% echo “Hello, world.” >> /proc/1261/fd/1

The text appears in the first window. File descriptors besides standard input, output, and error appear in the fd subdirectory, too. Listing 7.6 presents a program that simply opens a file descriptor to a file specified on the command line and then loops forever.
Listing 7.6 (open-and-spin.c) Open a File for Reading
#include #include #include #include #include

int main (int argc, char* argv[]) { const char* const filename = argv[1]; int fd = open (filename, O_RDONLY); printf (“in process %d, file descriptor %d is open to %s\n”, (int) getpid (), (int) fd, filename); while (1); return 0; }

Try running it in one window:
% ./open-and-spin /etc/fstab in process 2570, file descriptor 3 is open to /etc/fstab

In another window, take a look at the /proc.
% ls -l /proc/2570/fd total 0 lrwx-----1 samuel

fd

subdirectory corresponding to this process in

samuel

64 Jan 30 01:30 0 -> /dev/pts/2

158 Chapter 7 The /proc File System

lrwx-----lrwx-----lr-x------

1 samuel 1 samuel 1 samuel

samuel samuel samuel

64 Jan 30 01:30 1 -> /dev/pts/2 64 Jan 30 01:30 2 -> /dev/pts/2 64 Jan 30 01:30 3 -> /etc/fstab

Notice the entry for file descriptor 3, linked to the file /etc/fstab opened on this descriptor. File descriptors can be opened on sockets or pipes, too (see Chapter 5 for more information about these). In such a case, the target of the symbolic link corresponding to the file descriptor will state “socket” or “pipe” instead of pointing to an ordinary file or device.

7.2.6

Process Memory Statistics

The statm entry contains a list of seven numbers, separated by spaces. Each number is a count of the number of pages of memory used by the process in a particular category.The categories, in the order the numbers appear, are listed here: The total process size The size of the process resident in physical memory The memory shared with other processes—that is, memory mapped both by this process and at least one other (such as shared libraries or untouched copyon-write pages) The text size of the process—that is, the size of loaded executable code The size of shared libraries mapped into this process The memory used by this process for its stack The number of dirty pages—that is, pages of memory that have been modified by the program n n n n n n n

7.2.7 Process Statistics
The status entry contains a variety of information about the process, formatted for comprehension by humans. Among this information is the process ID and parent process ID, the real and effective user and group IDs, memory usage, and bit masks specifying which signals are caught, ignored, and blocked.

7.3 Hardware Information
Several of the other entries in the /proc file system provide access to information about the system hardware. Although these are typically of interest to system configurators and administrators, the information may occasionally be of use to application programmers as well.We’ll present some of the more useful entries here.

7.3

Hardware Information 159

7.3.1

CPU Information

As shown previously, /proc/cpuinfo contains information about the CPU or CPUs running the GNU/Linux system.The Processor field lists the processor number; this is 0 for single-processor systems.The Vendor, CPU Family, Model, and Stepping fields enable you to determine the exact model and revision of the CPU. More useful, the Flags field shows which CPU flags are set, which indicates the features available in this CPU. For example, “mmx” indicates the availability of the extended MMX instructions.3 Most of the information returned from /proc/cpuinfo is derived from the cpuid x86 assembly instruction.This instruction is the low-level mechanism by which a program obtains information about the CPU. For a greater understanding of the output of /proc/cpuinfo, see the documentation of the cpuid instruction in Intel’s IA-32 Intel Architecture Software Developer’s Manual,Volume 2: Instruction Set Reference.This manual is available from http://developer.intel.com/design. The last element, bogomips, is a Linux-specific value. It is a measurement of the processor’s speed spinning in a tight loop and is therefore a rather poor indicator of overall processor speed.

7.3.2 Device Information
The /proc/devices file lists major device numbers for character and block devices available to the system. See Chapter 6, “Devices,” for information about types of devices and device numbers.

7.3.3

PCI Bus Information

The /proc/pci file lists a summary of devices attached to the PCI bus or buses.These are actual PCI expansion cards and may also include devices built into the system’s motherboard, plus AGP graphics cards.The listing includes the device type; the device and vendor ID; a device name, if available; information about the features offered by the device; and information about the PCI resources used by the device.

7.3.4 Serial Port Information
The /proc/tty/driver/serial file lists configuration information and statistics about serial ports. Serial ports are numbered from 0.4 Configuration information about serial ports can also be obtained, as well as modified, using the setserial command. However, /proc/tty/driver/serial displays additional statistics about each serial port’s interrupt counts.
3. See the IA-32 Intel Architecture Software Developer’s Manual for documentation about MMX instructions, and see Chapter 9, “Inline Assembly Code,” in this book for information on how to use these and other special assembly instructions in GNU/Linux programs. 4. Note that under DOS and Windows, serial ports are numbered from 1, so COM1 corresponds to serial port number 0 under Linux.

160 Chapter 7 The /proc File System

For example, this line from /proc/tty/driver/serial might describe serial port 1 (which would be COM2 under Windows):
1: uart:16550A port:2F8 irq:3 baud:9600 tx:11 rx:0

This indicates that the serial port is run by a 16550A-type UART, uses I/O port 0x2f8 and IRQ 3 for communication, and runs at 9,600 baud.The serial port has seen 11 transmit interrupts and 0 receive interrupts. See Section 6.4, “Hardware Devices,” for information about serial devices.

7.4 Kernel Information
Many of the entries in /proc provide access to information about the running kernel’s configuration and state. Some of these entries are at the top level of /proc; others are under /proc/sys/kernel.

7.4.1

Version Information

The file /proc/version contains a long string describing the kernel’s release number and build version. It also includes information about how the kernel was built: the user who compiled it, the machine on which it was compiled, the date it was compiled, and the compiler release that was used—for example:
% cat /proc/version Linux version 2.2.14-5.0 (root@porky.devel.redhat.com) (gcc version egcs-2.91.66 19990314/Linux (egcs-1.1.2 release)) #1 Tue Mar 7 21:07:39 EST 2000

This indicates that the system is running a 2.2.14 release of the Linux kernel, which was compiled with EGCS release 1.1.2. (EGCS, the Experimental GNU Compiler System, was a precursor to the current GCC project.) The most important items in this output, the OS name and kernel version and revision, are available in separate /proc entries as well.These are /proc/sys/ kernel/ostype, /proc/sys/kernel/osrelease, and /proc/sys/kernel/version, respectively.
% cat /proc/sys/kernel/ostype Linux % cat /proc/sys/kernel/osrelease 2.2.14-5.0 % cat /proc/sys/kernel/version #1 Tue Mar 7 21:07:39 EST 2000

7.4.2 Hostname and Domain Name
The /proc/sys/kernel/hostname and /proc/sys/kernel/domainname entries contain the computer’s hostname and domain name, respectively.This information is the same as that returned by the uname system call, described in Section 8.15.

7.5

Drives, Mounts, and File Systems 161

7.4.3 Memory Usage
The /proc/meminfo entry contains information about the system’s memory usage. Information is presented both for physical memory and for swap space.The first three lines present memory totals, in bytes; subsequent lines summarize this information in kilobytes—for example:
% cat /proc/meminfo total: used: free: shared: buffers: cached: Mem: 529694720 519610368 10084352 82612224 10977280 82108416 Swap: 271392768 44003328 227389440 MemTotal: 517280 kB MemFree: 9848 kB MemShared: 80676 kB Buffers: 10720 kB Cached: 80184 kB BigTotal: 0 kB BigFree: 0 kB SwapTotal: 265032 kB SwapFree: 222060 kB

This shows 512MB physical memory, of which about 9MB is free, and 258MB of swap space, of which 216MB is free. In the row corresponding to physical memory, three other values are presented: The Shared column displays total shared memory currently allocated on the system (see Section 5.1, “Shared Memory”). The Buffers column displays the memory allocated by Linux for block device buffers.These buffers are used by device drivers to hold blocks of data being read from and written to disk. The Cached column displays the memory allocated by Linux to the page cache. This memory is used to cache accesses to mapped files. n n n

You can use the

free

command to display the same memory information.

7.5 Drives, Mounts, and File Systems
The /proc file system also contains information about the disk drives present in the system and the file systems mounted from them.

7.5.1 File Systems
The /proc/filesystems entry displays the file system types known to the kernel. Note that this list isn’t very useful because it is not complete: File systems can be loaded and unloaded dynamically as kernel modules.The contents of /proc/filesystems list only file system types that either are statically linked into the kernel or are currently loaded. Other file system types may be available on the system as modules but might not be loaded yet.

162 Chapter 7 The /proc File System

7.5.2

Drives and Partitions

The /proc file system includes information about devices connected to both IDE controllers and SCSI controllers (if the system includes them). On typical systems, the /proc/ide subdirectory may contain either or both of two subdirectories, ide0 and ide1, corresponding to the primary and secondary IDE controllers on the system.5 These contain further subdirectories corresponding to physical devices attached to the controllers.The controller or device directories may be absent if Linux has not recognized any connected devices.The full paths corresponding to the four possible IDE devices are listed in Table 7.1.
Table 7.1 Controller Primary Primary Secondary Secondary Full Paths Corresponding to the Four Possible IDE Devices Device Master Slave Master Slave Subdirectory
/proc/ide/ide0/hda/ /proc/ide/ide0/hdb/ /proc/ide/ide1/hdc/ /proc/ide/ide1/hdd/

See Section 6.4, “Hardware Devices,” for more information about IDE device names. Each IDE device directory contains several entries providing access to identification and configuration information for the device. A few of the most useful are listed here: model contains the device’s model identification string. media contains the device’s media type. Possible values are disk, cdrom, tape, floppy, and UNKNOWN. capacity contains the device’s capacity, in 512-byte blocks. Note that for CDROM devices, the value will be 231 –1, not the capacity of the disk in the drive. Note that the value in capacity represents the capacity of the entire physical disk; the capacity of file systems contained in partitions of the disk will be smaller. n n n

For example, these commands show how to determine the media type and device identification for the master device on the secondary IDE controller. In this case, it turns out to be a Toshiba CD-ROM drive.
% cat /proc/ide/ide1/hdc/media cdrom % cat /proc/ide/ide1/hdc/model TOSHIBA CD-ROM XM-6702B

5. If properly configured, the Linux kernel can support additional IDE controllers.These are numbered sequentially from ide2.

7.5

Drives, Mounts, and File Systems 163

If SCSI devices are present in the system, /proc/scsi/scsi contains a summary of their identification values. For example, the contents might look like this:
% cat /proc/scsi/scsi Attached devices: Host: scsi0 Channel: 00 Id: 00 Lun: 00 Vendor: QUANTUM Model: ATLAS_V__9_WLS Type: Direct-Access Host: scsi0 Channel: 00 Id: 04 Lun: 00 Vendor: QUANTUM Model: QM39100TD-SW Type: Direct-Access

Rev: 0230 ANSI SCSI revision: 03 Rev: N491 ANSI SCSI revision: 02

This computer contains one single-channel SCSI controller (designated “scsi0”), to which two Quantum disk drives are connected, with SCSI device IDs 0 and 4. The /proc/partitions entry displays the partitions of recognized disk devices. For each partition, the output includes the major and minor device number, the number of 1024-byte blocks, and the device name corresponding to that partition. The /proc/sys/dev/cdrom/info entry displays miscellaneous information about the capabilities of CD-ROM drives.The fields are self-explanatory:
% cat /proc/sys/dev/cdrom/info CD-ROM information, Id: cdrom.c 2.56 1999/09/09 drive name: hdc drive speed: 48 drive # of slots: 0 Can close tray: 1 Can open tray: 1 Can lock tray: 1 Can change speed: 1 Can select disk: 0 Can read multisession: Can read MCN: 1 Reports media changed: Can play audio: 1

1 1

7.5.3

Mounts

The /proc/mounts file provides a summary of mounted file systems. Each line corresponds to a single mount descriptor and lists the mounted device, the mount point, and other information. Note that /proc/mounts contains the same information as the ordinary file /etc/mtab, which is automatically updated by the mount command. These are the elements of a mount descriptor: The first element on the line is the mounted device (see Chapter 6). For special file systems such as the /proc file system, this is none. The second element is the mount point, the place in the root file system at which the file system contents appear. For the root file system itself, the mount point is listed as /. For swap drives, the mount point is listed as swap. n n

164 Chapter 7 The /proc File System

n

n

The third element is the file system type. Currently, most GNU/Linux systems use the ext2 file system for disk drives, but DOS or Windows drives may be mounted with other file system types, such as fat or vfat. Most CD-ROMs contain an iso9660 file system. See the man page for the mount command for a list of file system types. The fourth element lists mount flags.These are options that were specified when the mount was added. See the man page for the mount command for an explanation of flags for the various file system types.

In

last two elements are always 0 and have no meaning. See the man page for fstab for details about the format of mount descriptors.6 GNU/Linux includes functions to help you parse mount descriptors; see the man page for the getmntent function for information on using these.

/proc/mounts, the

7.5.4 Locks
Section 8.3, “fcntl: Locks and Other File Operations,” describes how to use the fcntl system call to manipulate read and write locks on files.The /proc/locks entry describes all the file locks currently outstanding in the system. Each row in the output corresponds to one lock. For locks created with fcntl, the first two entries on the line are POSIX ADVISORY. The third is WRITE or READ, depending on the lock type.The next number is the process ID of the process holding the lock.The following three numbers, separated by colons, are the major and minor device numbers of the device on which the file resides and the inode number, which locates the file in the file system.The remainder of the line lists values internal to the kernel that are not of general utility. Turning the contents of /proc/locks into useful information takes some detective work.You can watch /proc/locks in action, for instance, by running the program in Listing 8.2 to create a write lock on the file /tmp/test-file.
% touch /tmp/test-file % ./lock-file /tmp/test-file file /tmp/test-file opening /tmp/test-file locking locked; hit enter to unlock...

In another window, look at the contents of

/proc/locks.

% cat /proc/locks 1: POSIX ADVISORY WRITE 5467 08:05:181288 0 2147483647 d1b5f740 00000000 dfea7d40 00000000 00000000

6.The

/etc/fstab

file lists the static mount configuration of the GNU/Linux system.

7.6

System Statistics 165

There may be other lines of output, too, corresponding to locks held by other programs. In this case, 5467 is the process ID of the lock-file program. Use ps to figure out what this process is running.
% ps 5467 PID TTY 5467 pts/28 STAT S TIME COMMAND 0:00 ./lock-file /tmp/test-file

The locked file, /tmp/test-file, resides on the device that has major and minor device numbers 8 and 5, respectively.These numbers happen to correspond to /dev/sda5.
% df /tmp Filesystem 1k-blocks /dev/sda5 8459764 % ls -l /dev/sda5 brw-rw---1 root disk Used Available Use% Mounted on 5094292 2935736 63% / 8, 5 May 5 1998 /dev/sda5

The file

/tmp/test-file

itself is at inode 181,288 on that device.

% ls --inode /tmp/test-file 181288 /tmp/test-file

See Section 6.2, “Device Numbers,” for more information about device numbers.

7.6 System Statistics
Two entries in /proc contain useful system statistics.The /proc/loadavg file contains information about the system load.The first three numbers represent the number of active tasks on the system—processes that are actually running—averaged over the last 1, 5, and 15 minutes.The next entry shows the instantaneous current number of runnable tasks—processes that are currently scheduled to run rather than being blocked in a system call—and the total number of processes on the system.The final entry is the process ID of the process that most recently ran. The /proc/uptime file contains the length of time since the system was booted, as well as the amount of time since then that the system has been idle. Both are given as floating-point values, in seconds.
% cat /proc/uptime 3248936.18 3072330.49

The program in Listing 7.7 extracts the uptime and idle time from the system and displays them in friendly units.
Listing 7.7 (print-uptime.c) Print the System Uptime and Idle Time
#include /* Summarize a duration of time to standard output. TIME is the amount of time, in seconds, and LABEL is a short descriptive label. void print_time (char* label, long time) {

*/

continues

166 Chapter 7 The /proc File System

Listing 7.7 Continued
/* Conversion constants. */ const long minute = 60; const long hour = minute * 60; const long day = hour * 24; /* Produce output. */ printf (“%s: %ld days, %ld:%02ld:%02ld\n”, label, time / day, (time % day) / hour, (time % hour) / minute, time % minute); } int main () { FILE* fp; double uptime, idle_time; /* Read the system uptime and accumulated idle time from /proc/uptime. fp = fopen (“/proc/uptime”, “r”); fscanf (fp, “%lf %lf\n”, &uptime, &idle_time); fclose (fp); /* Summarize it. */ print_time (“uptime “, (long) uptime); print_time (“idle time”, (long) idle_time); return 0; }

*/

The uptime command and the sysinfo system call (see Section 8.14, “sysinfo: Obtaining System Statistics”) also can obtain the system’s uptime.The uptime command also displays the load averages found in /proc/loadavg.

