Re: [PATCH] alternative^2 to sys_indirect: socket, socketpair

[Alan Cox <alan@lxorguk.ukuu.org.uk>]

Comments all come down to implementation details and it basically looks
sound and fairly small clean change (especially when the various wrappers
and _flags changes were replaced by just fixing up the callers to
functions in a final implementation.

Add a new accept() and it looks good to me

> -	fd = get_unused_fd();
> +	fd = get_unused_fd_flags(flags);

Better to fix up the existing users who don't pass flags (and check why
they need not to), Implementation detail

> -int sock_map_fd(struct socket *sock)
> +static int sock_map_fd_flags(struct socket *sock, int flags)

No need to rename these, the compiler errors will let us quickly
find/fix any users.

> +int sock_map_fd(struct socket *sock)
> +{
> +	return sock_map_fd_flags(sock, 0);
> +}

Ditto

> +	/* Extract the close-on-exec flag.  */
> +	if ((type & SOCK_CLOEXEC) != 0) {
> +		fflags = O_CLOEXEC;
> +		type &= ~SOCK_CLOEXEC;

> +	/* Extract the close-on-exec flag.  */
> +	if ((type & SOCK_CLOEXEC) != 0) {
> +		fflags = O_CLOEXEC;
> +		type &= ~SOCK_CLOEXEC;
> +	}

Should probably be done once in one place in case we ever add future flags



-- 
--
	"Alan, I'm getting a bit worried about you."
				-- Linus Torvalds