DEV Community

loading...

How to add system call (syscall) to the kernel, compile and test it?

omergulen profile image Ömer Gülen Updated on ・6 min read

One of my example syscall methods will print Hello world to the kernel buffer.

The second one gets PID of a process and prints the elapsed time of it to the kernel buffer in my other tutorial:

1. Setup environment

I prefer to get root and apply the steps as root to prevent getting permission failures on the way. If you don't want to apply the steps as a root, you can just use sudo for root required commands.

su -
Enter fullscreen mode Exit fullscreen mode

will switch to root user after entering the password of the root. Then, first you can check the active kernel version of your OS with:

uname -r
Enter fullscreen mode Exit fullscreen mode

This prints out 4.19.0-6-amd64 in my case. Then, we need to get the source of a kernel. I will use slightly newer version (4.20.1) from my version with following wget command.

wget https://mirrors.edge.kernel.org/pub/linux/kernel/v4.x/linux-4.20.1.tar.xz
Enter fullscreen mode Exit fullscreen mode

This will download the source of 4.20.1. It may take a while, like 2-3 minutes depending on your internet speed.

Extract the compressed kernel code with

tar -xvf linux-4.20.1.tar.xz
Enter fullscreen mode Exit fullscreen mode

It will create a folder named linux-4.20.1.tar.xz and extract the compressed code into that folder.

Now we will change our directory to new kernel code.

cd linux-4.20.1
Enter fullscreen mode Exit fullscreen mode

2. Add "Hello world" syscall to kernel

I prefer creating new folder for my own stuff while adding a new syscall.

mkdir hello && cd hello
Enter fullscreen mode Exit fullscreen mode

After that, I will create a C file for my syscall implementation. I prefer vim hello.c to create and edit the file and insert following C code.

#include <linux/kernel.h>

asmlinkage long sys_hello(void) 
{
        //printk prints to the kernel’s log file.
        printk("Hello world\n");
        return 0;
}
Enter fullscreen mode Exit fullscreen mode

We need to create a Makefile in the hello directory.

vim Makefile
Enter fullscreen mode Exit fullscreen mode

and then insert this:

obj-y := hello.o

Then, go to the parent directory (kernel source main directory):

cd ..
Enter fullscreen mode Exit fullscreen mode

We need to add our new syscall directory to Makefile, in this way it will compile our syscall, too. To achieve this, search for core-y in the Makefile then, find the

In vim you can do search with /core-y after pressing ESC.

core-y += kernel/ mm/ fs/ ipc/ security/ crypto/ block/

line and add hello/ to the end of this line.

As a result, the line should be looking like this:

core-y += kernel/ mm/ fs/ ipc/ security/ crypto/ block/ hello/

Next step is adding the new system call to the system call table.
If you are on a 32-bit system you’ll need to change syscall_32.tbl file.

vim arch/x86/entry/syscalls/syscall_32.tbl
Enter fullscreen mode Exit fullscreen mode

For 64-bit, change syscall_64.tbl.

vim arch/x86/entry/syscalls/syscall_64.tbl
Enter fullscreen mode Exit fullscreen mode

I will continue with my 64-bit OS and my steps will be accordingly.

We need to keep table's structure while adding our syscall. So that we will add our line to the end of the line. My last syscall has number of 547, so that I will use 548, you also should use N+1.

My syscall data:

548 64 hello sys_hello

Example file:

546     ...     ...     ...
547     ...     ...     ...
548     64      hello   sys_hello
Enter fullscreen mode Exit fullscreen mode

Now, we need to add our syscall method signature to syscalls header file which is syscalls.h.

vim include/linux/syscalls.h
Enter fullscreen mode Exit fullscreen mode

Then, add the following line to the end of the document before the #endif statement:

asmlinkage long sys_hello(void);
Enter fullscreen mode Exit fullscreen mode

3. Compiling our kernel

Before starting to compile you need to install a few packages. Type the following commands in your terminal to install required packages:

apt-get install gcc &&
apt-get install libncurses5-dev &&
apt-get install bison &&
apt-get install flex &&
apt-get install libssl-dev &&
apt-get install libelf-dev &&
apt-get update &&
apt-get upgrade &&
apt-get make
Enter fullscreen mode Exit fullscreen mode

Now, you can configure your kernel by using the config menu by executing:

make menuconfig
Enter fullscreen mode Exit fullscreen mode

IMPORTANT NOTE: While entering this command, you might want to maximize your terminal screen. Otherwise, you might get and error and a pop-up screen will not appear.

