Merge pull request torvalds#383 from motomuman/pipe-netfd-network

lkl: introduce PIPE network backend for virtio device

tools/lkl/include/lkl.h

@@ -450,6 +450,16 @@ struct lkl_netdev *lkl_netdev_raw_create(const char *ifname);
  */
 struct lkl_netdev *lkl_netdev_macvtap_create(const char *path, int offload);
 
+/**
+ * lkl_netdev_pipe_create - create pipe net_device for the virtio
+ * net backend
+ *
+ * @ifname - a file name for the rx and tx pipe device. need to be configured
+ * on host in advance. delimiter is "|". e.g. "rx_name|tx_name".
+ * @offload - offload bits for the device
+ */
+struct lkl_netdev *lkl_netdev_pipe_create(char *ifname, int offload);
+
 /*
  * lkl_register_dbg_handler- register a signal handler that loads a debug lib.
  *

tools/lkl/lib/Build

@@ -20,3 +20,4 @@ lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_raw.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_macvtap.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_dpdk.o
 lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_vde.o
+lkl-$(CONFIG_AUTO_LKL_POSIX_HOST) += virtio_net_pipe.o

tools/lkl/lib/hijack/init.c

@@ -284,6 +284,8 @@ hijack_init(void)
 			nd = lkl_netdev_macvtap_create(ifparams, offload);
 		} else if ((strcmp(iftype, "dpdk") == 0)) {
 			nd = lkl_netdev_dpdk_create(ifparams, offload, mac);
+		} else if ((strcmp(iftype, "pipe") == 0)) {
+			nd = lkl_netdev_pipe_create(ifparams, offload);
 		} else {
 			if (offload) {
 				fprintf(stderr,

tools/lkl/lib/virtio_net_fd.c

@@ -23,7 +23,8 @@
 struct lkl_netdev_fd {
 	struct lkl_netdev dev;
 	/* file-descriptor based device */
-	int fd;
+	int fd_rx;
+	int fd_tx;
 	/*
 	 * Controlls the poll mask for fd. Can be acccessed concurrently from
 	 * poll, tx, or rx routines but there is no need for syncronization
@@ -48,7 +49,7 @@ static int fd_net_tx(struct lkl_netdev *nd, struct iovec *iov, int cnt)
 		container_of(nd, struct lkl_netdev_fd, dev);
 
 	do {
-		ret = writev(nd_fd->fd, iov, cnt);
+		ret = writev(nd_fd->fd_tx, iov, cnt);
 	} while (ret == -1 && errno == EINTR);
 
 	if (ret < 0) {
@@ -72,7 +73,7 @@ static int fd_net_rx(struct lkl_netdev *nd, struct iovec *iov, int cnt)
 		container_of(nd, struct lkl_netdev_fd, dev);
 
 	do {
-		ret = readv(nd_fd->fd, (struct iovec *)iov, cnt);
+		ret = readv(nd_fd->fd_rx, (struct iovec *)iov, cnt);
 	} while (ret == -1 && errno == EINTR);
 
 	if (ret < 0) {
@@ -93,9 +94,12 @@ static int fd_net_poll(struct lkl_netdev *nd)
 {
 	struct lkl_netdev_fd *nd_fd =
 		container_of(nd, struct lkl_netdev_fd, dev);
-	struct pollfd pfds[2] = {
+	struct pollfd pfds[3] = {
 		{
-			.fd = nd_fd->fd,
+			.fd = nd_fd->fd_rx,
+		},
+		{
+			.fd = nd_fd->fd_tx,
 		},
 		{
 			.fd = nd_fd->pipe[0],
@@ -107,21 +111,21 @@ static int fd_net_poll(struct lkl_netdev *nd)
 	if (nd_fd->poll_rx)
 		pfds[0].events |= POLLIN|POLLPRI;
 	if (nd_fd->poll_tx)
-		pfds[0].events |= POLLOUT;
+		pfds[1].events |= POLLOUT;
 
 	do {
-		ret = poll(pfds, 2, -1);
+		ret = poll(pfds, 3, -1);
 	} while (ret == -1 && errno == EINTR);
 
 	if (ret < 0) {
 		perror("virtio net fd poll");
 		return 0;
 	}
 
-	if (pfds[1].revents & (POLLHUP|POLLNVAL))
+	if (pfds[2].revents & (POLLHUP|POLLNVAL))
 		return LKL_DEV_NET_POLL_HUP;
 
