Implement socket server code that properly handles async I/O, partial

messages, multiple messages per buffer and important things like
connection limits and file descriptor accounting.  It works with TCP
connections as well as switch(4).  The ofrelay.c part replaces
networking that was in ofp.c and will soon handle all socket
connections of switchd.  It is called "ofrelay" because it will be
used as client, server, and forwarder.

OK rzalamena@

1 files changed, 58 insertions, 196 deletions

diff --git a/usr.sbin/switchd/ofp.c b/usr.sbin/switchd/ofp.c
index 6bd4944bad4..ac1bb57f3d6 100644
--- a/usr.sbin/switchd/ofp.c
+++ b/usr.sbin/switchd/ofp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: ofp.c,v 1.9 2016/09/29 18:25:54 reyk Exp $	*/
+/*	$OpenBSD: ofp.c,v 1.10 2016/09/30 11:57:57 reyk Exp $	*/
 
 /*
  * Copyright (c) 2013-2016 Reyk Floeter <reyk@openbsd.org>
@@ -38,45 +38,27 @@
 #include "switchd.h"
 #include "ofp_map.h"
 
-int	 ofp_dispatch_control(int, struct privsep_proc *, struct imsg *);
 int	 ofp_dispatch_parent(int, struct privsep_proc *, struct imsg *);
+int	 ofp_dispatch_control(int, struct privsep_proc *, struct imsg *);
 void	 ofp_run(struct privsep *, struct privsep_proc *, void *);
 int	 ofp_add_device(struct switchd *, int, const char *);
 
-static unsigned int	 id = 0;
-
 static struct privsep_proc procs[] = {
 	{ "control",	PROC_CONTROL,	ofp_dispatch_control },
-	{ "parent",	PROC_PARENT,	ofp_dispatch_parent },
-	{ "ofcconn",	PROC_OFCCONN,	NULL }
+	{ "parent",	PROC_PARENT,	ofp_dispatch_parent }
 };
 
-static TAILQ_HEAD(, switch_connection) conn_head =
-    TAILQ_HEAD_INITIALIZER(conn_head);
-
 void
 ofp(struct privsep *ps, struct privsep_proc *p)
 {
-	struct switchd		*sc = ps->ps_env;
-	struct switch_server	*srv = &sc->sc_server;
-
-	if ((sc->sc_tap = switchd_tap()) == -1)
-		fatal("tap");
-
-	log_info("listen on %s", print_host(&srv->srv_addr, NULL, 0));
-
-	if ((srv->srv_fd = switchd_listen((struct sockaddr *)
-	    &srv->srv_addr)) == -1)
-		fatal("listen");
-
+	ofrelay(ps, p);
 	proc_run(ps, p, procs, nitems(procs), ofp_run, NULL);
 }
 
 void
 ofp_run(struct privsep *ps, struct privsep_proc *p, void *arg)
 {
-	struct switchd		*sc = ps->ps_env;
-	struct switch_server	*srv = &sc->sc_server;
+	struct switchd	*sc = ps->ps_env;
 
 	/*
 	 * pledge in the ofp process:
@@ -87,8 +69,10 @@ ofp_run(struct privsep *ps, struct privsep_proc *p, void *arg)
 	if (pledge("stdio inet recvfd", NULL) == -1)
 		fatal("pledge");
 
-	event_set(&srv->srv_ev, srv->srv_fd, EV_READ, ofp_accept, srv);
-	event_add(&srv->srv_ev, NULL);
+	TAILQ_INIT(&sc->sc_conns);
+	sc->sc_tap = -1;
+
+	ofrelay_run(ps, p, NULL);
 }
 
 int
@@ -107,51 +91,45 @@ ofp_dispatch_control(int fd, struct privsep_proc *p, struct imsg *imsg)
 int
 ofp_dispatch_parent(int fd, struct privsep_proc *p, struct imsg *imsg)
 {
-	struct switchd			*sc = p->p_ps->ps_env;
+	struct privsep			*ps = p->p_ps;
+	struct switchd			*sc = ps->ps_env;
+	struct sockaddr_un		 un;
 	struct switch_device		*sdv;
-	struct switch_connection	*c;
+	struct switch_connection	*con;
 
