Julia Evans

Day 42: How to run a simple ELF executable, from scratch (I don't know)

in coding, hackerschool, kernel

I want to compile a 32-bit “Hello, world!” statically-linked ELF binary for Linux, and try to run it in my operating system. I’m trying to understand what I’ll have to do. The goal is to get everything just barely working, so that it will print the string to the screen and not crash the whole system.

I asked a question about this a little while ago, and got lots of helpful responses. Now I need to make it a bit more concrete, though.

I’ve discovered that this “set stuff up so that a program can run” business is called loading, and what I’m doing is writing a loader. Sweet.

Right now I’m doing this before implementing paging and virtual memory and not after, because this seems more fun than virtual memory for now. If this is a very bad idea, I would like to know.

Things I’ll have to do

  1. Compile “hello-c” for a 32-bit OS, with gcc -m32 -static
  2. Parse the ELF headers (using the wikipedia article, and this great picture as a reference)
  3. Add an interrupt handler for int 80 (or sysenter, we’ll see!), so that I can handle system calls.
  4. Write the actual system call implementations, as few as possible.
  5. Find e_entry, the entry point of the binary.
  6. Initialize registers? How?
  7. Change something, so that the memory addresses in the binary aren’t broken. Maybe? I still don’t 100% understand this.
  8. Finally: Jump to _start, the memory address in e_entry. I want to just do jmp address here. Then my program will run?

Things I won’t have to do (yet)

  1. Read the file into memory – I’m planning to just keep the file as a bunch of bytes in RAM to start, or possibly have a simple RAM filesystem later.
  2. Security, and making sure process can’t trample on each others’ address spaces.
  3. Scheduling.
  4. Set up a special heap for the process. I’m just going to allocate everyone’s memory in the same part of physical memory for now. And never free. Yeah.

Questions I have

  1. Do I need to make sure my binary is position independent?
  2. Do I need to implement virtual memory & paging? (I think not)
  3. Do I need to have a separate “user space” for the code to run in, or will it run in kernel space? (I think it will run in kernel space)
  4. Do I need to change something in the GOT and/or PLT to make the addresses work right? (I think yes? maybe?) Is there even a PLT in a static executable, or is that just for dynamic linking? Eep. Hmm.

Comments