Luna/kernel/src/thread/Thread.cpp

#include "thread/Thread.h"
#include "Log.h"
#include "arch/CPU.h"
#include "memory/MemoryManager.h"
#include "thread/Scheduler.h"
#include <bits/atfile.h>
#include <bits/open-flags.h>
#include <bits/signal.h>
#include <bits/waitpid.h>
#include <luna/Alignment.h>
#include <luna/Alloc.h>
#include <luna/Atomic.h>
#include <luna/PathParser.h>

static Atomic<pid_t> g_next_id;

LinkedList<Thread> g_threads;
LinkedList<Process> g_processes;

void Thread::init()
{
    g_next_id = 2;
}

Result<Thread*> new_thread()
{
    Thread* const thread = TRY(make<Thread>());

    thread->tid = g_next_id++;

    return thread;
}

pid_t next_thread_id()
{
    return g_next_id.load();
}

Result<int> Process::allocate_fd(int min, FileDescriptor& descriptor)
{
    if (min < 0 || min >= FD_MAX) return err(EINVAL);

    auto table = fd_table.lock();
    for (int i = min; i < FD_MAX; i++)
    {
        if (!(*table)[i].has_value())
        {
            (*table)[i] = descriptor;
            return i;
        }
    }

    return err(EMFILE);
}

Result<FileDescriptor*> Process::resolve_fd(int fd)
{
    if (fd < 0 || fd >= FD_MAX) return err(EBADF);

    auto table = fd_table.lock();

    Option<FileDescriptor>& maybe_descriptor = (*table)[fd];

    if (!maybe_descriptor.has_value()) return err(EBADF);

    return maybe_descriptor.value_ptr();
}

Credentials Process::credentials()
{
    auto credentials = auth.lock();
    return *credentials;
}

Result<Vector<gid_t>> Process::copy_groups()
{
    auto groups = extra_groups.lock();
    return groups->shallow_copy();
}

Result<int> Process::allocate_timerid()
{
    ScopedMutexLock lock(posix_timer_mutex);

    for (int i = 0; i < MAX_POSIX_TIMERS; i++)
    {
        if (!posix_timers[i].has_value())
        {
            posix_timers[i] = Timer {};
            return i;
        }
    }

    return err(EMFILE);
}

Result<Timer*> Process::resolve_timerid(int _tid)
{
    if (_tid < 0 || _tid >= MAX_POSIX_TIMERS) return err(EBADF);

    Option<Timer>& maybe_timer = posix_timers[_tid];

    if (!maybe_timer.has_value()) return err(EINVAL);

    return maybe_timer.value_ptr();
}

Result<SharedPtr<VFS::Inode>> Process::resolve_atfile(int dirfd, const String& path, bool allow_empty_path,
                                                      bool follow_last_symlink, SharedPtr<VFS::Inode>* parent_inode)
{
    if (parent_inode) *parent_inode = this->current_directory;

    if (PathParser::is_absolute(path.view())) return VFS::resolve_path(path.chars(), this, {}, follow_last_symlink);

    if (dirfd == AT_FDCWD) return VFS::resolve_path(path.chars(), this, this->current_directory, follow_last_symlink);

    auto descriptor = TRY(resolve_fd(dirfd));

    if (parent_inode) *parent_inode = descriptor->inode();

    if (path.is_empty() && allow_empty_path) return descriptor->inode();

    return VFS::resolve_path(path.chars(), this, descriptor->inode(), follow_last_symlink);
}

[[noreturn]] void Process::exit(int _status)
{
    check(this == Process::current()); // Process::exit() should only be called by the process itself.

#ifndef MOON_ENABLE_TESTING_FEATURES
    if (id == 1) fail("the init process exited");
#else
    if (id == 1) CPU::magic_exit(_status);
#endif

    Scheduler::for_each_thread(this, [](Thread* thread) {
        thread->quit();
        return true;
    });
    Scheduler::signal_reap_thread();

    thread_count = 0;

    status = _status;