 	switch (imsg->hdr.type) {
+	case IMSG_TAPFD:
+		if (sc->sc_tap != -1)
+			close(sc->sc_tap);
+		sc->sc_tap = imsg->fd;
+		return (0);
 	case IMSG_CTL_DEVICE_CONNECT:
 	case IMSG_CTL_DEVICE_DISCONNECT:
-		if (IMSG_DATA_SIZE(imsg) < sizeof(*sdv)) {
-			log_warnx("%s: IMSG_CTL_DEVICE_CONNECT: "
-			    "message size is wrong", __func__);
+		IMSG_SIZE_CHECK(imsg, sdv);
+		sdv = imsg->data;
+
+		if (strlcpy(un.sun_path, sdv->sdv_device,
+		    sizeof(un.sun_path)) >= sizeof(un.sun_path)) {
+			log_warnx("invalid device: %s", sdv->sdv_device);
 			return (0);
 		}
-		sdv = imsg->data;
+		un.sun_family = AF_UNIX;
+		un.sun_len = sizeof(un);
+
 		if (imsg->hdr.type == IMSG_CTL_DEVICE_CONNECT)
-			ofp_add_device(sc, imsg->fd, sdv->sdv_device);
-		else {
-			TAILQ_FOREACH(c, &conn_head, con_next) {
-				if (c->con_peer.ss_family == AF_UNIX &&
-				    strcmp(sdv->sdv_device,
-				    ((struct sockaddr_un *)&c->con_peer)
-				    ->sun_path) == 0)
-					break;
-			}
-			if (c)
-				ofp_close(c);
-		}
+			ofrelay_attach(&sc->sc_server,
+			    imsg->fd, (struct sockaddr *)&un);
+		else if ((con =
+		    switchd_connbyaddr(sc, (struct sockaddr *)&un)) != NULL)
+			ofp_close(con);
 		return (0);
 	default:
-
 		break;
 	}
 
 	return (-1);
 }
 
-void
-ofp_close(struct switch_connection *con)
-{
-	log_info("%s: connection %u closed", __func__, con->con_id);
-	event_del(&con->con_ev);
-	switch_remove(con->con_sc, con->con_switch);
-	close(con->con_fd);
-	TAILQ_REMOVE(&conn_head, con, con_next);
-}
-
 int
 ofp_validate_header(struct switchd *sc,
     struct sockaddr_storage *src, struct sockaddr_storage *dst,
@@ -186,46 +164,25 @@ ofp_validate_header(struct switchd *sc,
 	return (0);
 }
 
-void
-ofp_read(int fd, short event, void *arg)
+int
+ofp_input(struct switch_connection *con, struct ibuf *ibuf)
 {
-	uint8_t			 buf[SWITCHD_READ_BUFFER];
-	struct switch_connection *con = arg;
-	struct switch_control	*sw;
 	struct switchd		*sc = con->con_sc;
 	struct ofp_header	*oh;
-	ssize_t			 len;
-	struct ibuf		*ibuf = NULL;
-
-	event_add(&con->con_ev, NULL);
-	if ((event & EV_TIMEOUT))
-		goto fail;
-
-	if ((len = read(fd, buf, sizeof(buf))) == -1)
-		goto fail;
-	if (len == 0)
-		goto fail;
-
-	if ((ibuf = ibuf_new(buf, len)) == NULL)
-		goto fail;
-
-	sw = con->con_switch;
-	log_debug("%s: connection %d: %ld bytes from switch %u", __func__,
-	    con->con_id, len, sw == NULL ? 0 : sw->sw_id);
 
 	if ((oh = ibuf_seek(ibuf, 0, sizeof(*oh))) == NULL) {
 		log_debug("short header");
-		goto fail;
+		return (-1);
 	}
 
 	switch (oh->oh_version) {
 	case OFP_V_1_0:
 		if (ofp10_input(sc, con, oh, ibuf) != 0)
-			goto fail;
+			return (-1);
 		break;
 	case OFP_V_1_3:
 		if (ofp13_input(sc, con, oh, ibuf) != 0)
-			goto fail;
+			return (-1);
 		break;
 	case OFP_V_1_1:
 	case OFP_V_1_2:
@@ -234,147 +191,52 @@ ofp_read(int fd, short event, void *arg)
 		(void)ofp10_validate(sc,
 		    &con->con_peer, &con->con_local, oh, ibuf);
 		ofp10_hello(sc, con, oh, ibuf);
-		goto fail;
+		return (-1);
 	}
 
-	ibuf_release(ibuf);
-	return;
-
- fail:
-	ibuf_release(ibuf);
-	ofp_close(con);
+	return (0);
 }
 
 int
 ofp_output(struct switch_connection *con, struct ofp_header *oh,
     struct ibuf *obuf)
 {
-	struct iovec		 iov[2];
-	int			 cnt = 0;
-	void			*data;
-	ssize_t			 len;
-
-	if (oh != NULL) {
-		iov[cnt].iov_base = oh;
-		iov[cnt++].iov_len = sizeof(*oh);
-	}
+	struct ibuf	*buf;
 
