tcp: Avoid SO_ACCEPTCONN getsockopt() by noting listening/data sockets numbers

...the rest is reshuffling existing macros to use the bits we need in
TCP code.

Signed-off-by: Stefano Brivio <sbrivio@redhat.com>

5 files changed, 87 insertions, 56 deletions

diff --git a/passt.c b/passt.c
index 2fc88cf..3a56f67 100644
--- a/passt.c
+++ b/passt.c
@@ -613,40 +613,39 @@ static int tap_handler(struct ctx *c)
 /**
  * sock_handler() - Event handler for L4 sockets
  * @c:		Execution context
- * @fd:		File descriptor associated to event
+ * @s:		Socket associated to event
  * @events	epoll events
  */
-static void sock_handler(struct ctx *c, int fd, uint32_t events)
+static void sock_handler(struct ctx *c, int s, uint32_t events)
 {
 	socklen_t sl;
-	int so;
-
-	sl = sizeof(so);
-
-#define IN(x, proto)	(x >= c->proto.fd_min && x <= c->proto.fd_max)
-
-	if (IN(fd, udp)		&& !IN(fd, icmp)	&& !IN(fd, tcp))
-		so = IPPROTO_UDP;
-	else if (IN(fd, tcp)	&& !IN(fd, icmp)	&& !IN(fd, udp))
-		so = IPPROTO_TCP;
-	else if (IN(fd, icmp)	&& !IN(fd, udp)		&& !IN(fd, tcp))
-		so = IPPROTO_ICMP;	/* Fits ICMPv6 below, too */
-	else if (getsockopt(fd, SOL_SOCKET, SO_PROTOCOL, &so, &sl)) {
-		epoll_ctl(c->epollfd, EPOLL_CTL_DEL, fd, NULL);
-		close(fd);
+	int proto;
+
+	sl = sizeof(proto);
+
+	if (    FD_PROTO(s, udp)   && !FD_PROTO(s, icmp) && !FD_PROTO(s, tcp))
+		proto = IPPROTO_UDP;
+	else if (FD_PROTO(s, tcp)  && !FD_PROTO(s, icmp) && !FD_PROTO(s, udp))
+		proto = IPPROTO_TCP;
+	else if (FD_PROTO(s, icmp) && !FD_PROTO(s, udp)  && !FD_PROTO(s, tcp))
+		proto = IPPROTO_ICMP;	/* Fits ICMPv6 below, too */
+	else if (getsockopt(s, SOL_SOCKET, SO_PROTOCOL, &proto, &sl))
+		proto = -1;
+
+	if (proto == -1) {
+		epoll_ctl(c->epollfd, EPOLL_CTL_DEL, s, NULL);
+		close(s);
 		return;
 	}
 
-#undef IN
-
-	debug("%s: packet from socket %i", getprotobynumber(so)->p_name, fd);
+	debug("%s: packet from socket %i", getprotobynumber(proto)->p_name, s);
 
-	if (so == IPPROTO_ICMP || so == IPPROTO_ICMPV6)
-		icmp_sock_handler(c, fd, events);
-	else if (so == IPPROTO_TCP)
-		tcp_sock_handler(c, fd, events);
-	else if (so == IPPROTO_UDP)
-		udp_sock_handler(c, fd, events);
+	if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6)
+		icmp_sock_handler(c, s, events);
+	else if (proto == IPPROTO_TCP)
+		tcp_sock_handler(c, s, events);
+	else if (proto == IPPROTO_UDP)
+		udp_sock_handler(c, s, events);
 }
 
 /**
diff --git a/tcp.c b/tcp.c
index 1713dbc..5ba8f3f 100644
--- a/tcp.c
+++ b/tcp.c
@@ -1003,10 +1003,8 @@ static void tcp_conn_from_sock(struct ctx *c, int fd)
 	if (s == -1)
 		return;
 