    Scheduler::for_each_child(this, [](Process* child) {
        child->parent = Scheduler::init_process();
        return true;
    });

    if (is_session_leader())
    {
        kinfoln("process %d is exiting as a session leader, sending signals to session", id);
        // FIXME: Send SIGHUP only to the foreground process group if the session has a controlling terminal.
        Scheduler::for_each_in_session(sid, [this](Process* p) {
            if (p == this) return true;
            p->sid = 0;
            p->controlling_terminal = {};
            p->send_signal(SIGHUP);
            kinfoln("reparenting and sending SIGHUP to %d", p->id);
            return true;
        });
    }

    if (parent)
    {
        Scheduler::for_each_thread(parent, [&](Thread* t) {
            if (t->state == ThreadState::Waiting)
            {
                pid_t expected = -1;
                if (t->child_being_waited_for.compare_exchange_strong(expected, id))
                {
                    t->wake_up();
                    return false;
                }

                expected = id;
                if (t->child_being_waited_for.compare_exchange_strong(expected, id))
                {
                    t->wake_up();
                    return false;
                }
            }
            return true;
        });

        parent->send_signal(SIGCHLD);
    }

    kernel_yield();
    unreachable();
}

void Thread::quit()
{
    state = ThreadState::Dying;
}

void Thread::exit(bool yield)
{
    quit();

    process->thread_count--;
    if (process->thread_count == 0) { process->exit(0); }

    if (yield) kernel_yield();
}

enum class DefaultSignalAction
{
    Ignore,
    Terminate,
    Stop,
};

// FIXME: Implement coredumps for some signals.
static constexpr DefaultSignalAction default_actions[] = {
    DefaultSignalAction::Terminate, // SIGHUP
    DefaultSignalAction::Terminate, // SIGINT
    DefaultSignalAction::Terminate, // SIGQUIT (dump core)
    DefaultSignalAction::Terminate, // SIGILL  (dump core)
    DefaultSignalAction::Terminate, // SIGTRAP (dump core)
    DefaultSignalAction::Terminate, // SIGABRT (dump core)
    DefaultSignalAction::Ignore,    // SIGCHLD
    DefaultSignalAction::Terminate, // SIGFPE  (dump core)
    DefaultSignalAction::Terminate, // SIGKILL
    DefaultSignalAction::Stop,      // SIGSTOP
    DefaultSignalAction::Terminate, // SIGSEGV (dump core)
    DefaultSignalAction::Ignore,    // SIGCONT (Handled separately)
    DefaultSignalAction::Terminate, // SIGPIPE
    DefaultSignalAction::Terminate, // SIGALRM
    DefaultSignalAction::Terminate, // SIGTERM
    DefaultSignalAction::Terminate, // SIGTTIN
    DefaultSignalAction::Terminate, // SIGTTOU
    DefaultSignalAction::Ignore,    // SIGWINCH
};

void Thread::process_pending_signals(Registers* current_regs)
{
    interrupted = false;
    for (int i = 0; i < NSIG; i++)
    {
        int signo = i + 1;
        if (signo != SIGKILL && signo != SIGSTOP && signal_mask.get(i)) continue;
        if (pending_signals.get(i))
        {
            pending_signals.set(i, false);
            auto handler = signal_handlers[i];
            if (signo != SIGKILL && signo != SIGSTOP && handler.sa_handler == SIG_IGN) return;
            if (handler.sa_handler == SIG_DFL || signo == SIGKILL || signo == SIGSTOP)
            {
            default_signal:
                if (process->id == 1)
                {
                    kwarnln("signal: init got a signal it has no handler for, ignoring");
                    return;
                }

                auto action = default_actions[i];
                switch (action)
                {
                case DefaultSignalAction::Ignore: return;
                case DefaultSignalAction::Terminate:
                    kwarnln("Terminating thread %d with signal %d", tid, signo);
                    CPU::print_stack_trace_at(current_regs);
                    process->exit(signo | _SIGBIT);
                    unreachable();
                case DefaultSignalAction::Stop: stop();
                default: return;
                }
            }
            // If we fail to deliver the signal (usually because there's not enough space on the stack), execute the
            // default action. FIXME: Should this be changed?
            if (!deliver_signal(signo, current_regs)) goto default_signal;
            return;
        }
    }
}

int Thread::pending_signal_count()
{
    int result = 0;
    for (int i = 0; i < NSIG; i++)
    {
        if (pending_signals.get(i)) { result++; }
    }
    return result;
}

int Thread::pending_signal()
{
    for (int i = 0; i < NSIG; i++)
    {
        if (pending_signals.get(i)) { return i + 1; }
    }
    return 0;
}

