exit (nr 60) / exit_group (nr 231)

Linux Signature

void _exit(int status);          // nr 60
void exit_group(int status);     // nr 231

Description

• exit (60): Terminates the calling thread.
• exit_group (231): Terminates all threads in the calling process.

Both are handled identically in ostoo since each process currently has exactly one thread.

Current Implementation

1. Looks up the current PID.
2. If it’s a user process (not ProcessId::KERNEL), calls terminate_process:
   • Logs pid N exited with code C to serial.
   • Unblocks vfork parent: If this process was created by clone(CLONE_VFORK) and has not yet called execve, unblocks the parent thread so it can resume. Clears vfork_parent_thread.
   • Closes all fds: Releases IRQ handles, completion ports, pipes, channels, etc. while the process’s page tables are still active.
   • Frees user address space: Switches CR3 to the kernel boot PML4 and updates the scheduler’s thread record, then frees all user-half page tables and data frames via cleanup_user_address_space. Skipped for CLONE_VM children (shared PML4 still used by parent).
   • Marks zombie: Sets the process state to Zombie with the exit code.
   • Wakes parent: Queues SIGCHLD and unblocks the parent’s wait_thread if set.
   • Yields + dies: Donates remaining quantum to the parent, calls yield_now(), then kill_current_thread() marks the thread as Dead.
3. If it’s a kernel thread: prints a halt message and calls kill_current_thread().

Zombie processes are reaped by waitpid (when a parent collects exit status) or lazily by reap_zombies() at the start of spawn_process.

Source: osl/src/syscalls/process.rs — sys_exit, libkernel/src/process.rs — terminate_process

CR3 safety on exit

The process’s PML4 frame must not be freed while CR3 still references it. The frame allocator uses an intrusive free-list that overwrites the first 8 bytes of freed frames immediately; if the scheduler later reschedules the dying thread (before kill_current_thread runs), a TLB refill through the corrupted PML4 would triple-fault. terminate_process therefore switches to the kernel boot PML4 (stored in KERNEL_PML4_PHYS during memory::init_services) and updates the scheduler via set_current_cr3 before calling cleanup_user_address_space.

Future Work

• Properly distinguish exit (single thread) from exit_group (all threads) once multi-threaded processes are supported.
• Service auto-cleanup: remove service registry entries on process exit.