-	if (s < c->tcp.fd_min)
-		c->tcp.fd_min = s;
-	if (s > c->tcp.fd_max)
-		c->tcp.fd_max = s;
+	CHECK_SET_MIN_MAX(c->tcp.fd_, s);
+	CHECK_SET_MIN_MAX(c->tcp.fd_conn_, s);
 
 	if (sa_l.ss_family == AF_INET) {
 		struct sockaddr_in *sa4 = (struct sockaddr_in *)&sa_r;
@@ -1406,7 +1404,7 @@ static void tcp_connect_finish(struct ctx *c, int s)
 void tcp_sock_handler(struct ctx *c, int s, uint32_t events)
 {
 	socklen_t sl;
-	int so;
+	int accept;
 
 	if (tc[s].s == LAST_ACK) {
 		tcp_send_to_tap(c, s, ACK, NULL, 0);
@@ -1414,21 +1412,28 @@ void tcp_sock_handler(struct ctx *c, int s, uint32_t events)
 		return;
 	}
 
+
 	if (tc[s].s == SOCK_SYN_SENT) {
 		/* This can only be a socket error or a shutdown from remote */
 		tcp_rst(c, s);
 		return;
 	}
-
-	sl = sizeof(so);
-	if ((events & EPOLLERR) ||
-	    getsockopt(s, SOL_SOCKET, SO_ACCEPTCONN, &so, &sl)) {
+	if (IN_INTERVAL(c->tcp.fd_listen_min, c->tcp.fd_listen_max, s) &&
+	    !IN_INTERVAL(c->tcp.fd_conn_min, c->tcp.fd_conn_max, s))
+		accept = 1;
+	else if (IN_INTERVAL(c->tcp.fd_conn_min, c->tcp.fd_conn_max, s) &&
+		 !IN_INTERVAL(c->tcp.fd_listen_min, c->tcp.fd_listen_max, s))
+		accept = 0;
+	else if (getsockopt(s, SOL_SOCKET, SO_ACCEPTCONN, &accept, &sl))
+		accept = -1;
+
+	if ((events & EPOLLERR) || accept == -1) {
 		if (tc[s].s != CLOSED)
 			tcp_rst(c, s);
 		return;
 	}
 
-	if (so) {
+	if (accept) {
 		tcp_conn_from_sock(c, s);
 		return;
 	}
@@ -1466,15 +1471,24 @@ void tcp_sock_handler(struct ctx *c, int s, uint32_t events)
 int tcp_sock_init(struct ctx *c)
 {
 	in_port_t port;
+	int s = 0;
 
-	c->tcp.fd_min = INT_MAX;
-	c->tcp.fd_max = 0;
+	c->tcp.fd_min = c->tcp.fd_listen_min = c->tcp.fd_conn_min = INT_MAX;
+	c->tcp.fd_max = c->tcp.fd_listen_max = c->tcp.fd_conn_max = 0;
+	CHECK_SET_MIN_MAX(c->tcp.fd_listen_, s);
 
 	for (port = 0; port < (1 << 15) + (1 << 14); port++) {
-		if (c->v4 && sock_l4_add(c, 4, IPPROTO_TCP, port) < 0)
-			return -1;
-		if (c->v6 && sock_l4_add(c, 6, IPPROTO_TCP, port) < 0)
-			return -1;
+		if (c->v4) {
+			if ((s = sock_l4_add(c, 4, IPPROTO_TCP, port)) < 0)
+				return -1;
+			CHECK_SET_MIN_MAX(c->tcp.fd_listen_, s);
+		}
+
+		if (c->v6) {
+			if ((s = sock_l4_add(c, 6, IPPROTO_TCP, port)) < 0)
+				return -1;
+			CHECK_SET_MIN_MAX(c->tcp.fd_listen_, s);
+		}
 	}
 