8
Linux System Calls

O FAR, WE’VE PRESENTED A VARIETY OF FUNCTIONS that your program can invoke to perform system-related functions, such as parsing command-line options, manipulating processes, and mapping memory. If you look under the hood, you’ll find that these functions fall into two categories, based on how they are implemented. A library function is an ordinary function that resides in a library external to your program. Most of the library functions we’ve presented so far are in the standard C library, libc. For example, getopt_long and mkstemp are functions provided in the C library. A call to a library function is just like any other function call.The arguments are placed in processor registers or onto the stack, and execution is transferred to the start of the function’s code, which typically resides in a loaded shared library. A system call is implemented in the Linux kernel.When a program makes a system call, the arguments are packaged up and handed to the kernel, which takes over execution of the program until the call completes. A system call isn’t an ordinary function call, and a special procedure is required to transfer control to the kernel. However, the GNU C library (the implementation of the standard C library provided with GNU/Linux systems) wraps Linux system calls with functions so that you can call them easily. Low-level I/O functions such as open and read are examples of system calls on Linux. n n

S

168 Chapter 8 Linux System Calls

The set of Linux system calls forms the most basic interface between programs and the Linux kernel. Each call presents a basic operation or capability. Some system calls are very powerful and can exert great influence on the system. For instance, some system calls enable you to shut down the Linux system or to allocate system resources and prevent other users from accessing them.These calls have the restriction that only processes running with superuser privilege (programs run by the root account) can invoke them.These calls fail if invoked by a nonsuperuser process. Note that a library function may invoke one or more other library functions or system calls as part of its implementation. Linux currently provides about 200 different system calls. A listing of system calls for your version of the Linux kernel is in /usr/include/asm/unistd.h. Some of these are for internal use by the system, and others are used only in implementing specialized library functions. In this chapter, we’ll present a selection of system calls that are likely to be the most useful to application and system programmers. Most of these system calls are declared in .

8.1 Using strace
Before we start discussing system calls, it will be useful to present a command with which you can learn about and debug system calls.The strace command traces the execution of another program, listing any system calls the program makes and any signals it receives. To watch the system calls and signals in a program, simply invoke strace, followed by the program and its command-line arguments. For example, to watch the system calls that are invoked by the hostname1 command, use this command:
% strace hostname

This produces a couple screens of output. Each line corresponds to a single system call. For each call, the system call’s name is listed, followed by its arguments (or abbreviated arguments, if they are very long) and its return value.Where possible, strace conveniently displays symbolic names instead of numerical values for arguments and return values, and it displays the fields of structures passed by a pointer into the system call. Note that strace does not show ordinary function calls. In the output from strace hostname, the first line shows the execve system call that invokes the hostname program:2 execve(“/bin/hostname”, [“hostname”], [/* 49 vars */]) = 0

1. hostname invoked without any flags simply prints out the computer’s hostname to standard output. 2. In Linux, the exec family of functions is implemented via the execve system call.

8.2

access: Testing File Permissions 169

The first argument is the name of the program to run; the second is its argument list, consisting of only a single element; and the third is its environment list, which strace omits for brevity.The next 30 or so lines are part of the mechanism that loads the standard C library from a shared library file. Toward the end are system calls that actually help do the program’s work.The uname system call is used to obtain the system’s hostname from the kernel, uname({sys=”Linux”, node=”myhostname”, ...}) = 0

Observe that strace helpfully labels the fields (sys and node) of the structure argument.This structure is filled in by the system call—Linux sets the sys field to the operating system name and the node field to the system’s hostname.The uname call is discussed further in Section 8.15, “uname.” Finally, the write system call produces output. Recall that file descriptor 1 corresponds to standard output.The third argument is the number of characters to write, and the return value is the number of characters that were actually written. write(1, “myhostname\n”, 11) = 11

This may appear garbled when you run strace because the output from the hostname program itself is mixed in with the output from strace. If the program you’re tracing produces lots of output, it is sometimes more convenient to redirect the output from strace into a file. Use the option -o filename to do this. Understanding all the output from strace requires detailed familiarity with the design of the Linux kernel and execution environment. Much of this is of limited interest to application programmers. However, some understanding is useful for debugging tricky problems or understanding how other programs work.

8.2 access: Testing File Permissions
The access system call determines whether the calling process has access permission to a file. It can check any combination of read, write, and execute permission, and it can also check for a file’s existence. The access call takes two arguments.The first is the path to the file to check.The second is a bitwise or of R_OK, W_OK, and X_OK, corresponding to read, write, and execute permission.The return value is 0 if the process has all the specified permissions. If the file exists but the calling process does not have the specified permissions, access returns –1 and sets errno to EACCES (or EROFS, if write permission was requested for a file on a read-only file system). If the second argument is F_OK, access simply checks for the file’s existence. If the file exists, the return value is 0; if not, the return value is –1 and errno is set to ENOENT. Note that errno may instead be set to EACCES if a directory in the file path is inaccessible.

170 Chapter 8 Linux System Calls

The program shown in Listing 8.1 uses access to check for a file’s existence and to determine read and write permissions. Specify the name of the file to check on the command line.
Listing 8.1 (check-access.c) Check File Access Permissions
#include #include #include int main (int argc, char* argv[]) { char* path = argv[1]; int rval; /* Check file existence. */ rval = access (path, F_OK); if (rval == 0) printf (“%s exists\n”, path); else { if (errno == ENOENT) printf (“%s does not exist\n”, path); else if (errno == EACCES) printf (“%s is not accessible\n”, path); return 0; } /* Check read access. */ rval = access (path, R_OK); if (rval == 0) printf (“%s is readable\n”, path); else printf (“%s is not readable (access denied)\n”, path); /* Check write access. */ rval = access (path, W_OK); if (rval == 0) printf (“%s is writable\n”, path); else if (errno == EACCES) printf (“%s is not writable (access denied)\n”, path); else if (errno == EROFS) printf (“%s is not writable (read-only filesystem)\n”, path); return 0; }

For example, to check access permissions for a file named invoke it like this:

README

on a CD-ROM,

% ./check-access /mnt/cdrom/README /mnt/cdrom/README exists /mnt/cdrom/README is readable /mnt/cdrom/README is not writable (read-only filesystem)

8.3

fcntl: Locks and Other File Operations 171

8.3 fcntl: Locks and Other File Operations
The fcntl system call is the access point for several advanced operations on file descriptors.The first argument to fcntl is an open file descriptor, and the second is a value that indicates which operation is to be performed. For some operations, fcntl takes an additional argument.We’ll describe here one of the most useful fcntl operations, file locking. See the fcntl man page for information about the others. The fcntl system call allows a program to place a read lock or a write lock on a file, somewhat analogous to the mutex locks discussed in Chapter 5, “Interprocess Communication.” A read lock is placed on a readable file descriptor, and a write lock is placed on a writable file descriptor. More than one process may hold a read lock on the same file at the same time, but only one process may hold a write lock, and the same file may not be both locked for read and locked for write. Note that placing a lock does not actually prevent other processes from opening the file, reading from it, or writing to it, unless they acquire locks with fcntl as well. To place a lock on a file, first create and zero out a struct flock variable. Set the l_type field of the structure to F_RDLCK for a read lock or F_WRLCK for a write lock. Then call fcntl, passing a file descriptor to the file, the F_SETLCKW operation code, and a pointer to the struct flock variable. If another process holds a lock that prevents a new lock from being acquired, fcntl blocks until that lock is released. The program in Listing 8.2 opens a file for writing whose name is provided on the command line, and then places a write lock on it.The program waits for the user to hit Enter and then unlocks and closes the file.
Listing 8.2 (lock-file.c) Create a Write Lock with fcntl
#include #include #include #include

int main (int argc, char* argv[]) { char* file = argv[1]; int fd; struct flock lock; printf (“opening %s\n”, file); /* Open a file descriptor to the file. fd = open (file, O_WRONLY); printf (“locking\n”); /* Initialize the flock structure. */ memset (&lock, 0, sizeof(lock)); lock.l_type = F_WRLCK; /* Place a write lock on the file. */ fcntl (fd, F_SETLKW, &lock);

*/

continues

172 Chapter 8 Linux System Calls

Listing 8.2 Continued printf (“locked; hit Enter to unlock... “); /* Wait for the user to hit Enter. */ getchar (); printf (“unlocking\n”); /* Release the lock. */ lock.l_type = F_UNLCK; fcntl (fd, F_SETLKW, &lock); close (fd); return 0; }

Compile and run the program on a test file—say, /tmp/test-file—like this:
% cc -o lock-file lock-file.c % touch /tmp/test-file % ./lock-file /tmp/test-file opening /tmp/test-file locking locked; hit Enter to unlock...

Now, in another window, try running it again on the same file.
% ./lock-file /tmp/test-file opening /tmp/test-file locking

Note that the second instance is blocked while attempting to lock the file. Go back to the first window and press Enter: unlocking The program running in the second window immediately acquires the lock. If you prefer fcntl not to block if the call cannot get the lock you requested, use F_SETLK instead of F_SETLKW. If the lock cannot be acquired, fcntl returns –1 immediately. Linux provides another implementation of file locking with the flock call.The fcntl version has a major advantage: It works with files on NFS3 file systems (as long as the NFS server is reasonably recent and correctly configured). So, if you have access to two machines that both mount the same file system via NFS, you can repeat the previous example using two different machines. Run lock-file on one machine, specifying a file on an NFS file system, and then run it again on another machine, specifying the same file. NFS wakes up the second program when the lock is released by the first program.
3. Network File System (NFS) is a common network file sharing technology, comparable to Windows’ shares and network drives.

8.4

fsync and fdatasync: Flushing Disk Buffers 173

8.4 fsync and fdatasync: Flushing Disk Buffers
On most operating systems, when you write to a file, the data is not immediately written to disk. Instead, the operating system caches the written data in a memory buffer, to reduce the number of required disk writes and improve program responsiveness.When the buffer fills or some other condition occurs (for instance, enough time elapses), the system writes the cached data to disk all at one time. Linux provides caching of this type as well. Normally, this is a great boon to performance. However, this behavior can make programs that depend on the integrity of disk-based records unreliable. If the system goes down suddenly—for instance, due to a kernel crash or power outage—any data written by a program that is in the memory cache but has not yet been written to disk is lost. For example, suppose that you are writing a transaction-processing program that keeps a journal file.The journal file contains records of all transactions that have been processed so that if a system failure occurs, the state of the transaction data can be reconstructed. It is obviously important to preserve the integrity of the journal file— whenever a transaction is processed, its journal entry should be sent to the disk drive immediately. To help you implement this, Linux provides the fsync system call. It takes one argument, a writable file descriptor, and flushes to disk any data written to this file. The fsync call doesn’t return until the data has physically been written. The function in Listing 8.3 illustrates the use of fsync. It writes a single-line entry to a journal file.
Listing 8.3 (write_journal_entry.c) Write and Sync a Journal Entry
#include #include #include #include #include

const char* journal_filename = “journal.log”; void write_journal_entry (char* entry) { int fd = open (journal_filename, O_WRONLY | O_CREAT | O_APPEND, 0660); write (fd, entry, strlen (entry)); write (fd, “\n”, 1); fsync (fd); close (fd); }

Another system call, fdatasync does the same thing. However, although fsync guarantees that the file’s modification time will be updated, fdatasync does not; it guarantees only that the file’s data will be written.This means that in principal, fdatasync can execute faster than fsync because it needs to force only one disk write instead of two.

174 Chapter 8 Linux System Calls

However, in current versions of Linux, these two system calls actually do the same thing, both updating the file’s modification time. The fsync system call enables you to force a buffer write explicitly.You can also open a file for synchronous I/O, which causes all writes to be committed to disk immediately.To do this, specify the O_SYNC flag when opening the file with the open call.

8.5 getrlimit and setrlimit: Resource Limits
The getrlimit and setrlimit system calls allow a process to read and set limits on the system resources that it can consume.You may be familiar with the ulimit shell command, which enables you to restrict the resource usage of programs you run;4 these system calls allow a program to do this programmatically. For each resource there are two limits, the hard limit and the soft limit.The soft limit may never exceed the hard limit, and only processes with superuser privilege may change the hard limit.Typically, an application program will reduce the soft limit to place a throttle on the resources it uses. Both getrlimit and setrlimit take as arguments a code specifying the resource limit type and a pointer to a structrlimit variable.The getrlimit call fills the fields of this structure, while the setrlimit call changes the limit based on its contents.The rlimit structure has two fields: rlim_cur is the soft limit, and rlim_max is the hard limit. Some of the most useful resource limits that may be changed are listed here, with their codes: RLIMIT_CPU—The maximum CPU time, in seconds, used by a program.This is the amount of time that the program is actually executing on the CPU, which is not necessarily the same as wall-clock time. If the program exceeds this time limit, it is terminated with a SIGXCPU signal. RLIMIT_DATA—The maximum amount of memory that a program can allocate for its data. Additional allocation beyond this limit will fail. RLIMIT_NPROC—The maximum number of child processes that can be running for this user. If the process calls fork and too many processes belonging to this user are running on the system, fork fails. RLIMIT_NOFILE—The maximum number of file descriptors that the process may have open at one time. n n n n

See the setrlimit man page for a full list of system resources. The program in Listing 8.4 illustrates setting the limit on CPU time consumed by a program. It sets a 1-second CPU time limit and then spins in an infinite loop. Linux kills the process soon afterward, when it exceeds 1 second of CPU time.
4. See the man page for your shell for more information about ulimit. 8.6

getrusage: Process Statistics 175

Listing 8.4 (limit-cpu.c) CPU Time Limit Demonstration
#include #include #include int main () { struct rlimit rl; /* Obtain the current limits. */ getrlimit (RLIMIT_CPU, &rl); /* Set a CPU limit of 1 second. */ rl.rlim_cur = 1; setrlimit (RLIMIT_CPU, &rl); /* Do busy work. */ while (1); return 0; }

When the program is terminated by interpreting the signal:
% ./limit_cpu CPU time limit exceeded

SIGXCPU, the

shell helpfully prints out a message

8.6 getrusage: Process Statistics
The getrusage system call retrieves process statistics from the kernel. It can be used to obtain statistics either for the current process by passing RUSAGE_SELF as the first argument, or for all terminated child processes that were forked by this process and its children by passing RUSAGE_CHILDREN.The second argument to rusage is a pointer to a struct rusage variable, which is filled with the statistics. A few of the more interesting fields in struct rusage are listed here: ru_utime—A struct timeval field containing the amount of user time, in seconds, that the process has used. User time is CPU time spent executing the user program, rather than in kernel system calls. ru_stime—A struct timeval field containing the amount of system time, in seconds, that the process has used. System time is the CPU time spent executing system calls on behalf of the process. ru_maxrss—The largest amount of physical memory occupied by the process’s data at one time over the course of its execution. n n n

The getrusage man page lists all the available fields. See Section 8.7, “gettimeofday: Wall-Clock Time,” for information about struct timeval.