bool Thread::will_ignore_pending_signal()
{
    for (int i = 0; i < NSIG; i++)
    {
        if (pending_signals.get(i))
        {
            int signo = i + 1;
            if (signo == SIGKILL || signo == SIGSTOP) return false;
            if (signal_mask.get(i)) continue;
            auto handler = signal_handlers[i];
            if (handler.sa_handler == SIG_IGN) return true;
            if (handler.sa_handler == SIG_DFL && default_actions[i] == DefaultSignalAction::Ignore) return true;
            return false;
        }
    }
    return false;
}

void Process::send_signal(int signo)
{
    Scheduler::for_each_thread(this, [signo](Thread* t) {
        t->send_signal(signo);
        return false;
    });
}

void Thread::send_signal(int signo)
{
    if (is_kernel) return;
    if (state == ThreadState::Exited || state == ThreadState::Dying) return;

    if (state == ThreadState::Stopped && signo == SIGCONT)
    {
        wake_up();
        return;
    }

    check(signo > 0 && signo <= NSIG);
    pending_signals.set(signo - 1, true);

    if (state == ThreadState::Waiting || state == ThreadState::Sleeping || is_in_kernel(&regs))
    {
        if (state == ThreadState::Stopped && signo != SIGKILL) return;
        interrupted = true;
        wake_up();
    }
}

static constexpr usize MAX_STACK_SIZE = 8 * 1024 * 1024; // 8 MB

bool Thread::check_stack_on_exception(u64 stack_pointer)
{
    if (stack_pointer < stack.bottom() || stack_pointer >= stack.top())
        kwarnln("Abnormal stack (Stack pointer outside the normal range, %.16lx-%.16lx)", stack.bottom(), stack.top());

    // Check whether the stack pointer is within 8 pages of the bottom of the stack
    u64 threshold = stack.bottom() - (ARCH_PAGE_SIZE * 8);

    if (stack_pointer >= threshold && stack_pointer < stack.bottom())
    {
        kwarnln("Likely stack overflow (CPU exception a few pages below the stack)");

        // Try to grow the stack
        usize stack_space_remaining = MAX_STACK_SIZE - stack.bytes();
        if (!stack_space_remaining)
        {
            kwarnln("Failed to grow stack: this thread already used up all its stack space");
            return false;
        }

        usize exceeded_bytes = align_up<ARCH_PAGE_SIZE>(stack.bottom() - stack_pointer);
        if (exceeded_bytes > stack_space_remaining)
        {
            kwarnln("Failed to grow stack: this thread needs more space than the one it has remaining (%zu bytes out "
                    "of %zu remaining)",
                    exceeded_bytes, stack_space_remaining);
            return false;
        }

        auto address_space = process->address_space.lock();

        // If we can, we'll add 2 more pages of buffer space, otherwise we use whatever we can.
        usize bytes_to_grow = min(stack_space_remaining, exceeded_bytes + 2 * ARCH_PAGE_SIZE);
        auto maybe_base = (*address_space)->grow_region(stack.bottom(), bytes_to_grow / ARCH_PAGE_SIZE, true);
        if (maybe_base.has_error())
        {
            kwarnln("Failed to grow stack: could not allocate virtual memory space (%s)", maybe_base.error_string());
            return false;
        }

        u64 base = maybe_base.release_value();
        auto result = MemoryManager::alloc_at_zeroed(base, bytes_to_grow / ARCH_PAGE_SIZE,
                                                     MMU::ReadWrite | MMU::NoExecute | MMU::User);
        if (result.has_error())
        {
            (*address_space)->free_region(base, bytes_to_grow / ARCH_PAGE_SIZE);
            kwarnln("Failed to grow stack: could not allocate physical pages (%s)", result.error_string());
            return false;
        }

        kinfoln("Stack expanded from %lx (%zu bytes) to %lx (%zu bytes)", stack.bottom(), stack.bytes(), base,
                stack.bytes() + bytes_to_grow);

        stack = Stack { base, stack.bytes() + bytes_to_grow };
        return true;
    }

    return false;
}

void Thread::stop()
{
    state = ThreadState::Stopped;
    kernel_yield();
}

Process* Process::current()
{
    return Scheduler::current()->process;
}