-	if (pfds[1].revents & POLLIN) {
+	if (pfds[2].revents & POLLIN) {
 		char tmp[PIPE_BUF];
 
 		ret = read(nd_fd->pipe[0], tmp, PIPE_BUF);
@@ -138,7 +142,7 @@ static int fd_net_poll(struct lkl_netdev *nd)
 		ret |= LKL_DEV_NET_POLL_RX;
 	}
 
-	if (pfds[0].revents & POLLOUT) {
+	if (pfds[1].revents & POLLOUT) {
 		nd_fd->poll_tx = 0;
 		ret |= LKL_DEV_NET_POLL_TX;
 	}
@@ -161,7 +165,8 @@ static void fd_net_free(struct lkl_netdev *nd)
 	struct lkl_netdev_fd *nd_fd =
 		container_of(nd, struct lkl_netdev_fd, dev);
 
-	close(nd_fd->fd);
+	close(nd_fd->fd_rx);
+	close(nd_fd->fd_tx);
 	free(nd_fd);
 }
 
@@ -173,7 +178,7 @@ struct lkl_dev_net_ops fd_net_ops =  {
 	.free = fd_net_free,
 };
 
-struct lkl_netdev *lkl_register_netdev_fd(int fd)
+struct lkl_netdev *lkl_register_netdev_fd(int fd_rx, int fd_tx)
 {
 	struct lkl_netdev_fd *nd;
 
@@ -186,7 +191,8 @@ struct lkl_netdev *lkl_register_netdev_fd(int fd)
 
 	memset(nd, 0, sizeof(*nd));
 
-	nd->fd = fd;
+	nd->fd_rx = fd_rx;
+	nd->fd_tx = fd_tx;
 	if (pipe(nd->pipe) < 0) {
 		perror("pipe");
 		free(nd);

tools/lkl/lib/virtio_net_fd.h

@@ -7,10 +7,11 @@ struct ifreq;
  * lkl_register_netdev_linux_fdnet - register a file descriptor-based network
  * device as a NIC
  *
- * @fd - a POSIX file descriptor number for input/output
+ * @fd_rx - a POSIX file descriptor number for input
+ * @fd_tx - a POSIX file descriptor number for output
  * @returns a struct lkl_netdev_linux_fdnet entry for virtio-net
  */
-struct lkl_netdev *lkl_register_netdev_fd(int fd);
+struct lkl_netdev *lkl_register_netdev_fd(int fd_rx, int fd_tx);
 
 
 /**

tools/lkl/lib/virtio_net_pipe.c

@@ -0,0 +1,68 @@
+/*
+ * pipe based virtual network interface feature for LKL
+ * Copyright (c) 2017,2016 Motomu Utsumi
+ *
+ * Author: Motomu Utsumi <motomuman@gmail.com>
+ *
+ * Current implementation is linux-specific.
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <string.h>
+#include <fcntl.h>
+
+#include "virtio.h"
+#include "virtio_net_fd.h"
+
+struct lkl_netdev *lkl_netdev_pipe_create(char *ifname, int offload)
+{
+	struct lkl_netdev *nd;
+	int fd_rx, fd_tx;
+	char *ifname_rx = NULL, *ifname_tx = NULL;
+
+	ifname_rx = strtok(ifname, "|");
+	if (ifname_rx == NULL) {
+		fprintf(stderr, "invalid ifname format: %s\n", ifname);
+		return NULL;
+	}
+
+	ifname_tx = strtok(NULL, "|");
+	if (ifname_tx == NULL) {
+		fprintf(stderr, "invalid ifname format: %s\n", ifname);
+		return NULL;
+	}
+
+	if (strtok(NULL, "|") != NULL) {
+		fprintf(stderr, "invalid ifname format: %s\n", ifname);
+		return NULL;
+	}
+
+	fd_rx = open(ifname_rx, O_RDWR|O_NONBLOCK);
+	if (fd_rx < 0) {
+		perror("can not open ifname_rx pipe");
+		return NULL;
+	}
+
+	fd_tx = open(ifname_tx, O_RDWR|O_NONBLOCK);
+	if (fd_tx < 0) {
+		perror("can not open ifname_tx pipe");
+		close(fd_rx);
+		return NULL;
+	}
+
+	nd = lkl_register_netdev_fd(fd_rx, fd_tx);
+	if (!nd) {
+		perror("failed to register to.");
+		close(fd_rx);
+		close(fd_tx);
+		return NULL;
+	}
+
+	/*
+	 * To avoid mismatch with LKL otherside,
+	 * we always enabled vnet hdr
+	 */
+	nd->has_vnet_hdr = 1;
+	return nd;
+}