176 Chapter 8 Linux System Calls

The function in Listing 8.5 prints out the current process’s user and system time.
Listing 8.5 (print-cpu-times.c) Display Process User and System Times
#include #include #include #include

void print_cpu_time() { struct rusage usage; getrusage (RUSAGE_SELF, &usage); printf (“CPU time: %ld.%06ld sec user, %ld.%06ld sec system\n”, usage.ru_utime.tv_sec, usage.ru_utime.tv_usec, usage.ru_stime.tv_sec, usage.ru_stime.tv_usec); }

8.7 gettimeofday: Wall-Clock Time system call gets the system’s wall-clock time. It takes a pointer to a variable.This structure represents a time, in seconds, split into two fields.The tv_sec field contains the integral number of seconds, and the tv_usec field contains an additional number of microseconds.This struct timeval value represents the number of seconds elapsed since the start of the UNIX epoch, on midnight UTC on January 1, 1970.The gettimeofday call also takes a second argument, which should be NULL. Include if you use this system call. The number of seconds in the UNIX epoch isn’t usually a very handy way of representing dates.The localtime and strftime library functions help manipulate the return value of gettimeofday.The localtime function takes a pointer to the number of seconds (the tv_sec field of struct timeval) and returns a pointer to a struct tm object.This structure contains more useful fields, which are filled according to the local time zone: tm_hour, tm_min, tm_sec—The time of day, in hours, minutes, and seconds. tm_year, tm_mon, tm_day—The year, month, and date. tm_wday—The day of the week. Zero represents Sunday. tm_yday—The day of the year. tm_isdst—A flag indicating whether daylight savings time is in effect. gettimeofday struct timeval n n n n n

The

The strftime function additionally can produce from the struct tm pointer a customized, formatted string displaying the date and time.The format is specified in a manner similar to printf, as a string with embedded codes indicating which time fields to include. For example, this format string
“%Y-%m-%d %H:%M:%S”

8.8

The mlock Family: Locking Physical Memory 177

specifies the date and time in this form:
2001-01-14 13:09:42

Pass strftime a character buffer to receive the string, the length of that buffer, the format string, and a pointer to a struct tm variable. See the strftime man page for a complete list of codes that can be used in the format string. Notice that neither localtime nor strftime handles the fractional part of the current time more precise than 1 second (the tv_usec field of struct timeval). If you want this in your formatted time strings, you’ll have to include it yourself. Include if you call localtime or strftime. The function in Listing 8.6 prints the current date and time of day, down to the millisecond.
Listing 8.6 (print-time.c) Print Date and Time
#include #include #include #include

void print_time () { struct timeval tv; struct tm* ptm; char time_string[40]; long milliseconds; /* Obtain the time of day, and convert it to a tm struct. */ gettimeofday (&tv, NULL); ptm = localtime (&tv.tv_sec); /* Format the date and time, down to a single second. */ strftime (time_string, sizeof (time_string), “%Y-%m-%d %H:%M:%S”, ptm); /* Compute milliseconds from microseconds. */ milliseconds = tv.tv_usec / 1000; /* Print the formatted time, in seconds, followed by a decimal point and the milliseconds. */ printf (“%s.%03ld\n”, time_string, milliseconds); }

8.8 The mlock Family: Locking Physical Memory
The mlock family of system calls allows a program to lock some or all of its address space into physical memory.This prevents Linux from paging this memory to swap space, even if the program hasn’t accessed it for a while.

178 Chapter 8 Linux System Calls

A time-critical program might lock physical memory because the time delay of paging memory out and back may be too long or too unpredictable. High-security applications may also want to prevent sensitive data from being written out to a swap file, where they might be recovered by an intruder after the program terminates. Locking a region of memory is as simple as calling mlock with a pointer to the start of the region and the region’s length. Linux divides memory into pages and can lock only entire pages at a time; each page that contains part of the memory region specified to mlock is locked.The getpagesize function returns the system’s page size, which is 4KB on x86 Linux. For example, to allocate 32MB of address space and lock it into RAM, you would use this code: const int alloc_size = 32 * 1024 * 1024; char* memory = malloc (alloc_size); mlock (memory, alloc_size);

Note that simply allocating a page of memory and locking it with mlock doesn’t reserve physical memory for the calling process because the pages may be copy-onwrite.5 Therefore, you should write a dummy value to each page as well: size_t i; size_t page_size = getpagesize (); for (i = 0; i < alloc_size; i += page_size) memory[i] = 0;

The write to each page forces Linux to assign a unique, unshared memory page to the process for that page. To unlock a region, call munlock, which takes the same arguments as mlock. If you want your program’s entire address space locked into physical memory, call mlockall. This system call takes a single flag argument: MCL_CURRENT locks all currently allocated memory, but future allocations are not locked; MCL_FUTURE locks all pages that are allocated after the call. Use MCL_CURRENT|MCL_FUTURE to lock into memory both current and subsequent allocations. Locking large amounts of memory, especially using mlockall, can be dangerous to the entire Linux system. Indiscriminate memory locking is a good method of bringing your system to a grinding halt because other running processes are forced to compete for smaller physical memory resources and swap rapidly into and back out of memory (this is known as thrashing). If you lock too much memory, the system will run out of memory entirely and Linux will start killing off processes. For this reason, only processes with superuser privilege may lock memory with mlock or mlockall. If a nonsuperuser process calls one of these functions, it will fail, return –1, and set errno to EPERM. The munlockall call unlocks all memory locked by the current process, including memory locked with mlock and mlockall.
5. Copy-on-write means that Linux makes a private copy of a page of memory for a process only when that process writes a value somewhere into it.

8.9

mprotect: Setting Memory Permissions 179

A convenient way to monitor the memory usage of your program is to use the top command. In the output from top, the SIZE column displays the virtual address space size of each program (the total size of your program’s code, data, and stack, some of which may be paged out to swap space).The RSS column (for resident set size) shows the size of physical memory that each program currently resides in.The sum of all the RSS values for all running programs cannot exceed your computer’s physical memory size, and the sum of all address space sizes is limited to 2GB (for 32-bit versions of Linux). Include if you use any of the mlock system calls.

8.9 mprotect: Setting Memory Permissions
In Section 5.3, “Mapped Memory,” we showed how to use the mmap system call to map a file into memory. Recall that the third argument to mmap is a bitwise or of memory protection flags PROT_READ, PROT_WRITE, and PROT_EXEC for read, write, and execute permission, respectively, or PROT_NONE for no memory access. If a program attempts to perform an operation on a memory location that is not allowed by these permissions, it is terminated with a SIGSEGV (segmentation violation) signal. After memory has been mapped, these permissions can be modified with the mprotect system call.The arguments to mprotect are an address of a memory region, the size of the region, and a set of protection flags.The memory region must consist of entire pages:The address of the region must be aligned to the system’s page size, and the length of the region must be a page size multiple.The protection flags for these pages are replaced with the specified value.
Obtaining Page-Aligned Memory
Note that memory regions returned by malloc are typically not page-aligned, even if the size of the memory is a multiple of the page size. If you want to protect memory obtained from malloc, you will have to allocate a larger memory region and find a page-aligned region within it. Alternately, you can use the mmap system call to bypass malloc and allocate page-aligned memory directly from the Linux kernel. See Section 5.3, “Mapped Memory,” for details.

For example, suppose that your program allocates a page of memory by mapping /dev/zero, as described in Section 5.3.5, “Other Uses for mmap.”The memory is initially both readable and writable. int fd = open (“/dev/zero”, O_RDONLY); char* memory = mmap (NULL, page_size, PROT_READ | PROT_WRITE, MAP_PRIVATE, fd, 0); close (fd);

Later, your program could make the memory read-only by calling mprotect (memory, page_size, PROT_READ);

mprotect:

180 Chapter 8 Linux System Calls

An advanced technique to monitor memory access is to protect the region of memory using mmap or mprotect and then handle the SIGSEGV signal that Linux sends to the program when it tries to access that memory.The example in Listing 8.7 illustrates this technique.
Listing 8.7 (mprotect.c) Detect Memory Access Using mprotect
#include #include #include #include #include #include #include #include

static int alloc_size; static char* memory; void segv_handler (int signal_number) { printf (“memory accessed!\n”); mprotect (memory, alloc_size, PROT_READ | PROT_WRITE); } int main () { int fd; struct sigaction sa; /* Install segv_handler as the handler for SIGSEGV. memset (&sa, 0, sizeof (sa)); sa.sa_handler = &segv_handler; sigaction (SIGSEGV, &sa, NULL); */

/* Allocate one page of memory by mapping /dev/zero. Map the memory as write-only, initially. */ alloc_size = getpagesize (); fd = open (“/dev/zero”, O_RDONLY); memory = mmap (NULL, alloc_size, PROT_WRITE, MAP_PRIVATE, fd, 0); close (fd); /* Write to the page to obtain a private copy. */ memory[0] = 0; /* Make the memory unwritable. */ mprotect (memory, alloc_size, PROT_NONE); /* Write to the allocated memory region. memory[0] = 1; */

8.10

nanosleep: High-Precision Sleeping 181

/* All printf munmap return }

done; unmap the memory. (“all done\n”); (memory, alloc_size); 0;

*/

The program follows these steps: 1. The program installs a signal handler for SIGSEGV. 2. The program allocates a page of memory by mapping /dev/zero and writing a value to the allocated page to obtain a private copy. 3. The program protects the memory by calling mprotect with the PROT_NONE permission. 4. When the program subsequently writes to memory, Linux sends it SIGSEGV, which is handled by segv_handler.The signal handler unprotects the memory, which allows the memory access to proceed. 5. When the signal handler completes, control returns to main, where the program deallocates the memory using munmap.

8.10 nanosleep: High-Precision Sleeping
The nanosleep system call is a high-precision version of the standard UNIX sleep call. Instead of sleeping an integral number of seconds, nanosleep takes as its argument a pointer to a struct timespec object, which can express time to nanosecond precision. However, because of the details of how the Linux kernel works, the actual precision provided by nanosleep is 10 milliseconds—still better than that afforded by sleep.This additional precision can be useful, for instance, to schedule frequent operations with short time intervals between them. The struct timespec structure has two fields: tv_sec, the integral number of seconds, and tv_nsec, an additional number of milliseconds.The value of tv_nsec must be less than 109. The nanosleep call provides another advantage over sleep. As with sleep, the delivery of a signal interrupts the execution of nanosleep, which sets errno to EINTR and returns –1. However, nanosleep takes a second argument, another pointer to a struct timespec object, which, if not null, is filled with the amount of time remaining (that is, the difference between the requested sleep time and the actual sleep time). This makes it easy to resume the sleep operation. The function in Listing 8.8 provides an alternate implementation of sleep. Unlike the ordinary system call, this function takes a floating-point value for the number of seconds to sleep and restarts the sleep operation if it’s interrupted by a signal.

182 Chapter 8 Linux System Calls

Listing 8.8 (better_sleep.c) High-Precision Sleep Function
#include #include int better_sleep (double sleep_time) { struct timespec tv; /* Construct the timespec from the number of whole seconds... tv.tv_sec = (time_t) sleep_time; /* ... and the remainder in nanoseconds. */ tv.tv_nsec = (long) ((sleep_time - tv.tv_sec) * 1e+9);

*/

while (1) { /* Sleep for the time specified in tv. If interrupted by a signal, place the remaining time left to sleep back into tv. int rval = nanosleep (&tv, &tv); if (rval == 0) /* Completed the entire sleep time; all done. */ return 0; else if (errno == EINTR) /* Interrupted by a signal. Try again. */ continue; else /* Some other error; bail out. */ return rval; } return 0; }

*/

8.11 readlink: Reading Symbolic Links
The readlink system call retrieves the target of a symbolic link. It takes three arguments: the path to the symbolic link, a buffer to receive the target of the link, and the length of that buffer. Unusually, readlink does not NUL-terminate the target path that it fills into the buffer. It does, however, return the number of characters in the target path, so NUL-terminating the string is simple. If the first argument to readlink points to a file that isn’t a symbolic link, readlink sets errno to EINVAL and returns –1. The small program in Listing 8.9 prints the target of the symbolic link specified on its command line.

8.12

sendfile: Fast Data Transfers 183

Listing 8.9 (print-symlink.c) Print the Target of a Symbolic Link
#include #include #include int main (int argc, char* argv[]) { char target_path[256]; char* link_path = argv[1]; /* Attempt to read the target of the symbolic link. */ int len = readlink (link_path, target_path, sizeof (target_path)); if (len == -1) { /* The call failed. */ if (errno == EINVAL) /* It’s not a symbolic link; report that. */ fprintf (stderr, “%s is not a symbolic link\n”, link_path); else /* Some other problem occurred; print the generic message. */ perror (“readlink”); return 1; } else { /* NUL-terminate the target path. */ target_path[len] = ‘\0’; /* Print it. */ printf (“%s\n”, target_path); return 0; } }

For example, here’s how you could make a symbolic link and use read it back:
% ln -s /usr/bin/wc my_link % ./print-symlink my_link /usr/bin/wc

print-symlink

to

8.12 sendfile: Fast Data Transfers
The sendfile system call provides an efficient mechanism for copying data from one file descriptor to another.The file descriptors may be open to disk files, sockets, or other devices. Typically, to copy from one file descriptor to another, a program allocates a fixedsize buffer, copies some data from one descriptor into the buffer, writes the buffer out to the other descriptor, and repeats until all the data has been copied.This is inefficient in both time and space because it requires additional memory for the buffer and performs an extra copy of the data into that buffer.

184 Chapter 8 Linux System Calls

Using sendfile, the intermediate buffer can be eliminated. Call sendfile, passing the file descriptor to write to; the descriptor to read from; a pointer to an offset variable; and the number of bytes to transfer.The offset variable contains the offset in the input file from which the read should start (0 indicates the beginning of the file) and is updated to the position in the file after the transfer.The return value is the number of bytes transferred. Include in your program if it uses sendfile. The program in Listing 8.10 is a simple but extremely efficient implementation of a file copy.When invoked with two filenames on the command line, it copies the contents of the first file into a file named by the second. It uses fstat to determine the size, in bytes, of the source file.
Listing 8.10 (copy.c) File Copy Using sendfile
#include #include #include #include #include #include #include

int main (int argc, char* argv[]) { int read_fd; int write_fd; struct stat stat_buf; off_t offset = 0; /* Open the input file. */ read_fd = open (argv[1], O_RDONLY); /* Stat the input file to obtain its size. */ fstat (read_fd, &stat_buf); /* Open the output file for writing, with the same permissions as the source file. */ write_fd = open (argv[2], O_WRONLY | O_CREAT, stat_buf.st_mode); /* Blast the bytes from one file to the other. */ sendfile (write_fd, read_fd, &offset, stat_buf.st_size); /* Close up. */ close (read_fd); close (write_fd); return 0; }

The sendfile call can be used in many places to make copies more efficient. One good example is in a Web server or other network daemon, that serves the contents of a file over the network to a client program.Typically, a request is received from a socket connected to the client computer.The server program opens a local disk file to

8.13

setitimer: Setting Interval Timers 185

retrieve the data to serve and writes the file’s contents to the network socket. Using sendfile can speed up this operation considerably. Other steps need to be taken to make the network transfer as efficient as possible, such as setting the socket parameters correctly. However, these are outside the scope of this book.

8.13 setitimer: Setting Interval Timers
The setitimer system call is a generalization of the alarm call. It schedules the delivery of a signal at some point in the future after a fixed amount of time has elapsed. A program can set three different types of timers with setitimer: n n

n

If the timer code is ITIMER_REAL, the process is sent a SIGALRM signal after the specified wall-clock time has elapsed. If the timer code is ITIMER_VIRTUAL, the process is sent a SIGVTALRM signal after the process has executed for the specified time.Time in which the process is not executing (that is, when the kernel or another process is running) is not counted. If the timer code is ITIMER_PROF, the process is sent a SIGPROF signal when the specified time has elapsed either during the process’s own execution or the execution of a system call on behalf of the process.

The first argument to setitimer is the timer code, specifying which timer to set. The second argument is a pointer to a struct itimerval object specifying the new settings for that timer.The third argument, if not null, is a pointer to another struct itimerval object that receives the old timer settings. A struct itimerval variable has two fields: it_value is a struct timeval field that contains the time until the timer next expires and a signal is sent. If this is 0, the timer is disabled. it_interval is another struct timeval field containing the value to which the timer will be reset after it expires. If this is 0, the timer will be disabled after it expires. If this is nonzero, the timer is set to expire repeatedly after this interval. n n

The struct timeval type is described in Section 8.7, “gettimeofday:Wall-Clock Time.” The program in Listing 8.11 illustrates the use of setitimer to track the execution time of a program. A timer is configured to expire every 250 milliseconds and send a SIGVTALRM signal.
Listing 8.11 (itemer.c) Timer Example
#include #include #include #include

continues

186 Chapter 8 Linux System Calls

Listing 8.11 Continued void timer_handler (int signum) { static int count = 0; printf (“timer expired %d times\n”, ++count); } int main () { struct sigaction sa; struct itimerval timer; /* Install timer_handler as the signal handler for SIGVTALRM. memset (&sa, 0, sizeof (sa)); sa.sa_handler = &timer_handler; sigaction (SIGVTALRM, &sa, NULL); */

/* Configure the timer to expire after 250 msec... */ timer.it_value.tv_sec = 0; timer.it_value.tv_usec = 250000; /* ... and every 250 msec after that. */ timer.it_interval.tv_sec = 0; timer.it_interval.tv_usec = 250000; /* Start a virtual timer. It counts down whenever this process is executing. */ setitimer (ITIMER_VIRTUAL, &timer, NULL); /* Do busy work. while (1); } */

8.14 sysinfo: Obtaining System Statistics
The sysinfo system call fills a structure with system statistics. Its only argument is a pointer to a struct sysinfo. Some of the more interesting fields of struct sysinfo that are filled include these: uptime—Time elapsed since the system booted, in seconds totalram—Total available physical RAM freeram—Free physical RAM procs—Number of processes on the system n n n n

See the

sysinfo

man page for a full description of

structsysinfo. Include

, , and

if you use sysinfo. The program in Listing 8.12 prints some statistics about the current system.

