Function req_reply

hsmd/hsmd.c:252–277

This is the common pattern for the tail of each handler in this file. */

/* This is the common pattern for the tail of each handler in this file. */
static struct io_plan *req_reply(struct io_conn *conn,
				 struct client *c,
				 const u8 *msg_out TAKES)
{
	/*~ Write this out, then read the next one.  This works perfectly for
	 * a simple request/response system like this.
	 *
	 * Internally, the ccan/io subsystem gathers all the file descriptors,
	 * figures out which want to write and read, asks the OS which ones
	 * are available, and for those file descriptors, tries to do the
	 * reads/writes we've asked it.  It handles retry in the case where a
	 * read or write is done partially.
	 *
	 * Since the OS does buffering internally (on my system, over 100k
	 * worth) writes will normally succeed immediately.  However, if the
	 * client is slow or malicious, and doesn't read from the socket as
	 * fast as we're writing, eventually the socket buffer will fill up;
	 * we don't care, because ccan/io will wait until there's room to
	 * write this reply before it will read again.  The client just hurts
	 * themselves, and there's no Denial of Service on us.
	 *
	 * If we were to queue outgoing messages ourselves, we *would* have to
	 * consider such scenarios; this is why our daemons generally avoid
	 * buffering from untrusted parties. */
	return io_write_wire(conn, msg_out, client_read_next, c);
}

/*~ This encrypts the content of the `struct secret hsm_secret` and
 * stores it in hsm_secret, this is called instead of create_hsm() if