tools/lkl/lib/virtio_net_raw.c

@@ -58,5 +58,5 @@ struct lkl_netdev *lkl_netdev_raw_create(const char *ifname)
 	fd_flags = fcntl(fd, F_GETFD, NULL);
 	fcntl(fd, F_SETFL, fd_flags | O_NONBLOCK);
 
-	return lkl_register_netdev_fd(fd);
+	return lkl_register_netdev_fd(fd, fd);
 }

tools/lkl/lib/virtio_net_tap.c

@@ -69,7 +69,7 @@ struct lkl_netdev *lkl_netdev_tap_init(const char *path, int offload,
 		close(fd);
 		return NULL;
 	}
-	nd = lkl_register_netdev_fd(fd);
+	nd = lkl_register_netdev_fd(fd, fd);
 	if (!nd) {
 		perror("failed to register to.");
 		close(fd);

tools/lkl/tests/hijack-test.sh

@@ -72,6 +72,71 @@ elif [ "${CROSS_COMPILE}" = "aarch64-linux-android-" ] ; then
 (echo "$ans" | grep "100756k") || true
 fi
 
+echo "== PIPE tests =="
+if [ -z `which mkfifo` ]; then
+    echo "WARNIG: no mkfifo command, skipping PIPE tests."
+else
+
+fifo1=${work_dir}/fifo1
+fifo2=${work_dir}/fifo2
+mkfifo ${fifo1}
+mkfifo ${fifo2}
+
+# Make sure our device has the addresses we expect
+addr=$(LKL_HIJACK_NET_IFTYPE=pipe \
+	LKL_HIJACK_NET_IFPARAMS="${fifo1}|${fifo2}" \
+	LKL_HIJACK_NET_IP=192.168.13.2 \
+	LKL_HIJACK_NET_NETMASK_LEN=24 \
+	LKL_HIJACK_NET_MAC="aa:bb:cc:dd:ee:ff" \
+	${hijack_script} ip addr)
+echo "$addr" | grep eth0
+echo "$addr" | grep 192.168.13.2
+echo "$addr" | grep "aa:bb:cc:dd:ee:ff"
+
+# Copy ping so we're allowed to run it under LKL
+cp $(which ping) .
+cp $(which ping6) .
+
+# Ping receiver
+LKL_HIJACK_NET_IFTYPE=pipe \
+	LKL_HIJACK_NET_IFPARAMS="${fifo1}|${fifo2}" \
+	LKL_HIJACK_NET_IP=192.168.13.1 \
+	LKL_HIJACK_NET_NETMASK_LEN=24 \
+	LKL_HIJACK_NET_GATEWAY=192.168.13.2 \
+	LKL_HIJACK_NET_IPV6=fc03::1   \
+	LKL_HIJACK_NET_NETMASK6_LEN=64\
+	LKL_HIJACK_NET_GATEWAY6=fc03::2 \
+	${hijack_script} sleep 10 &
+
+sleep 5
+
+# Ping under LKL
+sudo LKL_HIJACK_NET_IFTYPE=pipe \
+	LKL_HIJACK_NET_IFPARAMS="${fifo2}|${fifo1}" \
+	LKL_HIJACK_NET_IP=192.168.13.2 \
+	LKL_HIJACK_NET_NETMASK_LEN=24 \
+	LKL_HIJACK_NET_GATEWAY=192.168.13.1 \
+	LKL_HIJACK_NET_IPV6=fc03::2   \
+	LKL_HIJACK_NET_NETMASK6_LEN=64\
+	LKL_HIJACK_NET_GATEWAY6=fc03::1 \
+	${hijack_script} ./ping 192.168.13.1 -c 1
+
+# Ping 6 under LKL
+sudo LKL_HIJACK_NET_IFTYPE=pipe \
+	LKL_HIJACK_NET_IFPARAMS="${fifo2}|${fifo1}" \
+	LKL_HIJACK_NET_IP=192.168.13.2 \
+	LKL_HIJACK_NET_NETMASK_LEN=24 \
+	LKL_HIJACK_NET_GATEWAY=192.168.13.1 \
+	LKL_HIJACK_NET_IPV6=fc03::2   \
+	LKL_HIJACK_NET_NETMASK6_LEN=64\
+	LKL_HIJACK_NET_GATEWAY6=fc03::1 \
+	${hijack_script} ./ping6 fc03::1 -c 1
+
+wait
+rm ./ping
+rm ./ping6
+fi
+
 echo "== TAP tests =="
 if [ ! -c /dev/net/tun ]; then
     echo "WARNING: missing /dev/net/tun, skipping TAP and VDE tests."

