[ Curiosity,Experimentation ]

Random stuff from the parallel universe of Ones and Zeroes

Posts Tagged ‘gcc’

Writing a Linux character Device Driver

Posted by appusajeev on June 18, 2011

In this post, we would be writing a Linux device driver for a hypothetical character device which reverses any string that is given to it. i.e.  If we write any string to the device file represented by the device and then read that file, we get the string written earlier but reversed (for eg.,  myDev being our device, echo “hello” >/dev/myDev ; cat /dev/ myDev would print “olleh”).
We will be simulating the functionality of the device in the driver itself (and this is precisely what is done in emulation tools like Daemon Tools, Alcohol etc).

Download Code

We will be dealing with

  • Introduction and some basics
  • Creating a device file
  • The driver code
  • Compiling the driver
  • Loading and unloading the driver
  • Testing the driver

The post helps understand how to write a device driver, the significance of a device file and its role in the interaction between a user program and device driver.

Introduction

The devices in UNIX fall in two categories- Character devices and Block devices. Character devices can be compared to normal files in that we can read/write arbitrary bytes at a time (although for most part, seeking is not supported).They work with a stream of bytes. Block devices, on the other hand, operate on blocks of data, not arbitrary bytes. Usual block size is 512 bytes or larger powers of two. However, block devices can be accessed the same was as character devices, the driver does the block management. (Networking devices do not belong to these categories, the interface provided by these drivers in entirely different from that of char/block devices)

The beauty of UNIX is that devices are represented as files. Both character devices and block devices are represented by respective files in the /dev directory. This means that you can read and write into the device by manipulating those file using standard system calls like open, read, write, close etc.
For eg, you could directly write or read the hard disk by accessing /dev/sd* file – a dangerous act unless you know what you are doing (for those interested, try hexdump  –C  /dev/sda   –n  512 – what you see then is the boot sector of your hard disk !). As another example, you could directly see the contents of the RAM by reading /dev/mem.

Every device file represented in this manner is associated with the device driver of that device which is actually responsible for interacting with the device on behalf of the user request. So when you access a device file, the request is forwarded to the respective device driver which does the processing and returns the result.

For instance, you might be knowing about the files /dev/zero (an infinite source of zeroes), /dev/null (a data black hole), /dev/random ( a source of random numbers) etc. When you actually read these files, what happens is that a particular function in the device driver registered for the file is invoked which returns the respective data.

In our example, we will be developing a character device represented by the device file /dev/myDev.  The mechanisms for creating this file will be explained later.

Under the hood

Now how does Linux know which driver is associated with which file? For that, each device and its device file has associated with it, a unique Major number and a Minor number. No two devices have the same major number.  When a device file is opened, Linux examines its major number and forwards the call to the driver registered for that device.  Subsequent calls for read/write/close too are processed by the same driver. As far as kernel is concerned, only major number is important. Minor number is used to identify the specific device instance if the driver controls more than one device of a type.

To know the major, minor number of devices, use the ls – l command as shown below.

ls -l

ls -l

The starting ‘c’  means its a character device, 1 is the major number and 8 is the minor number.

A Linux driver is a Linux module which can be loaded and linked to the kernel at runtime. The driver operates in kernel space and becomes part of the kernel once loaded, the kernel being monolithic. It can then access the symbols exported by the kernel.

When the device driver module is loaded, the driver first registers itself as a driver for a particular device specifying a particular Major number.

It uses the call register_chrdev function for registration. The call takes the Major number, Minor number, device name and an address of a structure of the type file_operations(discussed later) as argument. In our example, we will be using a major number of 89 . The choice of major number is arbitrary but it has to be unique on the system.

The syntax of register_chrdev is

int register_chrdev(unsigned int major,const char *name,struct file_operations *fops)

Driver is unregistered by calling the unregister_chrdev function.

Since device driver is a kernel module, it should implement init_module and cleanup_module functions. The register_chrdev call is done in the init_module function  and unregister_chrdev call is done in the cleanup_module function.

The register_chrdev call returns a non-negative number on success. If we specify the Major number as 0, the kernel returns a Major number unique at that instant which can be used to create a device file.
A device file can be created either before the driver is loaded if we know the major and minor number beforehand or it can be created later after letting the driver specify a major number for us.

