Programming Project

1
Programming Project 2Linux Kernel Hacking

• CS-502 Operating SystemsFall 2006

2
Objective

• To learn how to work with an operating system
  kernel
• To understand some of the constraints and
  techniques of programming in a kernel (versus
  user space)

3
Method

• To build and install a new Linux kernel
• To add a new system call to the Linux kernel
• To get useful information from the data
  structures of a Linux kernel

4
Background User vs. Kernel mode

• Hardware provides two modes
• Indicated by bit in PSW
• Allows OS to protect itself system components
  against
• Faulty and malicious processes
• Some instructions designated as privileged
• Only executable in kernel mode
• System call, all traps, interrupts change mode
  to kernel
• return from system call resets it to user

5
Transition from User to Kernel Mode
6
Accessing the Kernel via System Call

• Normally embedded within a library routine
• User API never uses system calls directly
• System call mechanism is machine specific
• Different CPU architectures make system calls in
  different ways
• System call numbers different for various
  architectures
• Even for same operating system version!
• E.g., poll system call is 167 on PowerPC but
  168 on Intel 386 platforms (in SUSE Linux 9.3)

7
Inside Kernel, the OS can

• Read and modify data structures not in user
  address space
• Control devices forbidden to user processes
• Invoke operating system functions not available
  to user processes

8
In this project, we will

• Add a new system call to the Linux kernel
• It does nothing except announce its presence
• Upgrade that system call to provide information
  about the process
• Information not readily available via existing
  system calls
• Follow Linux naming numbering conventions

9
In this project, we wont

• bother to make a library routine to encapsulate
  our systems calls
• try to support them on all machine architectures

10
To get started

• Find kernel sources in /usr/src
• linux-2.6.11.4-20alinux-2.6.11.4-21.13linux-2.6
  .11.4-21.14
• Clone kernel source tree by linked copy
• cp al /usr/src/linux-2.6.11.4-20a myKernel
• Creates a linked copy of original tree in new
  directory called myKernel

11
Digression on Unix Hard Links

• Directory entries point to files
• Two entries may point to same file!
• Same or different directories
• Same or different name
• Permissions attached to file, not directory
• Called hard links (as opposed to symbolic links)
• Modifications to file seen via all hard links
• mv and rm commands change directories, not files!
• File goes away when all directory entries (i.e.,
  hard links) to that file are deleted.

12
Cloned Source Tree

• All directories are copied
• All directory entries in copy are linked back to
  files in original source tree
• To modify a file
• Unlink it (via mv or rm)
• Replace it with modified copy in that directory
• Original is preserved intact in other directory
• Note Versions of Linux kernel sources are linked
  copies of each other

13
To Modify a File

• mv file.c file.c
• Edit file.c in your favorite editor
• Save as file.c
• (Eventually) delete file.c
• EMACS and patch do this automatically
• Most other editors require you to do it manually

14
Part 1 Build Linux Kernel

• Make configuration
• As ordinary user
• Build source tree
• As ordinary user
• Install
• Needs root privileges

15
Make configuration

• In a command shell at top of your cloned
  directory, do one of
• make config
• Very long and tedious
• make menuconfig
• Still somewhat long and tedious also hokey
• make xconfig
• Nice
• make gconfig
• Really nice, but our virtual machines are missing
  some libraries to support this

16
Make configuration (continued)

• Edit General Setup to name your Local version
• No other edits necessary at this time.
• Save and quit
• If need to rebuild, use
• make oldconfig
• to reuse same configuration

17
To Build Kernel

• make gt make-output.txt
• Execute at top of your kernel tree
• Takes
• 40 minutes on csopt4
• one hour on 3 gigahertz Pentium
• Rebuilds after small edits are much faster
• Changing .h files can cause longer rebuilds

18
To Install Kernel

• Requires root privileges
• sudo make install modules_install
• Puts kernel, initrd file, symbols in /boot
• Changes links for vmlinuz, initrd
• Adds entries to /boot/grub/menu.lst
• So you can select which kernel to boot
• Some apparent bugs in this process
• Use YaST to repair manually

19
Running Your Kernel

• Restart your virtual machine
• Click in boot window
• Use arrow keys to select desired kernel or system
• To determine which kernel is running
• uname -a

20
Creating a Patch File

• In kernel tree, remove all object config files
  by
• make distclean
• One level above kernel tree, do
• diff urN original myKernel gt patch1
• To recreate your directory from patch
• cp al original newKernel
• cd newKernel
• patch p1 lt patch1

21
End of Part 1

• Questions?