tools/lkl/tests/net-test.c

@@ -144,6 +144,8 @@ static int test_net_init(int argc, char **argv)
 		nd = lkl_netdev_raw_create(ifname);
 	else if (iftype && ifname && (strncmp(iftype, "macvtap", 7) == 0))
 		nd = lkl_netdev_macvtap_create(ifname, 0);
+	else if (iftype && ifname && (strncmp(iftype, "pipe", 4) == 0))
+		nd = lkl_netdev_pipe_create(ifname, 0);
 
 	if (!nd) {
 		fprintf(stderr, "init netdev failed\n");

tools/lkl/tests/net.sh

@@ -1,4 +1,6 @@
-#!/bin/bash -e
+#!/bin/bash
+
+set -e
 
 # currently not supported mingw
 if [ "`printenv CONFIG_AUTO_LKL_POSIX_HOST`" != "y" ] ; then
@@ -17,15 +19,40 @@ GW=`ip route get ${TEST_HOST} |head -n1 | cut -d ' ' -f3`
 script_dir=$(cd $(dirname ${BASH_SOURCE:-$0}); pwd)
 cd ${script_dir}
 
+# Make a temporary directory to run tests in, since we'll be copying
+# things there.
+work_dir=$(mktemp -d)
+
 # And make sure we clean up when we're done
 function clear_work_dir {
+    rm -rf ${work_dir}
     ${SUDO} ip link set dev lkl_ptt1 down &> /dev/null || true
     ${SUDO} ip tuntap del dev lkl_ptt1 mode tap &> /dev/null || true
     ${SUDO} ip link del dev lkl_vtap0 type macvtap &> /dev/null || true
 }
 
 trap clear_work_dir EXIT
 
+echo "== PIPE (LKL net) tests =="
+if [ -z `which mkfifo` ]; then
+    echo "WARNIG: no mkfifo command, skipping PIPE tests."
+else
+
+fifo1=${work_dir}/fifo1
+fifo2=${work_dir}/fifo2
+mkfifo ${fifo1}
+mkfifo ${fifo2}
+hijack_script=${script_dir}/../bin/lkl-hijack.sh
+LKL_HIJACK_NET_IFTYPE=pipe \
+	LKL_HIJACK_NET_IFPARAMS="${fifo1}|${fifo2}" \
+	LKL_HIJACK_NET_IP=192.168.16.1 \
+	LKL_HIJACK_NET_NETMASK_LEN=24 \
+	${hijack_script} sleep 10 &
+sleep 5
+./net-test pipe "${fifo2}|${fifo1}" 192.168.16.1 192.168.16.2 24
+wait
+fi
+
 echo "== TAP (LKL net) tests =="
 if [ -c /dev/net/tun ]; then
     ${SUDO} ip link set dev lkl_ptt1 down || true
@@ -56,7 +83,8 @@ if ! [ -z $DST ]; then
     ${SUDO} ip link set dev ${IFNAME} promisc off
 
     echo "== macvtap (LKL net) tests =="
-    ${SUDO} ip link add link ${IFNAME} name lkl_vtap0 type macvtap mode passthru
+    ${SUDO} ip link add link ${IFNAME} name lkl_vtap0 \
+	    type macvtap mode passthru || true
     if ls /dev/tap* > /dev/null 2>&1 ; then
 	${SUDO} ip link set dev lkl_vtap0 up
 	${SUDO} chown ${USER} `ls /dev/tap*`