Contributer
- Yuxuan Shui
- Haochen Chen
Overview
This kernel follows a micro-kernel design.The kernel only does the basic process isolation, address space management and IPC, (almost) everything else is in user space. Including the console and the filesystem (so make sure they are there when you do the test).
Although due to time constraint, lack of experience in micro-kernel design and various other reasons, the result is extremly buggy and feature incomplete. At least we learned a lot, and had fun (at the expense of sleep) in the process.
Filesystem structure
The root directory contains 2 subdirectories, /tarfs for the tarfs, /sata for filesystem on the hard drive. /tarfs is readonly, and /sata is (in theory) read/write.
In case you are not able to run your shell for testing (highly likely), at least you can try the 'cd', 'ls', 'touch' and 'cat' builtin commands in our shell to see the filesystem in motion.
Syscalls
syscalls implemented in libc
Those syscalls are implemented in libc as user space functions:
- read, write, open, readdir, writedir, opendir, lseek
- fork, execve (Yep, loading ELF is done in user space)
- chdir, getcwd (We cheated a little here)
Actual syscalls
These are some of the most important syscalls that are actually implemented in the kernel, with brief explainations:
- get_thread_info: Get the thread info (mainly registers) of current running thread.
- asnew: Create a new sub address space. Can create a COW snapshot of current address space.
- sendpage: Used to transfer memory range between address spaces.
- munmap: Unmap a memory range.
- create_task: Create a new task, take an address space and thread info.
- wait_on, wait_on_port: Wait on file descriptors or a port.
- connect_port, open_port, request, respond, get_response: Used for IPC between processes. This API is TCP-like and stateful. The kernel will try to avoid memory copy as much as possible. I admit this API is not well designed, but in my defense, this is my first try.
What are not implemented
- pipe: We had some code for a pipe server in user space, but sadly we don't have time to finish it.
- sleep: The wait_on syscall takes an argument for timeout, but we are short of time to actually implement that.
- brk: Memory is allocated via sendpage syscall.
- waitpid
Other notes
This kernel generates a lot of logs through serial port. So don't (or do, if you are interested) start qemu with serial port.
Conclusion
This little kernel is extremly fragile, so please test with care and love. Thanks!.