8.15

uname 187

Listing 8.12 (sysinfo.c) Print System Statistics
#include #include #include #include

int main () { /* Conversion constants. */ const long minute = 60; const long hour = minute * 60; const long day = hour * 24; const double megabyte = 1024 * 1024; /* Obtain system statistics. */ struct sysinfo si; sysinfo (&si); /* Summarize interesting values. */ printf (“system uptime : %ld days, %ld:%02ld:%02ld\n”, si.uptime / day, (si.uptime % day) / hour, (si.uptime % hour) / minute, si.uptime % minute); printf (“total RAM : %5.1f MB\n”, si.totalram / megabyte); printf (“free RAM : %5.1f MB\n”, si.freeram / megabyte); printf (“process count : %d\n”, si.procs); return 0; }

8.15 uname
The uname system call fills a structure with various system information, including the computer’s network name and domain name, and the operating system version it’s running. Pass uname a single argument, a pointer to a struct utsname object. Include if you use uname. The call to uname fills in these fields: sysname—The name of the operating system (such as Linux). release, version—The Linux kernel release number and version level. machine—Some information about the hardware platform running Linux. For x86 Linux, this is i386 or i686, depending on the processor. node—The computer’s unqualified hostname. __domain—The computer’s domain name. n n n n n

Each of these fields is a character string. The small program in Listing 8.13 prints the Linux release and version number and the hardware information.

188 Chapter 8 Linux System Calls

Listing 8.13 (print-uname) Print Linux Version Number and Hardware Information
#include #include int main () { struct utsname u; uname (&u); printf (“%s release %s (version %s) on %s\n”, u.sysname, u.release, u.version, u.machine); return 0; }

9
Inline Assembly Code

T

ODAY, FEW PROGRAMMERS USE ASSEMBLY LANGUAGE. Higher-level languages such as C and C++ run on nearly all architectures and yield higher productivity when writing and maintaining code. For occasions when programmers need to use assembly instructions in their programs, the GNU Compiler Collection permits programmers to add architecture-dependent assembly language instructions to their programs. GCC’s inline assembly statements should not be used indiscriminately. Assembly language instructions are architecture-dependent, so, for example, programs using x86 instructions cannot be compiled on PowerPC computers.To use them, you’ll require a facility in the assembly language for your architecture. However, inline assembly statements permit you to access hardware directly and can also yield faster code. An asm instruction allows you to insert assembly instructions into C and C++ programs. For example, this instruction asm (“fsin” : “=t” (answer) : “0” (angle));

is an x86-speciﬁc way of coding this C statement:1 answer = sin (angle);

1.The expression sin (angle) is usually implemented as a function call into the math library, but if you specify the -O1 or higher optimization ﬂag, GCC is smart enough to replace the function call with a single fsin assembly instruction.

190 Chapter 9 Inline Assembly Code

Observe that unlike ordinary assembly code instructions, asm statements permit you to specify input and output operands using C syntax. To read more about the x86 instruction set, which we will use in this chapter, see http://developer.intel.com/design/pentiumii/manuals/ and http://www.x86-64.org/documentation.

9.1 When to Use Assembly Code
Although asm statements can be abused, they allow your programs to access the computer hardware directly, and they can produce programs that execute quickly. You can use them when writing operating system code that directly needs to interact with hardware. For example, /usr/include/asm/io.h contains assembly instructions to access input/output ports directly.The Linux source code ﬁle /usr/src/linux/arch/i386/kernel/process.s provides another example, using hlt in idle loop code. See other Linux source code ﬁles in /usr/src/linux/arch/ and /usr/src/linux/drivers/. Assembly instructions can also speed the innermost loop of computer programs. For example, if the majority of a program’s running time is computing the sine and cosine of the same angles, this innermost loop could be recoded using the fsincos x86 instruction.2 See, for example, /usr/include/bits/mathinline.h, which wraps up into macros some inline assembly sequences that speed transcendental function computation. You should use inline assembly to speed up code only as a last resort. Current compilers are quite sophisticated and know a lot about the details of the processors for which they generate code.Therefore, compilers can often choose code sequences that may seem unintuitive or roundabout but that actually execute faster than other instruction sequences. Unless you understand the instruction set and scheduling attributes of your target processor very well, you’re probably better off letting the compiler’s optimizers generate assembly code for you for most operations. Occasionally, one or two assembly instructions can replace several lines of higherlevel language code. For example, determining the position of the most signiﬁcant nonzero bit of a nonzero integer using the C programming languages requires a loop or ﬂoating-point computations. Many architectures, including the x86, have a single assembly instruction (bsr) to compute this bit position.We’ll demonstrate the use of one of these in Section 9.4, “Example.”

2. Algorithmic or data structure changes may be more effective in reducing a program’s running time than using assembly instructions.

9.2

Simple Inline Assembly 191

9.2 Simple Inline Assembly
Here we introduce the syntax of shift a value 8 bits to the right: asm assembler instructions with an x86 example to

asm (“shrl $8, %0” : “=r” (answer) : “r” (operand) : “cc”);

The keyword asm is followed by a parenthetic expression consisting of sections separated by colons.The ﬁrst section contains an assembler instruction and its operands. In this example, shrl right-shifts the bits in its ﬁrst operand. Its ﬁrst operand is represented by %0. Its second operand is the immediate constant $8. The second section speciﬁes the outputs.The instruction’s one output will be placed in the C variable answer, which must be an lvalue.The string “=r” contains an equals sign indicating an output operand and an r indicating that answer is stored in a register. The third section speciﬁes the inputs.The C variable operand speciﬁes the value to shift.The string “r” indicates that it is stored in a register but omits an equals sign because it is an input operand, not an output operand. The fourth section indicates that the instruction changes the value in the condition code cc register.

9.2.1 Converting an asm to Assembly Instructions
GCC’s treatment of asm statements is very simple. It produces assembly instructions to deal with the asm’s operands, and it replaces the asm statement with the instruction that you specify. It does not analyze the instruction in any way. For example, GCC converts this program fragment double foo, bar; asm (“mycool_asm %1, %0” : “=r” (bar) : “r” (foo));

to these x86 assembly instructions: movl -8(%ebp),%edx movl -4(%ebp),%ecx #APP mycool_asm %edx, %edx #NO_APP movl %edx,-16(%ebp) movl %ecx,-12(%ebp)

Remember that foo and bar each require two words of stack storage on a 32-bit x86 architecture.The register ebp points to data on the stack. The ﬁrst two instructions copy foo into registers EDX and ECX on which mycool_asm operates.The compiler decides to use the same registers to store the answer, which is copied into bar by the ﬁnal two instructions. It chooses appropriate registers, even reusing the same registers, and copies operands to and from the proper locations automatically.

192 Chapter 9 Inline Assembly Code

9.3 Extended Assembly Syntax
In the subsections that follow, we describe the syntax rules for asm statements.Their sections are separated by colons. We will refer to this illustrative asm statement, which computes the Boolean expression x > y: asm (“fucomip %%st(1), %%st; seta %%al” : “=a” (result) : “u” (y), “t” (x) : “cc”, “st”);

First, fucomip compares its two operands x and y, and stores values indicating the result into the condition code register.Then seta converts these values into a 0 or 1 result.

9.3.1 Assembler Instructions
The ﬁrst section contains the assembler instructions, enclosed in quotation marks.The example asm contains two assembly instructions, fucomip and seta, separated by semicolons. If the assembler does not permit semicolons, use newline characters (\n) to separate instructions. The compiler ignores the contents of this ﬁrst section, except that one level of percentage signs is removed, so %% changes to %.The meaning of %%st(1) and other such terms is architecture-dependent. GCC will complain if you specify the -traditional option or the -ansi option when compiling a program containing asm statements.To avoid producing these errors, such as in header ﬁles, use the alternative keyword __asm__.

9.3.2

Outputs

The second section speciﬁes the instructions’ output operands using C syntax. Each operand is speciﬁed by an operand constraint string followed by a C expression in parentheses. For output operands, which must be lvalues, the constraint string should begin with an equals sign.The compiler checks that the C expression for each output operand is in fact an lvalue. Letters specifying registers for a particular architecture can be found in the GCC source code, in the REG_CLASS_FROM_LETTER macro. For example, the gcc/conﬁg/i386/i386.h conﬁguration ﬁle in GCC lists the register letters for the x86 architecture.3 Table 9.1 summarizes these.

3. You’ll need to have some familiarity with GCC’s internals to make sense of this ﬁle.

9.3

Extended Assembly Syntax 193

Table 9.1 R q f t u a b c d x y A D S

Register Letters for the Intel x86 Architecture Registers That GCC May Use General register (EAX, EBX, ECX, EDX, ESI, EDI, EBP, ESP) General register for data (EAX, EBX, ECX, EDX) Floating-point register Top ﬂoating-point register Second-from-top ﬂoating-point register EAX register EBX register ECX register EDX register SSE register (Streaming SIMD Extension register) MMX multimedia registers An 8-byte value formed from EAX and EDX Destination pointer for string operations (EDI) Source pointer for string operations (ESI)

Register Letter

Multiple operands in an asm statement, each speciﬁed by a constraint string and a C expression, are separated by commas, as illustrated in the example asm’s input section. You may specify up to 10 operands, denoted %0, %1, …, %9, in the output and input sections. If there are no output operands but there are input operands or clobbered registers, leave the output section empty or mark it with a comment like /* no outputs */.

9.3.3

Inputs

The third section speciﬁes the input operands for the assembler instructions.The constraint string for an input operand should not have an equals sign, which indicates an lvalue. Otherwise, an input operand’s syntax is the same as for output operands. To indicate that a register is both read from and written to in the same asm, use an input constraint string of the output operand’s number. For example, to indicate that an input register is the same as the ﬁrst output register number, use 0. Output operands are numbered left to right, starting with 0. Merely specifying the same C expression for an output operand and an input operand does not guarantee that the two values will be placed in the same register. This input section can be omitted if there are no input operands and the subsequent clobber section is empty.

194 Chapter 9 Inline Assembly Code

9.3.4

Clobbers

If an instruction modiﬁes the values of one or more registers as a side effect, specify the clobbered registers in the asm’s fourth section. For example, the fucomip instruction modiﬁes the condition code register, which is denoted cc. Separate strings representing clobbered registers with commas. If the instruction can modify an arbitrary memory location, specify memory. Using the clobber information, the compiler determines which values must be reloaded after the asm executes. If you don’t specify this information correctly, GCC may assume incorrectly that registers still contain values that have, in fact, been overwritten, which will affect your program’s correctness.

9.4 Example
The x86 architecture includes instructions that determine the positions of the least signiﬁcant set bit and the most signiﬁcant set bit in a word.The processor can execute these instructions quite efﬁciently. In contrast, implementing the same operation in C requires a loop and a bit shift. For example, the bsrl assembly instruction computes the position of the most signiﬁcant bit set in its ﬁrst operand, and places the bit position (counting from 0, the least signiﬁcant bit) into its second operand.To place the bit position for number into position, we could use this asm statement: asm (“bsrl %1, %0” : “=r” (position) : “r” (number));

One way you could implement the same operation in C is using this loop: long i; for (i = (number >> 1), position = 0; i != 0; ++position) i >>= 1;

To test the relative speeds of these two versions, we’ll place them in a loop that computes the bit positions for a large number of values. Listing 9.1 does this using the C loop implementation.The program loops over integers, from 1 up to the value speciﬁed on the command line. For each value of number, it computes the most signiﬁcant bit that is set. Listing 9.2 does the same thing using the inline assembly instruction. Note that in both versions, we assign the computed bit position to a volatile variable result.This is to coerce the compiler’s optimizer so that it does not eliminate the entire bit position computation; if the result is not used or stored in memory, the optimizer eliminates the computation as “dead code.”
Listing 9.1 (bit-pos-loop.c) Find Bit Position Using a Loop
#include #include int main (int argc, char* argv[]) { long max = atoi (argv[1]); long number;

9.4

Example 195

long i; unsigned position; volatile unsigned result; /* Repeat the operation for a large number of values. */ for (number = 1; number > 1), position = 0; i != 0; ++position) i >>= 1; /* The position of the most signiﬁcant set bit is the number of shifts we needed after the ﬁrst one. */ result = position; } return 0; }

Listing 9.2 (bit-pos-asm.c) Find Bit Position Using bsrl
#include #include int main (int argc, char* argv[]) { long max = atoi (argv[1]); long number; unsigned position; volatile unsigned result; /* Repeat the operation for a large number of values. */ for (number = 1; number hello sh: ./hello: Permission denied % ./hello sh: ./hello: Permission denied

We can’t read the file, which is why cat fails; we can’t write to the file, which is why echo fails; and we can’t run the file, which is why ./hello fails.

3.You’ll sometimes see the permission bits for a file referred to as the file’s mode.The name of the chmod command is short for “change mode.”

202 Chapter 10 Security Things are better if we are accessing the file as group: mitchell, who

is a member of the

csl

% id uid=501(mitchell) gid=501(mitchell) groups=501(mitchell),503(csl) % cat hello #!/bin/bash echo “Hello, world.” % ./hello Hello, world. % echo hi > hello bash: ./hello: Permission denied

We can list the contents of the file, and we can run it (it’s a simple shell script), but we still can’t write to it. If we run as the owner (samuel), we can even overwrite the file:
% id uid=502(samuel) gid=502(samuel) groups=502(samuel),503(csl) % echo hi > hello % cat hello hi

You can change the permissions associated with a file only if you are the file’s owner (or the superuser). For example, if you now want to allow everyone to execute the file, you can do this:
% chmod o+x hello % ls -l hello -rwxr-x--x 1 samuel

csl

3 Jan 22 16:38 hello

Note that there’s now an x at the end of the first string of characters.The o+x bit means that you want to add the execute permission for other people (not the file’s owner or members of its owning group).You could use g-w instead, to remove the write permission from the group. See the man page in section 1 for chmod for details about this syntax:
% man 1 chmod

Programmatically, you can use the stat system call to find the permissions associated with a file.This function takes two parameters: the name of the file you want to find out about, and the address of a data structure that is filled in with information about the file. See Appendix B,“Low-Level I/O,” Section B.2,“stat,” for a discussion of other information that you can obtain with stat. Listing 10.2 shows an example of using stat to obtain file permissions.
Listing 10.2 (stat-perm.c) Determine File Owner’s Write Permission
#include #include int main (int argc, char* argv[]) { const char* const filename = argv[1]; struct stat buf;

10.3

File System Permissions 203

/* Get file information. */ stat (filename, &buf); /* If the permissions are set such that the file’s owner can write to it, print a message. */ if (buf.st_mode & S_IWUSR) printf (“Owning user can write `%s’.\n”, filename); return 0; }

If you run this program on our

hello

program, it says:

% ./stat-perm hello Owning user can write ‘hello’.

