lkl: introduce raw socket based netdev backend

This patch introduces new backend for virtio net, which uses AF_PACKET
socket (a.k.a. raw socket) to bypass host kernel and uses LKL network
stack instead.  it is convinient since we don't have to add additional
net_device (e.g., tap) for LKL, and possibly faster than tuntap with
PACKET_QDISC_BYPASS socket option (available after Linux 3.14).  One
drawback is it requires root privilege (sudo or suid bit on) to use
this.

example usage is like this:

sudo LKL_HIJACK_NET_IFTYPE=raw LKL_HIJACK_NET_IFPARAMS=docker0 \
LKL_HIJACK_NET_IP=172.17.0.39 LKL_HIJACK_NET_NETMASK_LEN=24 \
./bin/lkl-hijack.sh ping  172.17.0.2

some benchmarks with netperf:

- TCP_RR
 raw(QDISC_BYPASS): 9519.31 Trans/sec
 tap:               9486.03 Trans/sec
- TCP_STREAM
 raw(QDISC_BYPASS): 2184.79 Mbps
 tap:               2130.39 Mbps
- UDP_STREAM
 raw(QDISC_BYPASS): 3654.32 Mbps
 tap:               3108.10 Mbps

Signed-off-by: Hajime Tazaki <thehajime@gmail.com>

tools/lkl/include/lkl.h

@@ -301,6 +301,14 @@ struct lkl_netdev *lkl_netdev_dpdk_create(const char *ifname);
  */
 struct lkl_netdev *lkl_netdev_vde_create(const char *switch_path);
 
+/**
+ * lkl_netdev_raw_create - create raw socket net_device for the virtio net
+ *                         backend
+ *
+ * @ifname - interface name for the snoop device.
+ */
+struct lkl_netdev *lkl_netdev_raw_create(const char *ifname);
+
 /*
  * lkl_register_dbg_handler- register a signal handler that loads a debug lib.
  *

tools/lkl/lib/Build

@@ -14,5 +14,6 @@ lkl-y += dbg_handler.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_linux_fdnet.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_tap.o
+lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_raw.o
 lkl-$(CONFIG_AUTO_LKL_VIRTIO_NET_DPDK) += virtio_net_dpdk.o
 lkl-$(CONFIG_AUTO_LKL_VIRTIO_NET_VDE) += virtio_net_vde.o

tools/lkl/lib/hijack/hijack.c

@@ -158,7 +158,7 @@ HOST_CALL(socket);
 int socket(int domain, int type, int protocol)
 {
 	CHECK_HOST_CALL(socket);
-	if (domain == AF_UNIX)
+	if (domain == AF_UNIX || domain == PF_PACKET)
 		return host_socket(domain, type, protocol);
 
 	return lkl_call(__lkl__NR_socket, 3, domain, type, protocol);

tools/lkl/lib/hijack/init.c

@@ -214,6 +214,8 @@ hijack_init(void)
 			nd = lkl_netdev_dpdk_create(ifparams);
 		else if (strcmp(iftype, "vde") == 0)
 			nd = lkl_netdev_vde_create(ifparams);
+		else if (strcmp(iftype, "raw") == 0)
+			nd = lkl_netdev_raw_create(ifparams);
 	}
 
 	if (nd) {

tools/lkl/lib/virtio_net_raw.c

@@ -0,0 +1,69 @@
+/*
+ * raw socket based virtual network interface feature for LKL
+ * Copyright (c) 2015,2016 Ryo Nakamura, Hajime Tazaki
+ *
+ * Author: Ryo Nakamura <upa@wide.ad.jp>
+ *         Hajime Tazaki <thehajime@gmail.com>
+ *
+ * Current implementation is linux-specific.
+ */
+
+#include <stdio.h>
+#include <errno.h>
+#include <string.h>
+#include <unistd.h>
+#include <net/if.h>
+#include <linux/if_ether.h>
+#include <linux/if_packet.h>
+#include <arpa/inet.h>
+#include <fcntl.h>
+
+#include "virtio.h"
+#include "virtio_net_linux_fdnet.h"
+
+/* since Linux 3.14 (man 7 packet) */
+#ifndef PACKET_QDISC_BYPASS
+#define PACKET_QDISC_BYPASS 20
+#endif
+
+struct lkl_netdev *lkl_netdev_raw_create(const char *ifname)
+{
+	struct lkl_netdev_linux_fdnet *nd;
+	int ret;
+	struct sockaddr_ll ll;
+	int fd, fd_flags, val;
+
+	fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
+	if (fd < 0) {
+		perror("socket");
+		return NULL;
+	}
+
+	memset(&ll, 0, sizeof(ll));
+	ll.sll_family = PF_PACKET;
+	ll.sll_ifindex = if_nametoindex(ifname);
+	ll.sll_protocol = htons(ETH_P_ALL);
+	ret = bind(fd, (struct sockaddr *)&ll, sizeof(ll));
+	if (ret) {
+		perror("bind");
+		close(fd);
+		return NULL;
+	}
+
+	val = 1;
+	ret = setsockopt(fd, SOL_PACKET, PACKET_QDISC_BYPASS, &val,
+			 sizeof(val));
+	if (ret)
+		perror("PACKET_QDISC_BYPASS, ignoring");
+
+	fd_flags = fcntl(fd, F_GETFD, NULL);
+	fcntl(fd, F_SETFL, fd_flags | O_NONBLOCK);
+
+	nd = lkl_register_netdev_linux_fdnet(fd);
+	if (!nd) {
+		perror("failed to register to.");
+		return NULL;
+	}
+
+	return (struct lkl_netdev *)nd;
+}

