From fb9b4ddec06c2552564abd8154889910abea2181 Mon Sep 17 00:00:00 2001
From: Martynas Pumputis <martynasp@gmail.com>
Date: Tue, 7 Jun 2016 14:14:53 +0100
Subject: [PATCH] Enable AWSVPC

This requires the following additional steps:

- Setting proxy_arp on on the bridge, because that's how we do
  the routing (check the comment at the place where we set it on).
- Disable ARP requests queueing, because it introduces unnecessary
  delay for the first ARP request resulting in a high latency of
  the first packet (on t2.micro machines I've encountered ~700ms
  latency). Also, the queue is bounded, so we might risking to drop
  the requests.
---
 weave | 44 ++++++++++++++++++++++++++++++++++++--------
 1 file changed, 36 insertions(+), 8 deletions(-)

diff --git a/weave b/weave
index 53af6c0d..1aab3d26 100755
--- a/weave
+++ b/weave
@@ -181,6 +181,7 @@ exec_remote() {
         -e PROXY_HOST="$PROXY_HOST" \
         -e COVERAGE \
         -e CHECKPOINT_DISABLE \
+        -e AWSVPC   \
         $WEAVEEXEC_DOCKER_ARGS $EXEC_IMAGE --local "$@"
 }
 
@@ -532,9 +533,10 @@ random_mac() {
 
 util_op() {
     if command_exists weaveutil ; then
-        weaveutil "$@"
+        WEAVE_HTTP_ADDR=$WEAVE_HTTP_ADDR weaveutil "$@"
     else
         docker run --rm --privileged --net=host --pid=host $(docker_sock_options) \
+            -e WEAVE_HTTP_ADDR \
             --entrypoint=/usr/bin/weaveutil $EXEC_IMAGE "$@"
     fi
 }
@@ -645,7 +647,7 @@ EOF
         fi
     fi
 
-    [ "$1" = "--without-ethtool" ] || ethtool_tx_off_$BRIDGE_TYPE $BRIDGE
+    [ "$1" = "--without-ethtool" -o -n "$AWSVPC" ] || ethtool_tx_off_$BRIDGE_TYPE $BRIDGE
 
     ip link set dev $BRIDGE up
 
@@ -660,6 +662,12 @@ expose_ip() {
         if ! ip addr show dev $BRIDGE | grep -qF $CIDR ; then
             ip addr add dev $BRIDGE $CIDR
             arp_update $BRIDGE $CIDR || true
+            # Remove a default route installed by the kernel, because awsvpc
+            # has installed it as well
+            if [ -n "$AWSVPC" ]; then
+                RCIDR=$(ip route list exact $CIDR proto kernel | head -n1 | cut -d' ' -f1)
+                [ -n "$RCIDR" ] && ip route del dev $BRIDGE proto kernel $RCIDR
+            fi
         fi
         [ -z "$FQDN" ] || when_weave_running put_dns_fqdn_no_check_alive weave:expose $FQDN $CIDR
     done
@@ -1501,22 +1509,19 @@ launch_router() {
     LAUNCHING_ROUTER=1
     check_forwarding_rules
     enforce_docker_bridge_addr_assign_type
-    create_bridge
-    docker_bridge_ip
-    # We set the router name to the bridge MAC, which in turn is
-    # derived from the system UUID (if available), and thus stable
-    # across reboots.
-    PEERNAME=$(cat /sys/class/net/$BRIDGE/address)
+
     # backward compatibility...
     if is_cidr "$1" ; then
         echo "WARNING: $1 parameter ignored; 'weave launch' no longer takes a CIDR as the first parameter" >&2
         shift 1
     fi
+
     CONTAINER_PORT=$PORT
     ARGS=
     IPRANGE=
     IPRANGE_SPECIFIED=
 
+    docker_bridge_ip
     DNS_ROUTER_OPTS="--dns-listen-address $DOCKER_BRIDGE_IP:53"
     NO_DNS_OPT=
 
@@ -1558,6 +1563,10 @@ launch_router() {
             --no-restart)
                 RESTART_POLICY=
                 ;;
+            --awsvpc)
+                AWSVPC_ARGS="--awsvpc"
+                AWSVPC=1
+                ;;
             *)
                 ARGS="$ARGS '$(echo "$1" | sed "s|'|'\"'\"'|g")'"
                 ;;
@@ -1565,6 +1574,13 @@ launch_router() {
         shift
     done
     eval "set -- $ARGS"
+
+    create_bridge
+    # We set the router name to the bridge MAC, which in turn is
+    # derived from the system UUID (if available), and thus stable
+    # across reboots.
+    PEERNAME=$(cat /sys/class/net/$BRIDGE/address)
+
     if [ -z "$IPRANGE_SPECIFIED" ] ; then
         IPRANGE="10.32.0.0/12"
         if ! check_overlap $IPRANGE ; then
@@ -1601,11 +1617,23 @@ launch_router() {
         --ipalloc-range "$IPRANGE" \
         --dns-effective-listen-address $DOCKER_BRIDGE_IP \
         $DNS_ROUTER_OPTS $NO_DNS_OPT \
+        $AWSVPC_ARGS \
         --http-addr $HTTP_ADDR \
         "$@")
     setup_router_iface_$BRIDGE_TYPE
     wait_for_status $CONTAINER_NAME http_call $HTTP_ADDR
     populate_router
+    if [ -n "$AWSVPC" ]; then
+        expose_ip
+        # Set proxy_arp on the bridge, so that it could accept packets destined
+        # to containers within the same subnet but running on remote hosts.
+        # Without it, exact routes on each container are required.
+        echo 1 >/proc/sys/net/ipv4/conf/$BRIDGE/proxy_arp
+        # Avoid delaying the first ARP request. Also, setting it to 0 avoids
+        # placing the request into a bounded queue as it can be seen:
+        # https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/tree/net/ipv4/arp.c?id=refs/tags/v4.6.1#n819
+        echo 0 >/proc/sys/net/ipv4/neigh/$BRIDGE/proxy_delay
+    fi
 }
 
 # Recreate the parameter values that are set when the router is first launched