22
Part 2 Adding a System Call

• See Silbershatz, pp 74-78
• Similar problem statement
• Many details are different (due to version of
  Linux)
• Clone kernel tree from Part1
• Use make oldconfig before building

23
Linux Conventions (all versions)

• If your library routine is alarm,
• then the corresponding system call is sys_alarm
• and the corresponding function prototype for
  its implementation is
• asmlinkage unsigned long sys_alarm (unsigned int
  seconds)
• Note that asmlinkage is a compiler directive that
  tells how to compile the function call
• in gcc

24
Linux Conventions (continued)

• To invoke alarm system call from a user or
  library routine, use macro
• _syscall1(unsigned long, alarm, unsigned int
  seconds)
• _syscalln has n2 arguments
• Return type
• Name of actual system call (in user space)
• Arguments to system call function
• This macro defines the function
• unsigned long alarm(unsigned int seconds)

25
helloworld System Call

• / This is the text of the helloworld system call
  /
• asmlinkage int sys_helloworld(void)
  printk(KERN_EMERG Hello, world!) return
  0
• Add to the file kernel/sys.c

26
printk(), the Kernel Debug Print Tool

• Very robust
• May be called from (almost) anywhere in kernel
• Same calling convention as printf()
• Writes to system log
• Output survives crashes (almost all of the time)
• To read output, see
• /var/log/messages
• Needs root privileges to read
• Circular log, newest messages at end
• See Linux Kernel Development, 2nd edition, by
  Robert Love, Chapter 18.

27
Registering your System Call

• include/asm-i386/unistd.h
• Add entry for your call number
• Increment total number of calls
• arch/i386/kernel/entry.S
• Lists entry points for system calls
• Must be kept in numerical order!
• Number must correspond to entry in unistd.h
• Rebuild and install your kernel

28
Testing your System Call

• include ltlinux/errno.hgtinclude
  ltsys/syscall.hgtinclude ltlinux/unistd.hgtinclude
  ltstdio.hgtdefine __NR_helloworld 288 / or
  whatever you set it in unistd.h
  /_syscall0(int, helloworld)main ()
  printf(The return code from the helloworld
  system call is d\n, helloworld())
• Check log for the message!

29
Create Patch File

• patch2 is difference between kernel tree for
  Part1 and kernel tree for Part2

30
End of Part 2

• Questions?

31
Part 3 Get Process Information

• Modify you kernel of Part 2 to add system call to
  get information about process
• System call is
• int getprinfo(struct prinfo info)
• info is pointer to area to store results
• Returns zero if successful, error code if not
• See handout for definition of struct prinfo

32
Information needed for prinfo

• See task_struct in include/linux/sched.h
• See getuid and getpid for examples of simple
  system calls
• See include/asm/current.h to find current process
  information
• Use copy_to_user to safely copy data from kernel
  to user space
• Return EFAULT error code if info argument is not
  valid pointer in user space

33
copy_to_user and copy_from_user

• Functions to safely copy data to/from user space
• Check validity of pointer arguments
• Return zero if successful, number of bytes that
  fail if there is a problem
• Immune to page faults, pre-emption, etc.

34
Implementing getprinfo System Call

• Replace helloworld system call from Part 2
• Implement in kernel/timer.c
• Register in unistd.h and entry.S
• Use printk() to print debugging statements to
  system log

35
Testing getprinfo

• Write test program in user space
• Run multiple times from same shell, different
  shell
• Note differences in results
• Compare with what you can find about processes
  from ps command

36
Create patch3

• Patch3 is difference between Part 2 Part 3.

37
Submission

• Submit using web-based turnin program
• http//turnin.cs.wpi.edu8088/servlets/turnin/turn
  in.ss
• Include
• patch1, patch2, and patch3
• Write up explaining results of testing Part 3
• Starting point for your kernel tree
• Put your name on all documents and at top of
  every edited file!

38
Due Dates

• This project is due at start of class on Monday,
  October 16.
• Pace yourself
• Part 1 should be complete by October 2
• Part 2 should be complete by October 9
• Part 3 should be complete by October 16
• Report to instructor any difficulties

39
Questions?