Once the above command is used to configure the Linux kernel, you will get a pop up window with the list of
menus and you can select the items for the new configuration. If your unfamiliar with the configuration just check for the file systems menu and check whether “ext4” is chosen or not, if not select it and save the configuration.

OR: You can simply execute the following command and use the default configuration.

make defconfig
Enter fullscreen mode Exit fullscreen mode

Now, finally we can compile the kernel with the following command:

sudo make
Enter fullscreen mode Exit fullscreen mode

The compilation took 40 mins to 1 hour on my uni-core VM. As an alternative solution you can increase the core count of your VM in the VM settings and use them to compilation with the -jn parameter. n is the number of cores dedicated to your VM.

In my case, I've used the following ocmmand for 8-core VM which reduced the compilation time to 20 seconds to 1 minute:

make -j8
Enter fullscreen mode Exit fullscreen mode

4. Installing our kernel

After the successful compilation, to install/update kernel use the following command:

make modules_install install
Enter fullscreen mode Exit fullscreen mode

The command will create some files under /boot/ directory and it will automatically make a entry in your grub.cfg.
To check whether it made correct entry, check the files under /boot/ directory . If you have followed the steps without any error you will find the following files in it in addition to others.

5. Testing our syscall in new kernel

Now all we need to do is restart the system:

shutdown -r now
Enter fullscreen mode Exit fullscreen mode

After computer restarts, in grub's advanced options, you can see there are 2 options (without counting the recovery mode options).

Once your computer is up again, you can run the following command the check your kernel version:

uname -r
Enter fullscreen mode Exit fullscreen mode

which prints 4.20.1 in my OS after installing the kernel.

After checking the version of the kernel, we will test our syscall with tiny C program.

vim hello_test.c
Enter fullscreen mode Exit fullscreen mode

Insert the following C code into the hello_test.c file:

#include <stdio.h>
#include <linux/kernel.h>
#include <sys/syscall.h>
#include <unistd.h>

int main()
{
        long int helloCheck = syscall(548);
        printf("System call sys_hello returned %ld\n", helloCheck);
        return 0;
}
Enter fullscreen mode Exit fullscreen mode

Then compile and run it:

gcc hello_test.c -o hello.o && ./hello.o
Enter fullscreen mode Exit fullscreen mode

it prints System call sys_hello returned 0 if there is no problem on the execution and now we can check the kernel log buffer to see if our Hello world is there with the command below:

dmesg
Enter fullscreen mode Exit fullscreen mode

It will print tons of line and in the end we should be seeing our Hello world.

[    ........] ...
[    ........] ...
[    ........] ...
[    2.858022] IPv6: ADDRCONF(NETDEV_UP): enp0s3: link is not ready
[    2.858144] ip (1443) used greatest stack depth: 12424 bytes left
[    2.860434] Adding 10483708k swap on /dev/sda5.  Priority:-2 extents:1 across:10483708k 
[    3.391290] hrtimer: interrupt took 4499049 ns
[    4.887969] e1000: enp0s3 NIC Link is Up 1000 Mbps Full Duplex, Flow Control: RX
[    4.888216] IPv6: ADDRCONF(NETDEV_CHANGE): enp0s3: link becomes ready
[  306.497742] Hello world
Enter fullscreen mode Exit fullscreen mode

We can observe that our Hello world is printed to kernel log buffer. So it worked!