The S_IWUSR constant corresponds to write permission for the owning user.There are other constants for all the other bits. For example, S_IRGRP is read permission for the owning group, and S_IXOTH is execute permission for users who are neither the owning user nor a member of the owning group. If you store permissions in a variable, use the typedef mode_t for that variable. Like most system calls, stat will return -1 and set errno if it can’t obtain information about the file. You can use the chmod function to change the permission bits on an existing file. You call chmod with the name of the file you want to change and the permission bits you want set, presented as the bitwise or of the various permission constants mentioned previously. For example, this next line would make hello readable and executable by its owning user but would disable all other permissions associated with hello: chmod (“hello”, S_IRUSR

|

S_IXUSR);

The same permission bits apply to directories, but they have different meanings. If a user is allowed to read from a directory, the user is allowed to see the list of files that are present in that directory. If a user is allowed to write to a directory, the user is allowed to add or remove files from the directory. Note that a user may remove files from a directory if she is allowed to write to the directory, even if she does not have permission to modify the file she is removing. If a user is allowed to execute a directory, the user is allowed to enter that directory and access the files therein.Without execute access to a directory, a user is not allowed to access the files in that directory independent of the permissions on the files themselves. To summarize, let’s review how the kernel decides whether to allow a process to access a particular file. It checks to see whether the accessing user is the owning user, a member of the owning group, or someone else.The category into which the accessing user falls is used to determine which set of read/write/execute bits are checked.Then the kernel checks the operation that is being performed against the permission bits that apply to this user.4
4.The kernel may also deny access to a file if a component directory in its file path is inaccessible. For instance, if a process may not access the directory /tmp/private/, it may not read /tmp/private/data, even if the permissions on the latter are set to allow the access.

204 Chapter 10 Security

There is one important exception: Processes running as root (those with user ID 0) are always allowed to access any file, regardless of the permissions associated with it.

10.3.1 Security Hole: Programs Without Execute Permissions
Here’s a first example of where security gets very tricky.You might think that if you disallow execution of a program, then nobody can run it. After all, that’s what it means to disallow execution. But a malicious user can make a copy of the program, change the permissions to make it executable, and then run the copy! If you rely on users not being able to run programs that aren’t executable but then don’t prevent them from copying the programs, you have a security hole—a means by which users can perform some action that you didn’t intend.

10.3.2

Sticky Bits

In addition to read, write, and execute permissions, there is a magic bit called the sticky bit.5 This bit applies only to directories. A directory that has the sticky bit set allows you to delete a file only if you are the owner of the file. As mentioned previously, you can ordinarily delete a file if you have write access to the directory that contains it, even if you are not the file’s owner.When the sticky bit is set, you still must have write access to the directory, but you must also be the owner of the file that you want to delete. A few directories on the typical GNU/Linux system have the sticky bit set. For example, the /tmp directory, in which any user can place temporary files, has the sticky bit set.This directory is specifically designed to be used by all users, so the directory must be writable by everyone. But it would be bad if one user could delete another user’s files, so the sticky bit is set on the directory.Then only the owning user (or root, of course) can remove a file. You can see the sticky bit is set because of the t at the end of the permission bits when you run ls on /tmp:
% ls -ld /tmp drwxrwxrwt 12 root root 2048 Jan 24 17:51 /tmp

The corresponding constant to use with stat and chmod is S_ISVTX. If your program creates directories that behave like /tmp, in that lots of people put things there but shouldn’t be able to remove each other’s files, then you should set the sticky bit on the directory.You can set the sticky bit on a directory with the chmod command by invoking the following:
% chmod o+t directory

5.This name is anachronistic; it goes back to a time when setting the sticky bit caused a program to be retained in main memory even when it was done executing.The pages allocated to the program were “stuck” in memory.

10.4

Real and Effective IDs 205

To set the sticky bit programmatically, call chmod with the S_ISVTX mode flag. For example, to set the sticky bit of the directory specified by dir_path to those of the /tmp and give full read, write, and execute permissions to all users, use this call: chmod (dir_path, S_IRWXU

|

S_IRWXG

|

S_IRWXO

|

S_ISVTX);

10.4

Real and Effective IDs

Until now, we’ve talked about the user ID and group ID associated with a process as if there were only one such user ID and one such group ID. But, actually, it’s not quite that simple. Every process really has two user IDs: the effective user ID and the real user ID. (Of course, there’s also an effective group ID and real group ID. Just about everything that’s true about user IDs is also true about group IDs.) Most of the time, the kernel checks only the effective user ID. For example, if a process tries to open a file, the kernel checks the effective user ID when deciding whether to let the process access the file. The geteuid and getegid functions described previously return the effective user ID and the effective group ID. Corresponding getuid and getgid functions return the real user ID and real group ID. If the kernel cares about only the effective user ID, it doesn’t seem like there’s much point in having a distinction between a real user ID and an effective user ID. However, there is one very important case in which the real user ID matters. If you want to change the effective user ID of an already running process, the kernel looks at the real user ID as well as the effective user ID. Before looking at how you can change the effective user ID of a process, let’s examine why you would want to do such a thing by looking back at our accounting package. Suppose that there’s a server process that might need to look at any file on the system, regardless of the user who created it. Such a process must run as root because only root can be guaranteed to be capable of looking at any file. But now suppose that a request comes in from a particular user (say, mitchell) to access some file.The server process could carefully examine the permissions associated with the files in question and try to decide whether mitchell should be allowed to access those files. But that would mean duplicating all the processing that the kernel would normally do to check file access permissions. Reimplementing that logic would be complex, errorprone, and tedious. A better approach is simply to temporarily change the effective user ID of the process from root to mitchell and then try to perform the operations required. If mitchell is not allowed to access the data, the kernel will prevent the process from doing so and will return appropriate indications of error. After all the operations taken on behalf of mitchell are complete, the process can restore its original effective user ID to root.

206 Chapter 10 Security

Programs that authenticate users when they log in take advantage of the capability to change user IDs as well.These login programs run as root.When the user enters a username and password, the login program verifies the username and password in the system password database.Then the login program changes both the effective user ID and the real ID to be that of the user. Finally, the login program calls exec to start the user’s shell, leaving the user running a shell whose effective user ID and real user ID are that of the user. The function used to change the user IDs for a process is setreuid. (There is, of course, a corresponding setregid function as well.) This function takes two arguments.The first argument is the desired real user ID; the second is the desired effective user ID. For example, here’s how you would exchange the effective and real user IDs: setreuid (geteuid(), getuid ());

Obviously, the kernel won’t let just any process change its user IDs. If a process were allowed to change its effective user ID at will, then any user could easily impersonate any other user, simply by changing the effective user ID of one of his processes.The kernel will let a process running with an effective user ID of 0 change its user IDs as it sees fit. (Again, notice how much power a process running as root has! A process whose effective user ID is 0 can do absolutely anything it pleases.) Any other process, however, can do only one of the following things: Set its effective user ID to be the same as its real user ID Set its real user ID to be the same as its effective user ID Swap the two user IDs n n n

The first alternative would be used by our accounting process when it has finished accessing files as mitchell and wants to return to being root.The second alternative could be used by a login program after it has set the effective user ID to that of the user who just logged in. Setting the real user ID ensures that the user will never be able go back to being root. Swapping the two user IDs is almost a historical artifact; modern programs rarely use this functionality. You can pass -1 to either argument to setreuid if you want to leave that user ID alone.There’s also a convenience function called seteuid.This function sets the effective user ID, but it doesn’t modify the real user ID.The following two statements both do exactly the same thing: seteuid (id); setreuid (-1, id);

10.4.1 Setuid Programs
Using the previous techniques, you know how to make a root process impersonate another process temporarily and then return to being root.You also know how to make a root process drop all its special privileges by setting both its real user ID and its effective user ID.

10.4

Real and Effective IDs 207

Here’s a puzzle: Can you, running as a non-root user, ever become root? That doesn’t seem possible, using the previous techniques, but here’s proof that it can be done:
% whoami mitchell % su Password: ... % whoami root

The whoami command is just like id, except that it shows only the effective user ID, not all the other information.The su command enables you to become the superuser if you know the root password. How does su work? Because we know that the shell was originally running with both its real user ID and its effective user ID set to mitchell, setreuid won’t allow us to change either user ID. The trick is that the su program is a setuid program.That means that when it is run, the effective user ID of the process will be that of the file’s owner rather than the effective user ID of the process that performed the exec call. (The real user ID will still be that of the executing user.) To create a setuid program, you use chmod +s at the command line, or use the S_ISUID flag if calling chmod programmatically.6 For example, consider the program in Listing 10.3.
Listing 10.3 (setuid-test.c) Setuid Demonstration Program
#include #include int main () { printf (“uid=%d euid=%d\n”, (int) getuid (), (int) geteuid ()); return 0; }

Now suppose that this program is setuid and owned by put will look like this:
-rwsrws--x 1 root root

root. In

that case, the

ls

out-

11931 Jan 24 18:25 setuid-test

The s bits indicate that the file is not only executable (as an x bit would indicate) but also setuid and setgid.When we use this program, we get output like this:
% whoami mitchell % ./setuid-test uid=501 euid=0

6. Of course, there is a similar notion of a setgid program.When run, its effective group ID is the same as that of the group owner of the file. Most setuid programs are also setgid programs.

208 Chapter 10 Security

Note that the effective user ID is set to 0 when the program is run. You can use the chmod command with the u+s or g+s arguments to set the setuid and setgid bits on an executable file, respectively—for example:
% ls -l program -rwxr-xr-x 1 samuel % chmod g+s program % ls -l program -rwxr-sr-x 1 samuel % chmod u+s program % ls -l program -rwsr-sr-x 1 samuel csl 0 Jan 30 23:38 program

csl

0 Jan 30 23:38 program

csl

0 Jan 30 23:38 program

You can also use the chmod call with the S_ISUID or S_ISGID mode flags. su is capable of changing the effective user ID through this mechanism. It runs initially with an effective user ID of 0.Then it prompts you for a password. If the password matches the root password, it sets its real user ID to be root as well and then starts a new shell. Otherwise, it exits, unceremoniously leaving you as a nonprivileged user. Take a look at the permissions on the su program:
% ls -l /bin/su -rwsr-xr-x 1 root root 14188 Mar 7 2000 /bin/su

Notice that it’s owned by root and that the setuid bit is set. Note that su doesn’t actually change the user ID of the shell from which it was run. Instead, it starts a new shell process with the new user ID.The original shell is blocked until the new shell completes and su exits.

10.5 Authenticating Users
Often, if you have a setuid program, you don’t want to offer its services to everyone. For example, the su program lets you become root only if you know the root password.The program makes you prove that you are entitled to become root before going ahead with its actions.This process is called authentication—the su program is checking to see that you are authentic. If you’re administering a very secure system, you probably don’t want to let people log in just by typing an ordinary password. Users tend to write down passwords, and black hats tend to find them. Users tend to pick passwords that involve their birthdays, the names of their pets, and so forth.7 Passwords just aren’t all that secure.

7. It has been found that system administrators tend to pick the word god as their password more often than any other password. (Make of that what you will.) So, if you ever need root access on a machine and the sysadmin isn’t around, a little divine inspiration might be just what you need.

10.5

Authenticating Users 209

For example, many organizations now require the use of special “one-time” passwords that are generated by special electronic ID cards that users keep with them.The same password can’t be used twice, and you can’t get a valid password out of the ID card without entering a PIN. So, an attacker must obtain both the physical card and the PIN to break in. In a really secure facility, retinal scans or other kinds of biometric testing are used. If you’re writing a program that must perform authentication, you should allow the system administrator to use whatever means of authentication is appropriate for that installation. GNU/Linux comes with a very useful library that makes this very easy. This facility, called Pluggable Authentication Modules, or PAM, makes it easy to write applications that authenticate their users as the system administrator sees fit. It’s easiest to see how PAM works by looking at a simple PAM application. Listing 10.4 illustrates the use of PAM.
Listing 10.4 ( pam.c) PAM Example
#include #include #include int main () { pam_handle_t* pamh; struct pam_conv pamc; /* Set up the PAM conversation. */ pamc.conv = &misc_conv; pamc.appdata_ptr = NULL; /* Start a new authentication session. */ pam_start (“su”, getenv (“USER”), &pamc, &pamh); /* Authenticate the user. */ if (pam_authenticate (pamh, 0) != PAM_SUCCESS) fprintf (stderr, “Authentication failed!\n”); else fprintf (stderr, “Authentication OK.\n”); /* All done. */ pam_end (pamh, 0); return 0; }

To compile this program, you have to link it with two libraries: the a helper library called libpam_misc:
% gcc -o pam pam.c -lpam -lpam_misc

libpam

library and

210 Chapter 10 Security

This program starts off by building up a PAM conversation object.This object is used by the PAM library whenever it needs to prompt the user for information.The misc_conv function used in this example is a standard conversation function that uses the terminal for input and output.You could write your own function that pops up a dialog box, or that uses speech for input and output, or that provides even more exotic input and output methods. The program then calls pam_start.This function initializes the PAM library.The first argument is a service name.You should use a name that uniquely identifies your application. For example, if your application is named whizbang, you should probably use that for the service name, too. However, the program probably won’t work until the system administrator explicitly configures the system to work with your service. So, in this example, we use the su service, which says that our program should authenticate users in the same way that the su command does.You should not use this technique in a real program. Pick a real service name, and have your installation scripts help the system administrator to set up a correct PAM configuration for your application. The second argument is the name of the user whom you want to authenticate. In this example, we use the value of the USER environment variable. (Normally, this is the username that corresponds to the effective user ID of the current process, but that’s not always the case.) In most real programs, you would prompt for a username at this point.The third argument indicates the PAM conversation, discussed previously.The call to pam_start fills in the handle provided as the fourth argument. Pass this handle to subsequent calls to PAM library routines. Next, the program calls pam_authenticate.The second argument enables you to pass various flags; the value 0 means to use the default options.The return value from this function indicates whether authentication succeeded. Finally, the programs calls pam_end to clean up any allocated data structures. Let’s assume that the valid password for the current user is “password” (an exceptionally poor password).Then, running this program with the correct password produces the expected:
% ./pam Password: password Authentication OK.

If you run this program in a terminal, the password probably won’t actually appear when you type it in; it’s hidden to prevent others from peeking at your password over your shoulder as you type. However, if a hacker tries to use the wrong password, the PAM library will correctly indicate failure:
% ./pam Password: badguess Authentication failed!

10.6

More Security Holes 211

The basics covered here are enough for most simple programs. Full documentation about how PAM works is available in /usr/doc/pam on most GNU/Linux systems.

10.6 More Security Holes
Although this chapter will point out a few common security holes, you should by no means rely on this book to cover all possible security holes. A great many have already been discovered, and many more are out there waiting to be found. If you are trying to write secure code, there is really no substitute for having a security expert audit your code.

10.6.1 Buffer Overruns
Almost every major Internet application daemon, including the sendmail daemon, the finger daemon, the talk daemon, and others, has at one point been compromised through a buffer overrun. If you are writing any code that will ever be run as root, you absolutely must be aware of this particular kind of security hole. If you are writing a program that performs any kind of interprocess communication, you should definitely be aware of this kind of security hole. If you are writing a program that reads files (or might read files) that are not owned by the user executing the program, you should be aware of this kind of security hole.That last criterion applies to almost every program. Fundamentally, if you’re going to write GNU/Linux software, you ought to know about buffer overruns. The idea behind a buffer overrun attack is to trick a program into executing code that it did not intend to execute.The usual mechanism for achieving this feat is to overwrite some portion of the program’s process stack.The program’s stack contains, among other things, the memory location to which the program will transfer control when the current function returns.Therefore, if you can put the code that you want to have executed into memory somewhere and then change the return address to point to that piece of memory, you can cause the program to execute anything.When the program returns from the function it is executing, it will jump to the new code and execute whatever is there, running with the privileges of the current process. Clearly, if the current process is running as root, this would be a disaster. If the process is running as another user, it’s a disaster “only” for that user—and anybody else who depends on the contents of files owned by that user, and so forth. If the program is running as a daemon and listening for incoming network connections, the situation is even worse. A daemon typically runs as root. If it contains buffer overrun bugs, anyone who can connect via the network to a computer running the daemon can seize control of the computer by sending a malignant sequence of data to the daemon over the network. A program that does not engage in network communications is much safer because only users who are already able to log in to the computer running the program are able to attack it.