 	getrandom(&c->tcp.hash_secret, sizeof(c->tcp.hash_secret), GRND_RANDOM);
diff --git a/tcp.h b/tcp.h
index 2a79a75..4604281 100644
--- a/tcp.h
+++ b/tcp.h
@@ -14,11 +14,19 @@ void tcp_timer(struct ctx *c, struct timespec *ts);
  * @hash_secret:	128-bit secret for hash functions, ISN and hash table
  * @fd_min:		Lowest file descriptor number for TCP ever used
  * @fd_max:		Highest file descriptor number for TCP ever used
+ * @fd_listen_min:	Lowest file descriptor number for listening sockets
+ * @fd_listen_max:	Highest file descriptor number for listening sockets
+ * @fd_conn_min:	Lowest file descriptor number for connected sockets
+ * @fd_conn_max:	Highest file descriptor number for connected sockets
  */
 struct tcp_ctx {
 	uint64_t hash_secret[2];
 	int fd_min;
 	int fd_max;
+	int fd_listen_min;
+	int fd_listen_max;
+	int fd_conn_min;
+	int fd_conn_max;
 };
 
 #endif /* TCP_H */
diff --git a/util.c b/util.c
index cc96a1a..acd62a9 100644
--- a/util.c
+++ b/util.c
@@ -25,6 +25,7 @@
 #include <stdarg.h>
 
 #include "passt.h"
+#include "util.h"
 
 #define logfn(name, level)						\
 void name(const char *format, ...) {					\
@@ -189,20 +190,10 @@ int sock_l4_add(struct ctx *c, int v, uint16_t proto, uint16_t port)
 		return -1;
 	}
 
-#define CHECK_SET_MIN_MAX(ipproto, proto_ctx, fd)			\
-	if (proto == (ipproto)) {					\
-		if (fd < c->proto_ctx.fd_min)				\
-			c->proto_ctx.fd_min = (fd);			\
-		if (fd > c->proto_ctx.fd_max)				\
-			c->proto_ctx.fd_max = (fd);			\
-	}
-
-	CHECK_SET_MIN_MAX(IPPROTO_ICMP,		icmp,	fd);
-	CHECK_SET_MIN_MAX(IPPROTO_ICMPV6,	icmp,	fd);
-	CHECK_SET_MIN_MAX(IPPROTO_TCP,		tcp,	fd);
-	CHECK_SET_MIN_MAX(IPPROTO_UDP,		udp,	fd);
-
-#undef CHECK_SET_MIN_MAX
+	CHECK_SET_MIN_MAX_PROTO_FD(proto, IPPROTO_ICMP,		icmp,	fd);
+	CHECK_SET_MIN_MAX_PROTO_FD(proto, IPPROTO_ICMPV6,	icmp,	fd);
+	CHECK_SET_MIN_MAX_PROTO_FD(proto, IPPROTO_TCP,		tcp,	fd);
+	CHECK_SET_MIN_MAX_PROTO_FD(proto, IPPROTO_UDP,		udp,	fd);
 
 	if (proto == IPPROTO_ICMP || proto == IPPROTO_ICMPV6)
 		goto epoll_add;
diff --git a/util.h b/util.h
index 139638c..34b8266 100644
--- a/util.h
+++ b/util.h
@@ -8,6 +8,25 @@ void debug(const char *format, ...);
 #define debug(...) { }
 #endif
 
+#define CHECK_SET_MIN_MAX(basename, fd)					\
+	do {								\
+		if ((fd) < basename##min)				\
+			basename##min = (fd);				\
+		if ((fd) > basename##max)				\
+			basename##max = (fd);				\
+	} while (0)
+
+#define CHECK_SET_MIN_MAX_PROTO_FD(proto, ipproto, proto_ctx, fd)	\
+	do {								\
+		if ((proto) == (ipproto))				\
+			CHECK_SET_MIN_MAX(c->proto_ctx.fd_, (fd));	\
+	} while (0)
+
+#define IN_INTERVAL(a, b, x) ((x) >= (a) && (x) <= (b))
+
+#define FD_PROTO(x, proto)						\
+	(IN_INTERVAL(c->proto.fd_min, c->proto.fd_max, (x)))
+
 uint16_t csum_fold(uint32_t sum);
 uint16_t csum_ip4(void *buf, size_t len);
 void csum_tcp4(struct iphdr *iph);