Error - Fix

Debian Certificate Problem

On my first trial I have faced with an error such below:

make[1]: *** No rule to make target 'debian/certs/debian-uefi-certs.pem', needed by 'certs/x509_certificate_list'.  Stop.
make: *** [Makefile:1055: certs] Error 2
Enter fullscreen mode Exit fullscreen mode

- SOLUTION -

I've found the solution of this problem in the debian.org's bug report section. The issue was reported by Heinrich Schuchardt as the following:

The Debian kernels are currently built with CONFIG_SYSTEM_TRUSTED_KEYS="debian/certs benh@debian.org.cert.pem"
This was introduced with this https://alioth-lists-archive.debian.net/pipermail/kernel-svn-changes/2016-April/022904.html.

We are now two years after Ben's contribution and we are still not using kernel module signing (CONFIG_MODULE_SIG is not set in config).

As there is no need for the kernel trusting Ben's certificate, please, remove the setting.

Best regards

Heinrich Schuchardt

As a result, I've cleared the CONFIG_SYSTEM_TRUSTED_KEYS config in the configuration file.

In linux-4.20.1 (which is my kernel source directory), there is a file named .config which stores the configuration of the kernel. With the following command I opened it:

vim .config
Enter fullscreen mode Exit fullscreen mode

Then searched for CONFIG_SYSTEM_TRUSTED_KEYS and changed

CONFIG_SYSTEM_TRUSTED_KEYS="debian/certs benh@debian.org.cert.pem"
Enter fullscreen mode Exit fullscreen mode

into this:

CONFIG_SYSTEM_TRUSTED_KEYS=""
Enter fullscreen mode Exit fullscreen mode

then, saved and closed the file, and tried to compile again and there was no issue.

I will publish the second example in another article since this one is already so long.

Please comment below your questions and feedback.

Discussion

pic
Editor guide
Collapse
syaifulnizarrr profile image
syaifulnizarrr

Hi,

I'm running on x32 platform for my raspberry PI. I was wondering, since I had follow the steps accordingly with slight changes due to x32, and I have this error where it stated that my syscall hello is not implemented. How do I go about fix it?

Collapse
omergulen profile image
Ömer Gülen Author

Hello,
What did you add to your arch/x86/entry/syscalls/syscall_32.tbl file exactly? Also, other than this, what did you do differently to match up with a slight change which is x32 bit.

Collapse
syaifulnizarrr profile image
syaifulnizarrr

Hi,

Linux raspberrypi 4.19.93.-v7+ for my raspberry pi. and it's a 32 bit. So for what I had added to my arch/x86/entry/syscalls/syscall_32.tbl was

387 i386 hello sys_calls __ia32_sys_hello

in addition, on hello.c file,

i tried to play around with assigning integers where if int a = 1, then it will print out like "Test" with but it says along the lines where it's not allowed to assign numbers?

So when I had tried to run using the make command, it gives me an error saying that my syscall hello is not implemented.

And addition, I found out that my kernel is not updated and my core is armv7

Collapse
jpo234 profile image
jpo234

The Pi is ARM based, so anything under the arch/x86 tree is irrelevant.

Collapse
phlash909 profile image
Phil Ashby

Interesting.. a good, step-by-step worked example of doing something unusual with a Linux kernel, I'm left wondering why you would do this for real though?

Lot's more info and articles to dissuade people from tinkering with the Linux ABI here :)

kernel.org/doc/html/latest/process...

Collapse
omergulen profile image
Ömer Gülen Author

I'm a college student and taking the Operating Systems course. As a programming assignment, our lecturer wanted us to add syscall which prints out the time elapsed since the start of a process with given PID.

So while I was doing my homework I used many tutorials and read many source and header files. As a result, I wanted to simplify it to other people, that's all I guess.

I'm sure there are many alternatives methods which people would prefer to syscalls but still, I just wanted to share :)

Thanks for your feedback!

Collapse
phlash909 profile image
Phil Ashby

Ah cool, thanks for helping others along, and sending me down a rabbit hole looking for ways to do this dynamically with loadable modules: TIL officially you cannot as the sys_call_table[] in the kernel is a fixed size, defined at build time; unofficially there are usually unused entries you can usurp..caveat hacker :)

Collapse
fjin9 profile image
fjin9

Hi,
When I run ./a.out, it returns -1 can you help me out?

Collapse
manishfoodtechs profile image
manish srivastava

nice .👍👍👍

Collapse
omergulen profile image