212 Chapter 10 Security

The buggy versions of finger, talk, and sendmail all shared a common flaw. Each used a fixed-length string buffer, which implied a constant upper limit on the size of the string but then allowed network clients to provide strings that overflowed the buffer. For example, they contained code similar to this:
#include int main () { /* Nobody in their right mind would have more than 32 characters in their username. Plus, I think UNIX allows only 8-character usernames. So, this should be plenty of space. */ char username[32]; /* Prompt the user for the username. */ printf (“Enter your username: “); /* Read a line of input. */ gets (username); /* Do other things here... */ return 0; }

The combination of the 32-character buffer with the gets function permits a buffer overrun.The gets function reads user input up until the next newline character and stores the entire result in the username buffer.The comments in the code are correct in that people generally have short usernames, so no well-meaning user is likely to type in more than 32 characters. But when you’re writing secure software, you must consider what a malicious attacker might do. In this case, the attacker might deliberately type in a very long username. Local variables such as username are stored on the stack, so by exceeding the array bounds, it’s possible to put arbitrary bytes onto the stack beyond the area reserved for the username variable.The username will overrun the buffer and overwrite parts of the surrounding stack, allowing the kind of attack described previously. Fortunately, it’s relatively easy to prevent buffer overruns.When reading strings, you should always use a function, such as getline, that either dynamically allocates a sufficiently large buffer or stops reading input if the buffer is full. For example, you could use this: char* username = getline (NULL, 0, stdin);

This call automatically uses malloc to allocate a buffer big enough to hold the line and returns it to you.You have to remember to call free to deallocate the buffer, of course, to avoid leaking memory. Your life will be even easier if you use C++ or another language that provides simple primitives for reading input. In C++, for example, you can simply use this: string username; getline (cin, username);

10.6

More Security Holes 213

The username string will automatically be deallocated as well; you don’t have to remember to free it.8 Of course, buffer overruns can occur with any statically sized array, not just with strings. If you want to write secure code, you should never write into a data structure, on the stack or elsewhere, without verifying that you’re not going to write beyond its region of memory.

10.6.2 Race Conditions in /tmp
Another very common problem involves the creation of files with predictable names, typically in the /tmp directory. Suppose that your program prog, running as root, always creates a temporary file called /tmp/prog and writes some vital information there. A malicious user can create a symbolic link from /tmp/prog to any other file on the system.When your program goes to create the file, the open system call will succeed. However, the data that you write will not go to /tmp/prog; instead, it will be written to some arbitrary file of the attacker’s choosing. This kind of attack is said to exploit a race condition.There is implicitly a race between you and the attacker.Whoever manages to create the file first wins. This attack is often used to destroy important parts of the file system. By creating the appropriate links, the attacker can trick a program running as root that is supposed to write a temporary file into overwriting an important system file instead. For example, by making a symbolic link to /etc/passwd, the attacker can wipe out the system’s password database.There are also ways in which a malicious user can obtain root access using this technique. One attempt at avoiding this attack is to use a randomized name for the file. For example, you could read from /dev/random to get some bits to use in the name of the file.This certainly makes it harder for a malicious user to guess the filename, but it doesn’t make it impossible.The attacker might just create a large number of symbolic links, using many potential names. Even if she has to try 10,000 times before wining the race condition, that one time could be disastrous. Another approach is to use the O_EXCL flag when calling open.This flag causes open to fail if the file already exists. Unfortunately, if you’re using the Network File System (NFS), or if anyone who’s using your program might ever be using NFS, that’s not a sufficiently robust approach because O_EXCL is not reliable when NFS is in use.You can’t ever really know for sure whether your code will be used on a system that uses NFS, so if you’re highly paranoid, don’t rely on using O_EXCL. In Chapter 2,“Writing Good GNU/Linux Software,” Section 2.1.7,“Using Temporary Files,” we showed how to use mkstemp to create temporary files. Unfortunately, what mkstemp does on Linux is open the file with O_EXCL after trying to pick a name that is hard to guess. In other words, using mkstemp is still insecure if /tmp is mounted over NFS.9 So, using mkstemp is better than nothing, but it’s not fully secure.
8. Some programmers believe that C++ is a horrible and overly complex language.Their arguments about multiple inheritance and other such complications have some merit, but it is easier to write code that avoids buffer overruns and other similar problems in C++ than in C. 9. Obviously, if you’re also a system administrator, you shouldn’t mount /tmp over NFS.

214 Chapter 10 Security

One approach that works is to call lstat on the newly created file (lstat is discussed in Section B.2, “stat”).The lstat function is like stat, except that if the file referred to is a symbolic link, lstat tells you about the link, not the file to which it refers. If lstat tells you that your new file is an ordinary file, not a symbolic link, and that it is owned by you, then you should be okay. Listing 10.5 presents a function that tries to securely open a file in /tmp.The authors of this book have not had it audited professionally, nor are we professional security experts, so there’s a good chance that it has a weakness, too.We do not recommend that you use this code without getting an audit, but it should at least convince you that writing secure code is tricky.To help dissuade you, we’ve deliberately made the interface difficult to use in real programs. Error checking is an important part of writing secure software, so we’ve included error-checking logic in this example.
Listing 10.5 (temp-file.c) Create a Temporary File
#include #include #include #include

/* Returns the file descriptor for a newly created temporary file. The temporary file will be readable and writable by the effective user ID of the current process but will not be readable or writable by anybody else. Returns -1 if the temporary file could not be created. */

int secure_temp_file () { /* This file descriptor points to /dev/random and allows us to get a good source of random bits. */ static int random_fd = -1; /* A random integer. */ unsigned int random; /* A buffer, used to convert from a numeric to a string representation of random. This buffer has fixed size, meaning that we potentially have a buffer overrun bug if the integers on this machine have a *lot* of bits. */ char filename[128]; /* The file descriptor for the new temporary file. */ int fd; /* Information about the newly created file. */ struct stat stat_buf; /* If we haven’t already opened /dev/random, do so now. not threadsafe.) */ if (random_fd == -1) { (This is

10.6

More Security Holes 215

/* Open /dev/random. Note that we’re assuming that /dev/random really is a source of random bits, not a file full of zeros placed there by an attacker. */ random_fd = open (“/dev/random”, O_RDONLY); /* If we couldn’t open /dev/random, give up. */ if (random_fd == -1) return -1; } /* Read an integer from /dev/random. */ if (read (random_fd, &random, sizeof (random)) != sizeof (random)) return -1; /* Create a filename out of the random number. */ sprintf (filename, “/tmp/%u”, random); /* Try to open the file. */ fd = open (filename, /* Use O_EXECL, even though it doesn’t work under NFS. */ O_RDWR | O_CREAT | O_EXCL, /* Make sure nobody else can read or write the file. */ S_IRUSR | S_IWUSR); if (fd == -1) return -1; /* Call lstat on the file, to make sure that it is not a symbolic link. */ if (lstat (filename, &stat_buf) == -1) return -1; /* If the file is not a regular file, someone has tried to trick us. */ if (!S_ISREG (stat_buf.st_mode)) return -1; /* If we don’t own the file, someone else might remove it, read it, or change it while we’re looking at it. */ if (stat_buf.st_uid != geteuid () || stat_buf.st_gid != getegid ()) return -1; /* If there are any more permission bits set on the file, something’s fishy. */ if ((stat_buf.st_mode & ~(S_IRUSR | S_IWUSR)) != 0) return -1; return fd; }

This function calls open to create the file and then calls lstat a few lines later to make sure that the file is not a symbolic link. If you’re thinking carefully, you’ll realize that there seems to be a race condition at this point. In particular, an attacker could remove the file and replace it with a symbolic link between the time we call open and the

216 Chapter 10 Security

time we call lstat.That won’t harm us directly because we already have an open file descriptor to the newly created file, but it will cause us to indicate an error to our caller.This attack doesn’t create any direct harm, but it does make it impossible for our program to get its work done. Such an attack is called a denial-of-service (DoS ) attack. Fortunately, the sticky bit comes to the rescue. Because the sticky bit is set on /tmp, nobody else can remove files from that directory. Of course, root can still remove files from /tmp, but if the attacker has root privilege, there’s nothing you can do to protect your program. If you choose to assume competent system administration, then /tmp will not be mounted via NFS. And if the system administrator was foolish enough to mount /tmp over NFS, then there’s a good chance that the sticky bit isn’t set, either. So, for most practical purposes, we think it’s safe to use mkstemp. But you should be aware of these issues, and you should definitely not rely on O_EXCL to work correctly if the directory in use is not /tmp—nor you should rely on the sticky bit being set anywhere else.

10.6.3 Using system or popen
The third common security hole that every programmer should bear in mind involves using the shell to execute other programs. As a toy example, let’s consider a dictionary server.This program is designed to accept connections via the Internet. Each client sends a word, and the server tells it whether that is a valid English word. Because every GNU/Linux system comes with a list of about 45,000 English words in /usr/dict/words, an easy way to build this server is to invoke the grep program, like this:
% grep -x word /usr/dict/words

Here, word is the word that the user is curious about.The exit code from grep will tell you whether that word appears in /usr/dict/words.10 Listing 10.6 shows how you might try to code the part of the server that invokes grep:
Listing 10.6 ( grep-dictionary.c) Search for a Word in the Dictionary
#include #include /* Returns a nonzero value if and only if the WORD appears in /usr/dict/words. */ int grep_for_word (const char* word) { size_t length; char* buffer; int exit_code;

10. If you don’t know about useful program.

grep, you

should look at the manual pages. It’s an incredibly

10.6

More Security Holes 217

/* Build up the string ‘grep -x WORD /usr/dict/words’. Allocate the string dynamically to avoid buffer overruns. */ length = strlen (“grep -x “) + strlen (word) + strlen (“ /usr/dict/words”) + 1; buffer = (char*) malloc (length); sprintf (buffer, “grep -x %s /usr/dict/words”, word); /* Run the command. */ exit_code = system (buffer); /* Free the buffer. */ free (buffer); /* If grep returned 0, then the word was present in the dictionary. */ return exit_code == 0; }

Note that by calculating the number of characters we need and then allocating the buffer dynamically, we’re sure to be safe from buffer overruns. Unfortunately, the use of the system function (described in Chapter 3, “Processes,” Section 3.2.1, “Using system”) is unsafe.This function invokes the standard system shell to run the command and then returns the exit value. But what happens if a malicious hacker sends a “word” that is actually the following line or a similar string? foo /dev/null; rm -rf /

In that case, the server will execute this command: grep -x foo /dev/null; rm -rf / /usr/dict/words

Now the problem is obvious.The user has turned one command, ostensibly the invocation of grep, into two commands because the shell treats a semicolon as a command separator.The first command is still a harmless invocation of grep, but the second removes all files on the entire system! Even if the server is not running as root, all the files that can be removed by the user running the server will be removed.The same problem can arise with popen (described in Section 5.4.4, “popen and pclose”), which creates a pipe between the parent and child process but still uses the shell to run the command. There are two ways to avoid these problems. One is to use the exec family of functions instead of system or popen.That solution avoids the problem because characters that the shell treats specially (such as the semicolon in the previous command) are not treated specially when they appear in the argument list to an exec call. Of course, you give up the convenience of system and popen.

218 Chapter 10 Security

The other alternative is to validate the string to make sure that it is benign. In the dictionary server example, you would make sure that the word provided contains only alphabetic characters, using the isalpha function. If it doesn’t contain any other characters, there’s no way to trick the shell into executing a second command. Don’t implement the check by looking for dangerous and unexpected characters; it’s always safer to explicitly check for the characters that you know are safe rather than try to anticipate all the characters that might cause trouble.

11
A Sample GNU/Linux Application

T

HIS CHAPTER IS WHERE IT ALL COMES TOGETHER.WE’LL DESCRIBE and implement a complete GNU/Linux program that incorporates many of the techniques described in this book.The program provides information about the system it’s running on via a Web interface. The program is a complete demonstration of some of the methods we’ve described for GNU/Linux programming and illustrated in shorter programs.This program is written more like “real-world” code, unlike most of the code listings that we presented in previous chapters. It can serve as a jumping-off point for your own GNU/Linux programs.

11.1 Overview
The example program is part of a system for monitoring a running GNU/Linux system. It includes these features: The program incorporates a minimal Web server. Local or remote clients access system information by requesting Web pages from the server via HTTP. The program does not serve static HTML pages. Instead, the pages are generated on the fly by modules, each of which provides a page summarizing one aspect of the system’s state. n n

220 Chapter 11 A Sample GNU/Linux Application

n

n

n

Modules are not linked statically into the server executable. Instead, they are loaded dynamically from shared libraries. Modules can be added, removed, or replaced while the server is running. The server services each connection in a child process.This enables the server to remain responsive even when individual requests take a while to complete, and it shields the server from failures in modules. The server does not require superuser privilege to run (as long as it is not run on a privileged port). However, this limits the system information that it can collect.

We provide four sample modules that demonstrate how modules might be written. They further illustrate some of the techniques for gathering system information presented previously in this book.The time module demonstrates using the gettimeofday system call.The issue module demonstrates low-level I/O and the sendfile system call.The diskfree module demonstrates the use of fork, exec, and dup2 by running a command in a child process.The processes module demonstrates the use of the /proc file system and various system calls.

11.1.1 Caveats
This program has many of the features you’d expect in an application program, such as command-line parsing and error checking. At the same time, we’ve made some simplifications to improve readability and to focus on the GNU/Linux-specific topics discussed in this book. Bear in mind these caveats as you examine the code. We don’t attempt to provide a full implementation of HTTP. Instead, we implement just enough for the server to interact with Web clients. A real-world program either would provide a more complete HTTP implementation or would interface with one of the various excellent Web server implementations1 available instead of providing HTTP services directly. Similarly, we don’t aim for full compliance with HTML specifications (see http://www.w3.org/MarkUp/).We generate simple HTML output that can be handled by popular Web browsers. The server is not tuned for high performance or minimum resource usage. In particular, we intentionally omit some of the network configuration code that you would expect in a Web server.This topic is outside the scope of this book. See one of the many excellent references on network application development, such as UNIX Network Programming,Volume 1: Networking APIs—Sockets and XTI, by W. Richard Stevens (Prentice Hall, 1997), for more information. n n n

1.The most popular open source Web server for GNU/Linux is the Apache server, available from http://www.apache.org.

11.2

Implementation 221

n

n

We make no attempt to regulate the resources (number of processes, memory use, and so on) consumed by the server or its modules. Many multiprocess Web server implementations service connections using a fixed pool of processes rather than creating a new child process for each connection. The server loads the shared library for a server module each time it is requested and then immediately unloads it when the request has been completed. A more efficient implementation would probably cache loaded modules.

HTTP
The Hypertext Transport Protocol (HTTP ) is used for communication between Web clients and servers. The client connects to the server by establishing a connection to a well-known port (usually port 80 for Internet Web servers, but any port may be used). HTTP requests and headers are composed of plain text. Once connected, the client sends a request to the server. A typical request is GET /page HTTP/1.0. The GET method indicates that the client is requesting that the server send it a Web page. The second element is the path to that page on the server. The third element is the protocol and version. Subsequent lines contain header fields, formatted similarly to email headers, which contain extra information about the client. The header ends with a blank line. The server sends back a response indicating the result of processing the request. A typical response is
HTTP/1.0 200 OK. The first element is the protocol version. The next two elements indicate the

result; in this case, result 200 indicates that the request was processed successfully. Subsequent lines contain header fields, formatted similarly to email headers. The header ends with a blank line. The server may then send arbitrary data to satisfy the request. Typically, the server responds to a page request by sending back HTML source for the Web page. In this case, the response headers will include Content-type: text/html, indicating that the result is HTML source. The HTML source follows immediately after the header. See the HTTP specification at http://www.w3.org/Protocols/ for more information.

11.2 Implementation
All but the very smallest programs written in C require careful organization to preserve the modularity and maintainability of the source code.This program is divided into four main source files. Each source file exports functions or variables that may be accessed by the other parts of the program. For simplicity, all exported functions and variables are declared in a single header file, server.h (see Listing 11.1), which is included by the other files. Functions that are intended for use within a single compilation unit only are declared static and are not declared in server.h.