tools/lkl/tests/net-test.c

@@ -89,7 +89,7 @@ static int test_icmp(char *str, int len)
 		return TEST_FAILURE;
 	}
 
-	ret = lkl_sys_recv(sock, buf, sizeof(buf), 0);
+	ret = lkl_sys_recv(sock, buf, sizeof(buf), MSG_DONTWAIT);
 	if (ret < 0) {
 		snprintf(str, len, "recv error (%s)", strerror(errno));
 		return TEST_FAILURE;
@@ -116,7 +116,7 @@ static int test_net_init(int argc, char **argv)
 	struct lkl_netdev *nd = NULL;
 
 	if (argc < 6) {
-		printf("usage %s <iftype: tap|dpdk> <ifname> <v4addr> <v4mask> <dstaddr> [gateway]\n", argv[0]);
+		printf("usage %s <iftype: tap|dpdk|raw> <ifname> <v4addr> <v4mask> <dstaddr> [gateway]\n", argv[0]);
 		exit(0);
 	}
 
@@ -127,14 +127,16 @@ static int test_net_init(int argc, char **argv)
 	dst = argv[5];
 
 	if (argc == 7)
-		gateway = argv[5];
+		gateway = argv[6];
 
 	if (iftype && ifname && (strncmp(iftype, "tap", 3) == 0))
 		nd = lkl_netdev_tap_create(ifname);
 #ifdef CONFIG_AUTO_LKL_VIRTIO_NET_DPDK
 	else if (iftype && ifname && (strncmp(iftype, "dpdk", 4) == 0))
 		nd = lkl_netdev_dpdk_create(ifname);
 #endif /* CONFIG_AUTO_LKL_VIRTIO_NET_DPDK */
+	else if (iftype && ifname && (strncmp(iftype, "raw", 3) == 0))
+		nd = lkl_netdev_raw_create(ifname);
 
 	if (!nd) {
 		fprintf(stderr, "init netdev failed\n");

tools/lkl/tests/net.sh

@@ -1,5 +1,10 @@
 #!/bin/bash -e
 
+IFNAME=`ip route |grep default | awk '{print $5}'`
+GW=`ip route |grep default | awk '{print $3}'`
+IPADDR=`echo $GW | sed -r "s/([0-9]+\.[0-9]+\.[0-9]+\.)([0-9]+)$/\1\`expr \2 + 10\`/"`
+PLEN=`ip rou |grep ${IFNAME} | grep "scope link" | awk '{print $1}' | sed "s/.*\/\(.*\)/\1/"`
+
 echo "== TAP (LKL net) tests =="
 if [ -c /dev/net/tun ]; then
     sudo ip link set dev lkl_ptt1 down || true
@@ -14,6 +19,15 @@ if [ -c /dev/net/tun ]; then
     sudo ip tuntap del dev lkl_ptt1 mode tap
 fi
 
+echo "== RAW socket (LKL net) tests =="
+# currently not supported mingw
+if [ -n "`printenv CONFIG_AUTO_LKL_POSIX_HOST`" ] ; then
+    sudo ip link set dev ${IFNAME} promisc on
+    # this won't work if IFNAME is wifi since it rewrites the src macaddr
+    sudo ./net-test raw ${IFNAME} ${IPADDR} ${PLEN} 8.8.8.8 ${GW}
+    sudo ip link set dev ${IFNAME} promisc off
+fi
+
 # we disabled this DPDK test because it's unlikely possible to describe
 # a generic set of commands for all environments to test with DPDK.  users
 # may customize those test commands for your host.