How to launch a process inside a minimal linux kernel environment? - c

Question

Is it possible to run a generic C program on top of the bare linux kernel?
I am using a virtualized linux minimal system with Qemu, where I load my custom kernel and a root filesystem, but I apparently can't run my user-space hello world program.
EDIT 1: I have updated the question to be more detailed. Here are the steps I have followed:
1] I have compiled the hello_world program with the following command, on my host machine:
gcc hello_world.c -o hello_world
The source code is the following:
#include <stdio.h>
int main(int argc, char const *argv[]) {
printf("Hello world!\n");
return 0;
}
By running the program on my host machine, it works fine.
I have checked as #nexus suggested the libraries against which the program is built, and are the following:
$ ldd hello_world
linux-vdso.so.1 => (0x00007ffc0dfc8000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007efd90658000)
/lib64/ld-linux-x86-64.so.2 (0x000055a57447f000)
2] I have build the root filesystem by executing buildroot, which comes by default from Ciro Santilli's linux-kernel-module-cheat. I have compiled the linux kernel and got its corresponding bzImage
3] I have mounted the root filesystem and copied my compiled program inside of it, as follows:
sudo mount -t ext2 rootfs.ext2 /home/andreww/Desktop/mountfile/
sudo cp hello_world /home/andreww/Desktop/mountfile/
sudo umount /home/andreww/Desktop/mountfile/
This feels wrong, I did it because I haven't found a way to compile the C program directly inside the Guest Machine.
On the guest machine I don't have gcc, indeed if I try to call it I get:
# -sh: gcc: not found
4] I have started the emulated system with the command
qemu-system-x86_64 \
-M pc \
-append 'root=/dev/vda $extra_append' \
-drive file=${images_dir}/rootfs.ext2,if=virtio,format=raw \
-kernel ${images_dir}/bzImage \
-m 256M \
-net nic,model=virtio \
-net user \
-enable-kvm \
-cpu host \
-smp threads=4
Which I have taken from Ciro Santilli's linux-kernel-module-cheat.
The rootfs.ext2 is the root filesystem which the system is started against, and the bzImage is the kernel image I am going to start the guest machine with.
5] The guest machine is emulated within a shell. This is what I expected since I am virtualizing only the bare kernel and not a whole operating system image.
I am displayed the buildroot login, and I login as root and run the following commands:
When I try to run my hello program I am displayed the following message:
-sh: ./hello_world: not found
I suppose that my program is never executed, even if I have seen by debugging that it's loaded as an elf file into the system.
If I type an random command which doesn't exists, I get the same error.
The goal is to make the hello_world program run in user-space, on top of the kernel, I don't want it to be a part of it.
I agree with you guys, I'm probably missing some libraries or something basic, but I have no idea on what to do.

Answer 1

You can basically run a process instead of init, if that is what you intend to do. To run a process instead of the systems init application, you can either replace the original init or pass init=/path/to/you/application as a kernel parameter to your VM.
But there are several things you should keep in mind:
All libraries need to be present in your VM at the same locations as on the system you have build the application on. This includes also the basic c-library it has been built against. You can get a list of all those libraries using
ldd hello_world
As an example, if you compile the following program
#include <stdio.h>
int main(int argc, char** argv) {
printf("hello world!\n");
return 0;
}
you might get something like this as a result for ldd
$ ldd test
linux-vdso.so.1 => (0x00007ffdba6ba000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f8559f50000)
/lib64/ld-linux-x86-64.so.2 (0x000055fc58ece000)
If you replace the init process, you need to take care of console interrupts and handle them properly. If you don't you risk really bad system crashes. As a result, it is a good idea to keep an init process and let that process start your application, if you do not really know what you are up to.
I have seen the output of your command in the past. Are you using a shell-script to start your application? If so, it might indicate that you are currently missing some libraries in your virtual environment.
Edit:
Maybe you could instead create a statically linked application which means to compile all libraries into the same file. On gcc you could use that -static switch for that. See the difference:
$ gcc -o test test.c
$ ldd test
linux-vdso.so.1 => (0x00007ffcd1f22000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007f088d43a000)
/lib64/ld-linux-x86-64.so.2 (0x00005580f8a2b000)
$ gcc -static -o test test.c
$ ldd test
not a dynamic executable
As already mentioned you can use ldd to find the required libraries. Another option I would normally use is objdump, as the output is more detailed.
So for my application that would mean something like this (I have shortened the output to the relevant section):
$ objdump -p test
test: file format elf64-x86-64
[...]
Dynamic Section:
NEEDED libc.so.6
INIT 0x00000000004003c8
FINI 0x00000000004005c4
INIT_ARRAY 0x0000000000600e10
INIT_ARRAYSZ 0x0000000000000008
FINI_ARRAY 0x0000000000600e18
FINI_ARRAYSZ 0x0000000000000008
[...]
So in my case, that is libc.so.6 I would need to copy from /lib/x86_64-linux-gnu/libc.so.6 on my host to /lib/x86_64-linux-gnu/libc.so.6 within the virtual machine.
Hope that helps.