222 Chapter 11 A Sample GNU/Linux Application

Listing 11.1 (server.h) Function and Variable Declarations
#ifndef SERVER_H #define SERVER_H #include #include /*** Symbols defined in common.c. /* The name of this program. */ extern const char* program_name; /* If nonzero, print verbose messages. extern int verbose; */ ************************************/

/* Like malloc, except aborts the program if allocation fails. extern void* xmalloc (size_t size); /* Like realloc, except aborts the program if allocation fails. extern void* xrealloc (void* ptr, size_t size); /* Like strdup, except aborts the program if allocation fails. extern char* xstrdup (const char* s);

*/

*/

*/

/* Print an error message for a failed call OPERATION, using the value of errno, and end the program. */ extern void system_error (const char* operation); /* Print an error message for failure involving CAUSE, including a descriptive MESSAGE, and end the program. */ extern void error (const char* cause, const char* message); /* Return the directory containing the running program’s executable. The return value is a memory buffer that the caller must deallocate using free. This function calls abort on failure. */ extern char* get_self_executable_directory ();

/*** Symbols defined in module.c

**************************************/

/* An instance of a loaded server module. */ struct server_module { /* The shared library handle corresponding to the loaded module. void* handle; /* A name describing the module. */ const char* name; /* The function that generates the HTML results for this module. void (* generate_function) (int); };

*/

*/

11.2

Implementation 223

/* The directory from which modules are loaded. extern char* module_dir;

*/

/* Attempt to load a server module with the name MODULE_PATH. If a server module exists with this path, loads the module and returns a server_module structure representing it. Otherwise, returns NULL. */ extern struct server_module* module_open (const char* module_path); /* Close a server module and deallocate the MODULE object. extern void module_close (struct server_module* module); */

/*** Symbols defined in server.c.

************************************/

/* Run the server on LOCAL_ADDRESS and PORT. */ extern void server_run (struct in_addr local_address, uint16_t port);

#endif

/* SERVER_H */

11.2.1 common.c Common Functions

(see Listing 11.2) contains functions of general utility that are used throughout the program.
Listing 11.2 (common.c) General Utility Functions
#include #include #include #include #include

#include “server.h” const char* program_name; int verbose; void* xmalloc (size_t size) { void* ptr = malloc (size); /* Abort if the allocation failed. if (ptr == NULL) abort (); else return ptr; }

*/

continues

224 Chapter 11 A Sample GNU/Linux Application

Listing 11.2 Continued void* xrealloc (void* ptr, size_t size) { ptr = realloc (ptr, size); /* Abort if the allocation failed. */ if (ptr == NULL) abort (); else return ptr; } char* xstrdup (const char* s) { char* copy = strdup (s); /* Abort if the allocation failed. if (copy == NULL) abort (); else return copy; }

*/

void system_error (const char* operation) { /* Generate an error message for errno. error (operation, strerror (errno)); }

*/

void error (const char* cause, const char* message) { /* Print an error message to stderr. */ fprintf (stderr, “%s: error: (%s) %s\n”, program_name, cause, message); /* End the program. */ exit (1); } char* get_self_executable_directory () { int rval; char link_target[1024]; char* last_slash; size_t result_length; char* result; /* Read the target of the symbolic link /proc/self/exe. */ rval = readlink (“/proc/self/exe”, link_target, sizeof (link_target)); if (rval == -1) /* The call to readlink failed, so bail. */ abort (); else

11.2

Implementation 225

/* NUL-terminate the target. */ link_target[rval] = ‘\0’; /* We want to trim the name of the executable file, to obtain the directory that contains it. Find the rightmost slash. */ last_slash = strrchr (link_target, ‘/’); if (last_slash == NULL || last_slash == link_target) /* Something strange is going on. */ abort (); /* Allocate a buffer to hold the resulting path. */ result_length = last_slash - link_target; result = (char*) xmalloc (result_length + 1); /* Copy the result. */ strncpy (result, link_target, result_length); result[result_length] = ‘\0’; return result; }

You could use these functions in other programs as well; the contents of this file might be included in a common code library that is shared among many projects: xmalloc, xrealloc, and xstrdup are error-checking versions of the C library functions malloc, realloc, and strdup, respectively. Unlike the standard versions, which return a null pointer if the allocation fails, these functions immediately abort the program when insufficient memory is available. Early detection of memory allocation failure is a good idea. Otherwise, failed allocations introduce null pointers at unexpected places into the program. Because allocation failures are not easy to reproduce, debugging such problems can be difficult. Allocation failures are usually catastrophic, so aborting the program is often an acceptable course of action. The error function is for reporting a fatal program error. It prints a message to stderr and ends the program. For errors caused by failed system calls or library calls, system_error generates part of the error message from the value of errno (see Section 2.2.3, “Error Codes from System Calls,” in Chapter 2, “Writing Good GNU/Linux Software”). get_self_executable_directory determines the directory containing the executable file being run in the current process.The directory path can be used to locate other components of the program, which are installed in the same place at runtime.This function works by examining the symbolic link /proc/self/exe in the /proc file system (see Section 7.2.1, “/proc/self,” in Chapter 7, “The /proc File System”). n n n

In addition, common.c defines two useful global variables: The value of program_name is the name of the program being run, as specified in its argument list (see Section 2.1.1, “The Argument List,” in Chapter 2).When the program is invoked from the shell, this is the path and name of the program as the user entered it. n 226 Chapter 11 A Sample GNU/Linux Application

n

The variable verbose is nonzero if the program is running in verbose mode. In this case, various parts of the program print progress messages to stdout.

11.2.2 Loading Server Modules module.c (see Listing 11.3) provides the implementation of dynamically loadable server modules. A loaded server module is represented by an instance of struct server_module, which is defined in server.h.

Listing 11.3 (module.c) Server Module Loading and Unloading
#include #include #include #include

#include “server.h” char* module_dir; struct server_module* module_open (const char* module_name) { char* module_path; void* handle; void (* module_generate) (int); struct server_module* module; /* Construct the full path of the module shared library we’ll try to load. */ module_path = (char*) xmalloc (strlen (module_dir) + strlen (module_name) + 2); sprintf (module_path, “%s/%s”, module_dir, module_name); /* Attempt to open MODULE_PATH as a shared library. */ handle = dlopen (module_path, RTLD_NOW); free (module_path); if (handle == NULL) { /* Failed; either this path doesn’t exist or it isn’t a shared library. */ return NULL; } /* Resolve the module_generate symbol from the shared library. */ module_generate = (void (*) (int)) dlsym (handle, “module_generate”); /* Make sure the symbol was found. */ if (module_generate == NULL) {

11.2

Implementation 227

/* The symbol is missing. While this is a shared library, it probably isn’t a server module. Close up and indicate failure. dlclose (handle); return NULL; }

*/

/* Allocate and initialize a server_module object. */ module = (struct server_module*) xmalloc (sizeof (struct server_module)); module->handle = handle; module->name = xstrdup (module_name); module->generate_function = module_generate; /* Return it, indicating success. */ return module; } void module_close (struct server_module* module) { /* Close the shared library. */ dlclose (module->handle); /* Deallocate the module name. */ free ((char*) module->name); /* Deallocate the module object. */ free (module); }

Each module is a shared library file (see Section 2.3.2, “Shared Libraries,” in Chapter 2) and must define and export a function named module_generate.This function generates an HTML Web page and writes it to the client socket file descriptor passed as its argument. module.c contains two functions: module_open attempts to load a server module with a given name.The name normally ends with the .so extension because server modules are implemented as shared libraries.This function opens the shared library with dlopen and resolves a symbol named module_generate from the library with dlsym (see Section 2.3.6, “Dynamic Loading and Unloading,” in Chapter 2). If the library can’t be opened, or if module_generate isn’t a name exported by the library, the call fails and module_open returns a null pointer. Otherwise, it allocates and returns a module object. module_close closes the shared library corresponding to the server module and deallocates the struct server_module object. n n

module.c

which

module_open

also defines a global variable module_dir.This is the path of the directory in attempts to find shared libraries corresponding to server modules.

228 Chapter 11 A Sample GNU/Linux Application

11.2.3 The Server server.c (see Listing 11.4) is the implementation of the minimal HTTP server.

Listing 11.4 (server.c) Server Implementation
#include #include #include #include #include #include #include #include #include #include #include

#include “server.h” /* HTTP response and header for a successful request. static char* ok_response = “HTTP/1.0 200 OK\n” “Content-type: text/html\n” “\n”; /* HTTP response, header, and body, indicating that we didn’t understand the request. */ static char* bad_request_response = “HTTP/1.0 400 Bad Request\n” “Content-type: text/html\n” “\n” “\n” “ \n” “ Bad Request\n” “ This server did not understand your request.\n” “ \n” “\n”; /* HTTP response, header, and body template, indicating that the requested document was not found. */ static char* not_found_response_template = “HTTP/1.0 404 Not Found\n” “Content-type: text/html\n” “\n” “\n” “ \n” “ Not Found\n” */

11.2

Implementation 229

“ The requested URL %s was not found on this server.\n” “ \n” “\n”; /* HTTP response, header, and body template, indicating that the method was not understood. */ static char* bad_method_response_template = “HTTP/1.0 501 Method Not Implemented\n” “Content-type: text/html\n” “\n” “\n” “ \n” “ Method Not Implemented\n” “ The method %s is not implemented by this server.\n” “ \n” “\n”; /* Handler for SIGCHLD, to clean up child processes that have terminated. */ static void clean_up_child_process (int signal_number) { int status; wait (&status); } /* Process an HTTP “GET” request for PAGE, and send the results to the file descriptor CONNECTION_FD. */ static void handle_get (int connection_fd, const char* page) { struct server_module* module = NULL; /* Make sure the requested page begins with a slash and does not contain any additional slashes -- we don’t support any subdirectories. */ if (*page == ‘/’ && strchr (page + 1, ‘/’) == NULL) { char module_file_name[64]; /* The page name looks OK. Construct the module name by appending “.so” to the page name. */ snprintf (module_file_name, sizeof (module_file_name), “%s.so”, page + 1); /* Try to open the module. */ module = module_open (module_file_name); } if (module == NULL) { /* Either the requested page was malformed, or we couldn’t open a module with the indicated name. Either way, return the HTTP response 404, Not Found. */

continues

230 Chapter 11 A Sample GNU/Linux Application

Listing 11.4 Continued char response[1024]; /* Generate the response message. */ snprintf (response, sizeof (response), not_found_response_template, page); /* Send it to the client. */ write (connection_fd, response, strlen (response)); } else { /* The requested module was loaded successfully.

*/

/* Send the HTTP response indicating success, and the HTTP header for an HTML page. */ write (connection_fd, ok_response, strlen (ok_response)); /* Invoke the module, which will generate HTML output and send it to the client file descriptor. */ (*module->generate_function) (connection_fd); /* We’re done with the module. */ module_close (module); } } /* Handle a client connection on the file descriptor CONNECTION_FD. static void handle_connection (int connection_fd) { char buffer[256]; ssize_t bytes_read; /* Read some data from the client. */ bytes_read = read (connection_fd, buffer, sizeof (buffer) - 1); if (bytes_read > 0) { char method[sizeof (buffer)]; char url[sizeof (buffer)]; char protocol[sizeof (buffer)]; /* Some data was read successfully. NUL-terminate the buffer so we can use string operations on it. */ buffer[bytes_read] = ‘\0’; /* The first line the client sends is the HTTP request, which is composed of a method, the requested page, and the protocol version. */ sscanf (buffer, “%s %s %s”, method, url, protocol); /* The client may send various header information following the request. For this HTTP implementation, we don’t care about it. However, we need to read any data the client tries to send. Keep on reading data until we get to the end of the header, which is delimited by a blank line. HTTP specifies CR/LF as the line delimiter. */ while (strstr (buffer, “\r\n\r\n”) == NULL) */

11.2

Implementation 231

bytes_read = read (connection_fd, buffer, sizeof (buffer)); /* Make sure the last read didn’t fail. If it did, there’s a problem with the connection, so give up. */ if (bytes_read == -1) { close (connection_fd); return; } /* Check the protocol field. We understand HTTP versions 1.0 and 1.1. */ if (strcmp (protocol, “HTTP/1.0”) && strcmp (protocol, “HTTP/1.1”)) { /* We don’t understand this protocol. Report a bad response. */ write (connection_fd, bad_request_response, sizeof (bad_request_response)); } else if (strcmp (method, “GET”)) { /* This server only implements the GET method. The client specified some other method, so report the failure. */ char response[1024]; snprintf (response, sizeof (response), bad_method_response_template, method); write (connection_fd, response, strlen (response)); } else /* A valid request. Process it. handle_get (connection_fd, url);

*/

} else if (bytes_read == 0) /* The client closed the connection before sending any data. Nothing to do. */ ; else /* The call to read failed. */ system_error (“read”); }

void server_run (struct in_addr local_address, uint16_t port) { struct sockaddr_in socket_address; int rval; struct sigaction sigchld_action; int server_socket; /* Install a handler for SIGCHLD that cleans up child processes that have terminated. */ memset (&sigchld_action, 0, sizeof (sigchld_action)); sigchld_action.sa_handler = &clean_up_child_process; sigaction (SIGCHLD, &sigchld_action, NULL);

continues

232 Chapter 11 A Sample GNU/Linux Application

Listing 11.4 Continued
/* Create a TCP socket. */ server_socket = socket (PF_INET, SOCK_STREAM, 0); if (server_socket == -1) system_error (“socket”); /* Construct a socket address structure for the local address on which we want to listen for connections. */ memset (&socket_address, 0, sizeof (socket_address)); socket_address.sin_family = AF_INET; socket_address.sin_port = port; socket_address.sin_addr = local_address; /* Bind the socket to that address. */ rval = bind (server_socket, &socket_address, sizeof (socket_address)); if (rval != 0) system_error (“bind”); /* Instruct the socket to accept connections. */ rval = listen (server_socket, 10); if (rval != 0) system_error (“listen”); if (verbose) { /* In verbose mode, display the local address and port number we’re listening on. */ socklen_t address_length; /* Find the socket’s local address. */ address_length = sizeof (socket_address); rval = getsockname (server_socket, &socket_address, &address_length); assert (rval == 0); /* Print a message. The port number needs to be converted from network byte order (big endian) to host byte order. */ printf (“server listening on %s:%d\n”, inet_ntoa (socket_address.sin_addr), (int) ntohs (socket_address.sin_port)); } /* Loop forever, handling connections. while (1) { struct sockaddr_in remote_address; socklen_t address_length; int connection; pid_t child_pid; */

/* Accept a connection. This call blocks until a connection is ready. */ address_length = sizeof (remote_address); connection = accept (server_socket, &remote_address, &address_length); if (connection == -1) { /* The call to accept failed. */ if (errno == EINTR)

11.2

Implementation 233

/* The call was interrupted by a signal. continue; else /* Something else went wrong. */ system_error (“accept”); }

Try again.

*/

/* We have a connection. Print a message if we’re running in verbose mode. */ if (verbose) { socklen_t address_length; /* Get the remote address of the connection. */ address_length = sizeof (socket_address); rval = getpeername (connection, &socket_address, &address_length); assert (rval == 0); /* Print a message. */ printf (“connection accepted from %s\n”, inet_ntoa (socket_address.sin_addr)); } /* Fork a child process to handle the connection. */ child_pid = fork (); if (child_pid == 0) { /* This is the child process. It shouldn’t use stdin or stdout, so close them. */ close (STDIN_FILENO); close (STDOUT_FILENO); /* Also this child process shouldn’t do anything with the listening socket. */ close (server_socket); /* Handle a request from the connection. We have our own copy of the connected socket descriptor. */ handle_connection (connection); /* All done; close the connection socket, and end the child process. */ close (connection); exit (0); } else if (child_pid > 0) { /* This is the parent process. The child process handles the connection, so we don’t need our copy of the connected socket descriptor. Close it. Then continue with the loop and accept another connection. */ close (connection); } else /* Call to fork failed. */ system_error (“fork”); } }

234 Chapter 11 A Sample GNU/Linux Application

These are the functions in server.c: server_run is the main entry point for running the server.This function starts the server and begins accepting connections, and does not return unless a serious error occurs.The server uses a TCP stream server socket (see Section 5.5.3, “Servers,” in Chapter 5, “Interprocess Communication”). The first argument to server_run specifies the local address at which connections are accepted. A GNU/Linux computer may have multiple network addresses, and each address may be bound to a different network interface.2 To restrict the server to accept connections from a particular interface, specify the corresponding network address. Specify the local address INADDR_ANY to accept connections for any local address. The second argument to server_run is the port number on which to accept connections. If the port number is already in use, or if it corresponds to a privileged port and the server is not being run with superuser privilege, the server fails.The special value 0 instructs Linux to select an unused port automatically. See the inet man page for more information about Internet-domain addresses and port numbers. The server handles each client connection in a child process created with fork (see Section 3.2.2, “Using fork and exec,” in Chapter 3, “Processes”).The main (parent) process continues accepting new connections while existing ones are being serviced.The child process invokes handle_connection and then closes the connection socket and exits. handle_connection processes a single client connection, using the socket file descriptor passed as its argument.This function reads data from the socket and attempts to interpret this as an HTTP page request. The server processes only HTTP version 1.0 and version 1.1 requests.When faced with a different protocol or version, it responds by sending the HTTP result code 400 and the message bad_request_response.The server understands only the HTTP GET method. If the client requests any other method, the server responds by sending the HTTP result code 501 and the message bad_method_response_template. If the client sends a well-formed GET request, handle_connection calls handle_get to service it.This function attempts to load a server module with a name generated from the requested page. For example, if the client requests the page named information, it attempts to load a server module named information.so. If the module can’t be loaded, handle_get sends the client the HTTP result code 404 and the message not_found_response_template. n n n

lo, the

2.Your computer might be configured to include such interfaces as eth0, an Ethernet card; local (loopback) network; or ppp0, a dial-up network connection.