Apart from those, the driver must also define certain callback functions that would be invoked when file operations are done on the device file. Ie. It must define functions that would be invoked by the kernel when a process uses open, read, write, close system calls on the file. Every driver must implement functions for processing these requests.
When register_chrdev call is done, the fourth argument is a structure that contains the addresses of these callback functions, callbacks for open, read, write, close system calls. The structure is of the type file_operations and has 4 main fields that should be set  – read,write,open and release. Each field must be assigned an address of a function that would be called when open, read,write , close system calls are called respectively.  For eg

file_operations structure initialisation

file_operations structure initialisation

It is important to note that all these callback functions have a predefined prototype although the name can be any.

Creating a device file

A device file is a special file. It can’t just be created using cat or gedit or shell redirection for that matter. The shell command mknod is usually used to create device file. The syntax of mknod is

mknod path type major minor

path:-path where the file to be created. It’s not necessary that the device file needs to be created in the /dev directory. It’s a mere convention. A device file can be created just about anywhere.

type: -‘c’ or ‘b’ . Whether the device being represented is a character device or a block device. In our example, we will be simulating a character device and hence we choose ‘c’.

major, minor:- the major and minor number of the device.

Heres how

mknod command

mknod command

chmod, though not necessary is done because, if not done, only processes will root permission can read or write to our device file.

The driver code

Given below is the code of the device driver

 Download Code

Device Driver Code

Device Driver Code

For debugging, I have included some printk messages in the code. To see those messages while the driver is in action, do dmesg|tail

Compiling the driver

A Linux module cannot just be compiled the way we compile normal C files.  cc filename.c won’t work. Compiling a Linux module is a separate process by its own. We use the help of kernel Makefile for compilation. The makefile we will be using is.

makefile for module compilation

makefile for module compilation

Here, we are making use of the kbuild mechanism used for compiling the kernel.
The result of compilation is a ko file (kernel object) which can then be loaded dynamically when required.

Loading and Unloading the Driver

Once the compilation is complete, we can use either insmod or modprobe command ( insmod myDev.ko  or modprobe myDev.ko, of course assuming the current directory contains the compiled module). The difference between insmod and modprobe is that modprobe automatically reads our module to find any dependencies on other modules and loads them before loading our module (these modules must be present in the standard path though!). insmod lacks this feature.

To test if the driver has been loaded successfully, do cat /proc/modules and cat /proc/devices.  We should see our module name in the first case and device name in the second.

cat /proc/modules

cat /proc/modules

cat /proc/devices

cat /proc/modules

To unload the driver, use rmmod command. (rmmod myDev.ko)

Testing the driver

To test the driver, we try writing something to the device file and then reading it. For example,

Testing the driver

Testing the driver

See the output. (The reason for the ‘ugly’ output is because echo automatically writes a newline character to the end of string. When the driver reverses the string, the newline is shifted to the front of the string and there is no newline at the end. Hence the result being ‘ugly’)

To see how this can be done from our program, I wrote a demo program given below

Interacting with the driver

Interacting with the driver

Compile it normally(or run make test) and run ./test  some_string  and see the output.

Testing the driver

Testing the driver

Note: You need to be root to compile the module, load the module and unload the module.

This driver interface presented here is an old one, there is a newer one but the old one is still supported.

Advertisements

Posted in Kernel | Tagged: , , , , | 69 Comments »

Implementing a System call in Linux Kernel 2.6.35

Posted by appusajeev on November 13, 2010

As we know, System calls are a set of services/functions provided to the programmer by the OS. These functions can be invoked in any language that provide some interface to the System call mechanism of the OS. Some common linux system calls are open,read,write etc.
While executing a system call, the calling process moves from user space to kernel space and back to user space when its completes executing the call.

There are around 338 system calls in linux kernel  2.6.35.7 by default. Presented here a howto on adding a new one into the kernel, a 339th one so that it will be available globally for any program. As example, we will implement the strcpy function as a system call so that it can be used without including string.h.

Obviously, you need the kernel source tree since some kernel modification is involved. Get it from kernel.org (any kernel version higher than 2.6.35.7 would work fine )and untar it to get linux-2.6.35.7. The paths used below all will be relative to this path.
We need to edit 4 files and 2 files need to be created.