-	if (ibuf_length(obuf)) {
-		if (oh != NULL && (ibuf_seek(obuf, 0, sizeof(*oh)) == NULL))
-			return (-1);
-		len = ibuf_dataleft(obuf);
-		if (len < 0) {
-			return (-1);
-		} else if (len > 0 &&
-		    (data = ibuf_getdata(obuf, len)) != NULL) {
-			iov[cnt].iov_base = data;
-			iov[cnt++].iov_len = len;
-		}
-	}
-
-	if (cnt == 0)
+	if ((buf = ibuf_static()) == NULL)
 		return (-1);
-
-	/* XXX */
-	if (writev(con->con_fd, iov, cnt) == -1)
+	if ((oh != NULL) &&
+	    (ibuf_add(buf, oh, sizeof(*oh)) == -1)) {
+		ibuf_release(buf);
 		return (-1);
+	}
+	if ((obuf != NULL) &&
+	    (ibuf_cat(buf, obuf) == -1)) {
+		ibuf_release(buf);
+		return (-1);
+	}
+	ibuf_close(&con->con_wbuf, buf);
 
 	return (0);
 }
 
 int
-ofp_add_device(struct switchd *sc, int fd, const char *name)
+ofp_open(struct privsep *ps, struct switch_connection *con)
 {
-	struct switch_connection	*con = NULL;
-	struct sockaddr_un		*sun;
-	struct switch_control		*sw;
-
-	if ((con = calloc(1, sizeof(*con))) == NULL) {
-		log_warn("calloc");
-		goto fail;
-	}
-	con->con_fd = fd;
-	con->con_sc = sc;
-	con->con_id = ++id;
-	sun = (struct sockaddr_un *)&con->con_peer;
-	sun->sun_family = AF_LOCAL;
-	strlcpy(sun->sun_path, name, sizeof(sun->sun_path));
+	struct switch_control	*sw;
 
 	/* Get associated switch, if it exists */
 	sw = switch_get(con);
 
-	log_info("%s: new device %u (%s) from switch %u",
-	    __func__, con->con_id, name, sw == NULL ? 0 : sw->sw_id);
-
-	bzero(&con->con_ev, sizeof(con->con_ev));
-	event_set(&con->con_ev, con->con_fd, EV_READ, ofp_read, con);
-	event_add(&con->con_ev, NULL);
-
-	TAILQ_INSERT_TAIL(&conn_head, con, con_next);
+	log_info("%s: new connection %u.%u from switch %u",
+	    __func__, con->con_id, con->con_instance,
+	    sw == NULL ? 0 : sw->sw_id);
 
 	return (0);
-fail:
-	if (fd != -1)
-		close(fd);
-	free(con);
-
-	return (-1);
 }
 
 void
-ofp_accept(int fd, short event, void *arg)
+ofp_close(struct switch_connection *con)
 {
-	struct switch_server	*server = arg;
-	struct switch_connection *con = NULL;
-	struct switchd		*sc = server->srv_sc;
-	struct switch_control	*sw;
-	struct sockaddr_storage	 ss;
-	socklen_t		 slen;
-	int			 s;
-
-	event_add(&server->srv_ev, NULL);
-	if ((event & EV_TIMEOUT))
-		return;
-
-	/* XXX accept_reserve() */
-	slen = sizeof(ss);
-	if ((s = accept(fd, (struct sockaddr *)&ss, &slen)) == -1) {
-		log_warn("accept");
-		goto fail;
-	}
-
-	if ((con = calloc(1, sizeof(*con))) == NULL) {
-		log_warn("calloc");
-		goto fail;
-	}
-
-	slen = sizeof(con->con_local);
-	if (getsockname(s, (struct sockaddr *)&con->con_local, &slen) == -1) {
-		log_warn("getsockname");
-		goto fail;
-	}
-
-	con->con_fd = s;
-	con->con_sc = sc;
-	con->con_id = ++id;
-	con->con_port = htons(socket_getport(&ss));
-	memcpy(&con->con_peer, &ss, sizeof(ss));
-
-	/* Get associated switch, if it exists */
-	sw = switch_get(con);
-
-	log_info("%s: new connection %u from switch %u",
-	    __func__, con->con_id, sw == NULL ? 0 : sw->sw_id);
-
-	bzero(&con->con_ev, sizeof(con->con_ev));
-	event_set(&con->con_ev, con->con_fd, EV_READ, ofp_read, con);
-	event_add(&con->con_ev, NULL);
-
-	TAILQ_INSERT_TAIL(&conn_head, con, con_next);
-
-	return;
- fail:
-	if (s != -1)
-		close(s);
-	free(con);
+	ofrelay_close(con);
 }