11.2

Implementation 235

n

If the client sends a page request that corresponds to a server module, handle_get sends a result code 200 header to the client, which indicates that the request was processed successfully and invokes the module’s module_generate function.This function generates the HTML source for a Web page and sends it to the Web client. server_run installs clean_up_child_process as the signal handler for SIGCHLD. This function simply cleans up terminated child processes (see Section 3.4.4, “Cleaning Up Children Asynchronously,” in Chapter 3).

11.2.4 The Main Program main.c (see Listing 11.5) provides the main function for the server program. Its responsibility is to parse command-line options, detect and report command-line errors, and configure and run the server.
Listing 11.5 (main.c) Main Server Program and Command-Line Parsing
#include #include #include #include #include #include #include #include

#include “server.h” /* Description of long options for getopt_long. static const struct option long_options[] = { { “address”, 1, NULL, ‘a’ }, { “help”, 0, NULL, ‘h’ }, { “module-dir”, 1, NULL, ‘m’ }, { “port”, 1, NULL, ‘p’ }, { “verbose”, 0, NULL, ‘v’ }, }; /* Description of short options for getopt_long. */ */

static const char* const short_options = “a:hm:p:v”; /* Usage summary text. */

static const char* const usage_template = “Usage: %s [ options ]\n” “ -a, --address ADDR Bind to local address (by default, bind\n” “ to all local addresses).\n”

continues

236 Chapter 11 A Sample GNU/Linux Application

Listing 11.5 Continued
“ “ “ “ “ -h, --help -m, --module-dir DIR -p, --port PORT -v, --verbose Print this information.\n” Load modules from specified directory\n” (by default, use executable directory).\n” Bind to specified port.\n” Print verbose messages.\n”;

/* Print usage information and exit. If IS_ERROR is nonzero, write to stderr and use an error exit code. Otherwise, write to stdout and use a non-error termination code. Does not return. */ static void print_usage (int is_error) { fprintf (is_error ? stderr : stdout, usage_template, program_name); exit (is_error ? 1 : 0); } int main (int argc, char* const argv[]) { struct in_addr local_address; uint16_t port; int next_option; /* Store the program name, which we’ll use in error messages. program_name = argv[0]; */

/* Set defaults for options. Bind the server to all local addresses, and assign an unused port automatically. */ local_address.s_addr = INADDR_ANY; port = 0; /* Don’t print verbose messages. */ verbose = 0; /* Load modules from the directory containing this executable. */ module_dir = get_self_executable_directory (); assert (module_dir != NULL); /* Parse options. */ do { next_option = getopt_long (argc, argv, short_options, long_options, NULL); switch (next_option) { case ‘a’: /* User specified -a or --address. */ { struct hostent* local_host_name; /* Look up the hostname the user specified. */ local_host_name = gethostbyname (optarg); if (local_host_name == NULL || local_host_name->h_length == 0)

11.2

Implementation 237

/* Could not resolve the name. */ error (optarg, “invalid host name”); else /* Hostname is OK, so use it. */ local_address.s_addr = *((int*) (local_host_name->h_addr_list[0])); } break; case ‘h’: /* User specified -h or --help. print_usage (0);

*/

case ‘m’: /* User specified -m or --module-dir. { struct stat dir_info;

*/

/* Check that it exists. */ if (access (optarg, F_OK) != 0) error (optarg, “module directory does not exist”); /* Check that it is accessible. */ if (access (optarg, R_OK | X_OK) != 0) error (optarg, “module directory is not accessible”); /* Make sure that it is a directory. */ if (stat (optarg, &dir_info) != 0 || !S_ISDIR (dir_info.st_mode)) error (optarg, “not a directory”); /* It looks OK, so use it. */ module_dir = strdup (optarg); } break; case ‘p’: /* User specified -p or --port. { long value; char* end;

*/

value = strtol (optarg, &end, 10); if (*end != ‘\0’) /* The user specified nondigits in the port number. */ print_usage (1); /* The port number needs to be converted to network (big endian) byte order. */ port = (uint16_t) htons (value); } break; case ‘v’: /* User specified -v or --verbose. verbose = 1; break;

*/

continues

238 Chapter 11 A Sample GNU/Linux Application

Listing 11.5 Continued case ‘?’: /* User specified an unrecognized option. print_usage (1); case -1: /* Done with options. break; */

*/

default: abort (); } } while (next_option != -1); /* This program takes no additional arguments. user specified any. */ if (optind != argc) print_usage (1); Issue an error if the

/* Print the module directory, if we’re running verbose. */ if (verbose) printf (“modules will be loaded from %s\n”, module_dir); /* Run the server. */ server_run (local_address, port); return 0; }

main.c n contains these functions: main invokes getopt_long (see Section 2.1.3, “Using getopt_long,” in Chapter 2) to parse command-line options. It provides both long and short option forms, the former in the long_options array and the latter in the short_options string. The default value for the server port is 0 and for a local address is INADDR_ANY. These can be overridden by the --port (-p) and --address (-a) options, respectively. If the user specifies an address, main calls the library function 3 gethostbyname to convert it to a numerical Internet address. The default value for the directory from which to load server modules is the directory containing the server executable, as determined by get_self_executable_directory.The user may override this with the --module-dir (-m) option; main makes sure that the specified directory is accessible. By default, verbose messages are not printed.The user may enable them by specifying the --verbose (-v) option.

3. gethostbyname performs name resolution using DNS, if necessary.

11.3

Modules 239

n

If the user specifies the --help (-h) option or specifies invalid options, main invokes print_usage, which prints a usage summary and exits.

11.3 Modules
We provide four modules to demonstrate the kind of functionality you could implement using this server implementation. Implementing your own server module is as simple as defining a module_generate function to return the appropriate HTML text.

11.3.1 Show Wall-Clock Time
The time.so module (see Listing 11.6) generates a simple page containing the server’s local wall-clock time.This module’s module_generate calls gettimeofday to obtain the current time (see Section 8.7, “gettimeofday:Wall-Clock Time,” in Chapter 8, “Linux System Calls”) and uses localtime and strftime to generate a text representation of it.This representation is embedded in the HTML template page_template.
Listing 11.6 (time.c) Server Module to Show Wall-Clock Time
#include #include #include #include

#include “server.h” /* A template for the HTML page this module generates. static char* page_template = “\n” “ \n” “ \n” “ \n” “ \n” “ The current time is %s.\n” “ \n” “\n”; void module_generate (int fd) { struct timeval tv; struct tm* ptm; char time_string[40]; FILE* fp; /* Obtain the time of day, and convert it to a tm struct. gettimeofday (&tv, NULL); ptm = localtime (&tv.tv_sec); */ */

continues

240 Chapter 11 A Sample GNU/Linux Application

Listing 11.6 Continued
/* Format the date and time, down to a single second. */ strftime (time_string, sizeof (time_string), “%H:%M:%S”, ptm); /* Create a stream corresponding to the client socket file descriptor. */ fp = fdopen (fd, “w”); assert (fp != NULL); /* Generate the HTML output. */ fprintf (fp, page_template, time_string); /* All done; flush the stream. */ fflush (fp); }

This module uses standard C library I/O routines for convenience.The fdopen call generates a stream pointer (FILE*) corresponding to the client socket file descriptor (see Section B.4, “Relation to Standard C Library I/O Functions,” in Appendix B, “Low-Level I/O”).The module writes to it using fprintf and flushes it using fflush to prevent the loss of buffered data when the socket is closed. The HTML page returned by the time.so module includes a element in the page header that instructs clients to reload the page every 5 seconds.This way the client displays the current time.

11.3.2 Show the GNU/Linux Distribution
The issue.so module (see Listing 11.7) displays information about the GNU/Linux distribution running on the server.This information is traditionally stored in the file /etc/issue.This module sends the contents of this file, wrapped in a element of an HTML page.
Listing 11.7 (issue.c) Server Module to Display GNU/Linux Distribution Information
#include #include #include #include #include #include

#include “server.h” /* HTML source for the start of the page we generate. static char* page_start = “\n” “ \n” */

11.3

Modules 241

“

\n”; */

/* HTML source for the end of the page we generate. static char* page_end = “ \n” “ \n” “\n”;

/* HTML source for the page indicating there was a problem opening /proc/issue. */ static char* error_page = “\n” “ \n” “ Error: Could not open /proc/issue.\n” “ \n” “\n”; /* HTML source indicating an error. */

static char* error_message = “Error reading /proc/issue.”; void module_generate (int fd) { int input_fd; struct stat file_info; int rval; /* Open /etc/issue. */ input_fd = open (“/etc/issue”, O_RDONLY); if (input_fd == -1) system_error (“open”); /* Obtain file information about it. */ rval = fstat (input_fd, &file_info); if (rval == -1) /* Either we couldn’t open the file or we couldn’t read from it. write (fd, error_page, strlen (error_page)); else { int rval; off_t offset = 0; /* Write the start of the page. */ write (fd, page_start, strlen (page_start)); /* Copy from /proc/issue to the client socket. */ rval = sendfile (fd, input_fd, &offset, file_info.st_size); if (rval == -1) /* Something went wrong sending the contents of /proc/issue. Write an error message. */ write (fd, error_message, strlen (error_message));

*/

continues

242 Chapter 11 A Sample GNU/Linux Application

Listing 11.7 Continued
/* End the page. */ write (fd, page_end, strlen (page_end)); } close (input_fd); }

The module first tries to open /etc/issue. If that file can’t be opened, the module sends an error page to the client. Otherwise, the module sends the start of the HTML page, contained in page_start.Then it sends the contents of /etc/issue using sendfile (see Section 8.12, “sendfile: Fast Data Transfers,” in Chapter 8). Finally, it sends the end of the HTML page, contained in page_end. You can easily adapt this module to send the contents of another file. If the file contains a complete HTML page, simply omit the code that sends the contents of page_start and page_end.You could also adapt the main server implementation to serve static files, in the manner of a traditional Web server. Using sendfile provides an extra degree of efficiency.

11.3.3 Show Free Disk Space
The diskfree.so module (see Listing 11.8) generates a page displaying information about free disk space on the file systems mounted on the server computer.This generated information is simply the output of invoking the df -h command. Like issue.so, this module wraps the output in a element of an HTML page.
Listing 11.8 (diskfree.c) Server Module to Display Information About Free Disk Space
#include #include #include #include #include

#include “server.h” /* HTML source for the start of the page we generate. static char* page_start = “\n” “ \n” “ \n”; */

11.3

Modules 243

/* HTML source for the end of the page we generate. static char* page_end = “ \n” “ \n” “\n”; void module_generate (int fd) { pid_t child_pid; int rval;

*/

/* Write the start of the page. */ write (fd, page_start, strlen (page_start)); /* Fork a child process. */ child_pid = fork (); if (child_pid == 0) { /* This is the child process. */ /* Set up an argument list for the invocation of df. char* argv[] = { “/bin/df”, “-h”, NULL };

*/

/* Duplicate stdout and stderr to send data to the client socket. rval = dup2 (fd, STDOUT_FILENO); if (rval == -1) system_error (“dup2”); rval = dup2 (fd, STDERR_FILENO); if (rval == -1) system_error (“dup2”); /* Run df to show the free space on mounted file systems. */ execv (argv[0], argv); /* A call to execv does not return unless an error occurred. */ system_error (“execv”); } else if (child_pid > 0) { /* This is the parent process. Wait for the child process to finish. */ rval = waitpid (child_pid, NULL, 0); if (rval == -1) system_error (“waitpid”); } else /* The call to fork failed. */ system_error (“fork”); /* Write the end of the page. */ write (fd, page_end, strlen (page_end)); }

*/

244 Chapter 11 A Sample GNU/Linux Application

While issue.so sends the contents of a file using sendfile, this module must invoke a command and redirect its output to the client.To do this, the module follows these steps: 1. First, the module creates a child process using fork (see Section 3.2.2, “Using fork and exec,” in Chapter 3). 2. The child process copies the client socket file descriptor to file descriptors STDOUT_FILENO and STDERR_FILENO, which correspond to standard output and standard error (see Section 2.1.4, “Standard I/O,” in Chapter 2).The file descriptors are copied using the dup2 call (see Section 5.4.3, “Redirecting the Standard Input, Output, and Error Streams,” in Chapter 5). All further output from the process to either of these streams is sent to the client socket. 3. The child process invokes the df command with the -h option by calling execv (see Section 3.2.2, “Using fork and exec,” in Chapter 3). 4. The parent process waits for the child process to exit by calling waitpid (see Section 3.4.2, “The wait System Calls,” in Chapter 3). You could easily adapt this module to invoke a different command and redirect its output to the client.

11.3.4 Summarize Running Processes
The processes.so module (see Listing 11.9) is a more extensive server module implementation. It generates a page containing a table that summarizes the processes currently running on the server system. Each process is represented by a row in the table that lists the PID, the executable program name, the owning user and group names, and the resident set size.
Listing 11.9 ( processes.c) Server Module to Summarize Processes
#include #include #include #include #include #include #include #include #include #include #include #include

#include “server.h” /* Set *UID and *GID to the owning user ID and group ID, respectively, of process PID. Return 0 on success, nonzero on failure. */

11.3

Modules 245

static int get_uid_gid (pid_t pid, uid_t* uid, gid_t* gid) { char dir_name[64]; struct stat dir_info; int rval; /* Generate the name of the process’s directory in /proc. */ snprintf (dir_name, sizeof (dir_name), “/proc/%d”, (int) pid); /* Obtain information about the directory. */ rval = stat (dir_name, &dir_info); if (rval != 0) /* Couldn’t find it; perhaps this process no longer exists. */ return 1; /* Make sure it’s a directory; anything else is unexpected. */ assert (S_ISDIR (dir_info.st_mode)); /* Extract the IDs we want. *uid = dir_info.st_uid; *gid = dir_info.st_gid; return 0; } /* Return the name of user UID. The return value is a buffer that the caller must allocate with free. UID must be a valid user ID. */ static char* get_user_name (uid_t uid) { struct passwd* entry; entry = getpwuid (uid); if (entry == NULL) system_error (“getpwuid”); return xstrdup (entry->pw_name); } /* Return the name of group GID. The return value is a buffer that the caller must allocate with free. GID must be a valid group ID. */ static char* get_group_name (gid_t gid) { struct group* entry; entry = getgrgid (gid); if (entry == NULL) system_error (“getgrgid”); return xstrdup (entry->gr_name); } */

continues

246 Chapter 11 A Sample GNU/Linux Application

Listing 11.9 Continued
/* Return the name of the program running in process PID, or NULL on error. The return value is a newly allocated buffer which the caller must deallocate with free. */ static char* get_program_name (pid_t pid) { char file_name[64]; char status_info[256]; int fd; int rval; char* open_paren; char* close_paren; char* result; /* Generate the name of the “stat” file in the process’s /proc directory, and open it. */ snprintf (file_name, sizeof (file_name), “/proc/%d/stat”, (int) pid); fd = open (file_name, O_RDONLY); if (fd == -1) /* Couldn’t open the stat file for this process. Perhaps the process no longer exists. */ return NULL; /* Read the contents. */ rval = read (fd, status_info, sizeof (status_info) - 1); close (fd); if (rval

Similar Documents

Test Taking

Test

Test

Test

Test

Test

Standardized Tests

Standardized Tests

Standardized Test

Test

Pre And Post-Test Assessment

Eco 410 Test Bank

Eco 410 Test Bank

Standardized Test Outline

Standardization Test

Popular Essays