spawn

fn spawn(ev: *Evented, options: process.SpawnOptions) process.SpawnError!Spawned {
    var cancel_region: CancelRegion = .init();
    defer cancel_region.deinit();

    // The child process does need to access (one end of) these pipes. However,
    // we must initially set CLOEXEC to avoid a race condition. If another thread
    // is racing to spawn a different child process, we don't want it to inherit
    // these FDs in any scenario; that would mean that, for instance, calls to
    // `poll` from the parent would not report the child's stdout as closing when
    // expected, since the other child may retain a reference to the write end of
    // the pipe. So, we create the pipes with CLOEXEC initially. After fork, we
    // need to do something in the new child to make sure we preserve the reference
    // we want. We could use `fcntl` to remove CLOEXEC from the FD, but as it
    // turns out, we `dup2` everything anyway, so there's no need!
    const pipe_flags: linux.O = .{ .CLOEXEC = true };

    const stdin_pipe = if (options.stdin == .pipe) try pipe2(pipe_flags) else undefined;
    errdefer if (options.stdin == .pipe) {
        ev.destroyPipe(stdin_pipe);
    };

    const stdout_pipe = if (options.stdout == .pipe) try pipe2(pipe_flags) else undefined;
    errdefer if (options.stdout == .pipe) {
        ev.destroyPipe(stdout_pipe);
    };

    const stderr_pipe = if (options.stderr == .pipe) try pipe2(pipe_flags) else undefined;
    errdefer if (options.stderr == .pipe) {
        ev.destroyPipe(stderr_pipe);
    };

    const any_ignore =
        options.stdin == .ignore or options.stdout == .ignore or options.stderr == .ignore;
    const dev_null_fd = if (any_ignore) try ev.null_fd.open(ev, &cancel_region, "/dev/null", .{
        .ACCMODE = .RDWR,
    }) else undefined;

    const prog_pipe: [2]fd_t = if (options.progress_node.index != .none) pipe: {
        // We use CLOEXEC for the same reason as in `pipe_flags`.
        const pipe = try pipe2(.{ .NONBLOCK = true, .CLOEXEC = true });
        _ = linux.fcntl(pipe[0], linux.F.SETPIPE_SZ, @as(u32, std.Progress.max_packet_len * 2));
        break :pipe pipe;
    } else .{ -1, -1 };
    errdefer ev.destroyPipe(prog_pipe);

    var arena_allocator = std.heap.ArenaAllocator.init(ev.allocator());
    defer arena_allocator.deinit();
    const arena = arena_allocator.allocator();

    // The POSIX standard does not allow malloc() between fork() and execve(),
    // and this allocator may be a libc allocator.
    // I have personally observed the child process deadlocking when it tries
    // to call malloc() due to a heap allocation between fork() and execve(),
    // in musl v1.1.24.
    // Additionally, we want to reduce the number of possible ways things
    // can fail between fork() and execve().
    // Therefore, we do all the allocation for the execve() before the fork().
    // This means we must do the null-termination of argv and env vars here.
    const argv_buf = try arena.allocSentinel(?[*:0]const u8, options.argv.len, null);
    for (options.argv, 0..) |arg, i| argv_buf[i] = (try arena.dupeSentinel(u8, arg, 0)).ptr;

    const env_block = env_block: {
        const prog_fd: i32 = if (prog_pipe[1] == -1) -1 else prog_fileno;
        if (options.environ_map) |environ_map| break :env_block try environ_map.createPosixBlock(arena, .{
            .zig_progress_fd = prog_fd,
        });
        break :env_block try ev.environ.process_environ.createPosixBlock(arena, .{
            .zig_progress_fd = prog_fd,
        });
    };

    // This pipe communicates to the parent errors in the child between `fork` and `execvpe`.
    // It is closed by the child (via CLOEXEC) without writing if `execvpe` succeeds.
    const err_pipe: [2]fd_t = try pipe2(.{ .CLOEXEC = true });
    errdefer ev.destroyPipe(err_pipe);

    try ev.scanEnviron(); // for PATH
    const PATH = ev.environ.string.PATH orelse default_PATH;

    const pid_result: pid_t = fork: {
        const rc = linux.fork();
        switch (linux.errno(rc)) {
            .SUCCESS => break :fork @intCast(rc),
            .AGAIN => return error.SystemResources,
            .NOMEM => return error.SystemResources,
            .NOSYS => return error.OperationUnsupported,
            else => |err| return unexpectedErrno(err),
        }
    };

    if (pid_result == 0) {
        defer comptime unreachable; // We are the child.
        // Note that the parent uring is no longer accessible, so we must no longer reference `ev`.
        var sync: CancelRegion.Sync = .{ .cancel_region = .initBlocked() };
        const err = setUpChild(&sync, .{
            .stdin_pipe = stdin_pipe[0],
            .stdout_pipe = stdout_pipe[1],
            .stderr_pipe = stderr_pipe[1],
            .dev_null_fd = dev_null_fd,
            .prog_pipe = prog_pipe[1],
            .argv_buf = argv_buf,
            .env_block = env_block,
            .PATH = PATH,
            .spawn = options,
        });
        writeAllSync(&sync, err_pipe[1], @ptrCast(&err)) catch {};
        const exit = if (builtin.single_threaded) linux.exit else linux.exit_group;
        exit(1);
    }

    const pid: pid_t = @intCast(pid_result); // We are the parent.
    errdefer comptime unreachable; // The child is forked; we must not error from now on

    ev.closeAsync(err_pipe[1]); // make sure only the child holds the write end open

    if (options.stdin == .pipe) ev.closeAsync(stdin_pipe[0]);
    if (options.stdout == .pipe) ev.closeAsync(stdout_pipe[1]);
    if (options.stderr == .pipe) ev.closeAsync(stderr_pipe[1]);

    if (prog_pipe[1] != -1) ev.closeAsync(prog_pipe[1]);

    options.progress_node.setIpcFile(ev, .{ .handle = prog_pipe[0], .flags = .{ .nonblocking = true } });

    return .{
        .pid = pid,
        .err_fd = err_pipe[0],
        .stdin = switch (options.stdin) {
            .pipe => .{ .handle = stdin_pipe[1], .flags = .{ .nonblocking = false } },
            else => null,
        },
        .stdout = switch (options.stdout) {
            .pipe => .{ .handle = stdout_pipe[0], .flags = .{ .nonblocking = false } },
            else => null,
        },
        .stderr = switch (options.stderr) {
            .pipe => .{ .handle = stderr_pipe[0], .flags = .{ .nonblocking = false } },
            else => null,
        },
    };
}