The Code

First, lets start off writing the code for strcpy. We need to include the file linux/linkage.h because it contains the macro asmlinkage which means that the system call expects the arguments on the stack and not in registers. Printk is the kernel alternative of printf, but with certain peculiar properties.

Code for the system call

Code for the system call

<1>” tells printk that we are giving that message the highest priority.

Create a folder named ‘test’ in the root of linux source directory and save this file as strcpy.c in that directory. Create a Makefile in that directory containing only the line.

obj-y:=strcpy.o

Thus, now the strcpy.c file and Makefile are present in

linux-2.6.35.7/test/strcpy.c
linux-2.6.35.7/test/Makefile

The Edits

The following files need to be edited.

1. linux-2.6.35.7/arch/x86/kernel/syscall_table_32.S

Append to the file the following line

.long sys_strcpy

linux-2.6.35/arch/x86/kernel/syscall_table_32.S

linux-2.6.35/arch/x86/kernel/syscall_table_32.S

2. linux-2.6.35.7/arch/x86/include/asm/unistd_32.h

This file contains the unique number associated with each system call. We can see the names of all the system calls and the number associated with each. After the last system call-number pair (around line 345), add a line

#define __NR_strcpy  338

(if 337 was the number associated with the last system call).
Then replace NR_syscalls directive denoting the total number of system calls by the previous number incremented by 1 ie the new value is

#define NR_syscalls 339

Note down the number 338, we need it later.

linux-2.6.35/arch/x86/include/asm/unistd_32.h

linux-2.6.35/arch/x86/include/asm/unistd_32.h

3. linux-2.6.35.7/include/linux/syscalls.h

This file contains the prototype of all the system calls. Here we append to the file the prototype of our function.
ie. We add  the line

asmlinkage  long sys_strcpy(char *dest,char *src);


linux-2.6.35/include/linux/syscalls.h

linux-2.6.35/include/linux/syscalls.h

4. Makefile

Open Makefile present in the root of source directory and find the area where core-y is defined and add the folder test to the end of that line as shown below

Makefile

Makefile

Next compile the kernel. Assuming you are familiar with kernel compilation, execute
make bzImage  –j4
The last argument is optional and is intended to speed up compilation on dual core CPUS
Once compilation is complete, install the kernel by executing the following command with root permission
make install
Once the kernel image is installed to /boot, reboot the system.

Testing

Now we need to check the newly done system call. Run the code below and feel the satisfaction 🙂

System call test

System call test

The kernel has performed the strcpy for us. Cool ! isn’t it

Note:
Execute the command dmesg now, you can see done printed in the last. Printk by default doesnt  print to the terminal. It writes to the kernel ring buffer which is printed by the dmesg

Try putting an infinite loop inside a system call, the system just drops dead. As it goes, the linux kernel does not preempt itself.

<1> means that we are giving that message the highest priority Create a folder name ‘test’ in the root of linux source directory and save this file as strcpy.c in that directory . Create a makefile in that directory containing the line Obj-y:=strcpy.o The Edits The files to be edited are 1 . /usr/src/linux-2.6.32.5/arch/x86/kernel/syscall_table_32.S Append the line  .long sys_strcpy to the file(Replace sys_strcpy with whatever name you want) 2. /usr/src/linux-2.6.32.5/arch/x86/include/asm/unistd_32.h This file contains the unique number associated with each system call. We can see the names of all the system calls and the number associated with each. After the last system call-number pair (around line 345), add a line
#define __NR_strcpy  338  (if 337 was the number associated with the last system call).
Then replace NR_syscalls directive denoting the total number of system calls by the previous number incremented by 1 ie the new value id
#define NR_syscalls 339 Note down the number 338, we need it later. 3. /usr/src/linux-2.6.32.5/include/linux/syscalls.h This file contains the prototypes of system calls. Here we append to the file the prototype of our file.
ie. We add  the line
asmlinkage  long sys_strcpy(char *dest,char *src);
4. Makefile
Open makefile and find the area where core-y is defined and add the folder test to the end of that line as shown below

Posted in Kernel | Tagged: , , | 26 Comments »