diff --git a/.gitmodules b/.gitmodules
new file mode 100644
index 0000000..e66bb90
--- /dev/null
+++ b/.gitmodules
@@ -0,0 +1,3 @@
+[submodule "travis"]
+	path = travis
+	url = ../travis
diff --git a/.travis.yml b/.travis.yml
new file mode 100644
index 0000000..453076f
--- /dev/null
+++ b/.travis.yml
@@ -0,0 +1,117 @@
+# Travis configuration for gardener.
+language: go
+services:
+  - docker
+
+go:
+   - 1.9.x
+
+
+###########################################################################
+before_install:
+# Coverage tools
+- go get github.com/mattn/goveralls
+- go get github.com/wadey/gocovmerge
+
+- echo Branch is ${TRAVIS_BRANCH} and Tag is $TRAVIS_TAG
+
+# Install test credentials.
+# The service account variables are uploaded to travis by running,
+# from root of repo directory:
+#  travis/setup_service_accounts_for_travis.sh
+#
+# All of the gcloud library calls will detect the GOOGLE_APPLICATION_CREDENTIALS
+# environment variable, and use that file for authentication.
+# If the application requires authentication outside the libraries, consider
+# also using travis/activate_service_account.sh
+- if [[ -n "$SERVICE_ACCOUNT_mlab_testing" ]] ; then
+  echo "$SERVICE_ACCOUNT_mlab_testing" > $TRAVIS_BUILD_DIR/creds.json ;
+  export GOOGLE_APPLICATION_CREDENTIALS=$TRAVIS_BUILD_DIR/creds.json ;
+  fi
+
+# These directories will be cached on successful "script" builds, and restored,
+# if available, to save time on future builds.
+cache:
+  directories:
+    - "$HOME/google-cloud-sdk/"
+
+install:
+# Install dependencies
+- GO_IMPORTS=$(go list -f '{{join .Imports "\n"}}{{"\n"}}{{join .TestImports "\n"}}' ./... | sort | uniq | grep -v etl-gardener)
+- go get -u -v -d $GO_IMPORTS
+
+script:
+# To start, run all the non-integration tests.
+- MODULES="inetdiag"
+- for module in $MODULES; do
+    COVER_PKGS=${COVER_PKGS}./$module/..., ;
+  done
+- COVER_PKGS=${COVER_PKGS::-1}  # Trim the trailing comma
+- EC=0
+# Note that for modules in subdirectories, this replaces separating slashes with _.
+- for module in $MODULES; do
+    go test -v -coverpkg=$COVER_PKGS -coverprofile=${module//\//_}.cov github.com/m-lab/tcp-info/$module ;
+    EC=$[ $EC || $? ] ;
+  done
+- echo "summary status $EC" ;
+- if [[ $EC != 0 ]]; then false; fi
+
+
+# Rerun modules with integration tests.  This means that some tests are repeated, but otherwise
+# we lose some coverage.  The corresponding cov files are overwritten, but that is OK since
+# the non-integration tests are repeated.  If we change the unit tests to NOT run when integration
+# test tag is set, then we would need to have separate cov files.
+# Note: we do not run integration tests from forked repos b/c the SA is unavailable.
+# Note that for modules in subdirectories, this replaces separating slashes with _.
+- if [[ -n "$SERVICE_ACCOUNT_mlab_testing" ]] ; then
+  for module in ; do
+    go test -v -coverpkg=$COVER_PKGS -coverprofile=${module//\//_}.cov github.com/m-lab/tcp-info/$module -tags=integration ;
+    EC=$[ $EC || $? ] ;
+  done ;
+  echo "summary status $EC" ;
+  if [[ $EC != 0 ]]; then false; fi ;
+  fi
+
+# Coveralls
+# Run "unit tests" with coverage.
+- $HOME/gopath/bin/gocovmerge *.cov > merge.cov
+- $HOME/gopath/bin/goveralls -coverprofile=merge.cov -service=travis-ci
+
+
+#################################################################################
+# Deployment Section
+#
+#  Overview:
+#   1.  Test in sandbox during development
+#   2.  Deploy to staging on commit to integration
+#   3.  Deploy to prod when a branch is tagged with prod-* or xxx-prod-*
+#
+#  We want to test individual components in sandbox, and avoid stepping on each
+#  other, so we do NOT automate deployment to sandbox.  Each person should
+#  use a branch name to trigger the single deployment that they are working on.
+#
+#  We want to soak all code in staging before deploying to prod.  To avoid
+#  incompatible components, we deploy ALL elements to staging when we merge
+#  to integration branch.
+#
+#  Deployments to prod are done by deliberately tagging a specific commit,
+#  typically in the integration branch, with a tag starting with prod-*.
+#  DO NOT just tag the latest version in integration, as someone may have
+#  pushed new code that hasn't had a chance to soak in staging.
+#
+#
+# Deploy steps never trigger on a new Pull Request. Deploy steps will trigger
+# on specific branch name patterns, after a merge to integration, or on
+# an explicit tag that matches "on:" conditions.
+#################################################################################
+
+
+deploy:
+#########################################
+## Sandbox
+
+#########################################
+## Staging
+
+#########################################
+## Production
diff --git a/README.md b/README.md
index 88796a7..6595d37 100644
--- a/README.md
+++ b/README.md
@@ -1,2 +1,8 @@
 # tcp-info
+| branch | travis-ci | report-card | coveralls |
+|--------|-----------|-----------|-------------|
+| master | [![Travis Build Status](https://travis-ci.org/m-lab/tcp-info.svg?branch=master)](https://travis-ci.org/m-lab/tcp-info) | [![Go Report Card](https://goreportcard.com/badge/github.com/m-lab/tcp-info)](https://goreportcard.com/report/github.com/m-lab/tcp-info) | [![Coverage Status](https://coveralls.io/repos/m-lab/tcp-info/badge.svg?branch=master)](https://coveralls.io/github/m-lab/tcp-info?branch=master) |
+
+
+
 Fast tcp-info collector in Go
diff --git a/inetdiag/inetdiag.go b/inetdiag/inetdiag.go
new file mode 100644
index 0000000..8c4b5ee
--- /dev/null
+++ b/inetdiag/inetdiag.go
@@ -0,0 +1,205 @@
+// Package inetdiag provides basic structs and utilities for INET_DIAG messaages.
+// Based on uapi/linux/inet_diag.h.
+package inetdiag
+
+// Pretty basic code slightly adapted from code copied from
+// https://gist.github.com/gwind/05f5f649d93e6015cf47ffa2b2fd9713
+// Original source no longer available at https://github.com/eleme/netlink/blob/master/inetdiag.go
+
+// Adaptations are Copyright 2018 M-Lab Authors
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+/* IMPORTANT NOTES
+This 2002 article describes Netlink Sockets
+https://pdfs.semanticscholar.org/6efd/e161a2582ba5846e4b8fea5a53bc305a64f3.pdf
+
+"Netlink messages are aligned to 32 bits and, generally speaking, they contain data that is
+expressed in host-byte order"
+*/
+
+import (
+	"fmt"
+	"log"
+	"net"
+	"syscall"
+	"unsafe"
+)
+
+// Constants from linux.
+const (
+	TCPDIAG_GETSOCK     = 18 // uapi/linux/inet_diag.h
+	SOCK_DIAG_BY_FAMILY = 20 // uapi/linux/sock_diag.h
+)
+
+// netinet/tcp.h
+const (
+	_               = iota
+	TCP_ESTABLISHED = iota
+	TCP_SYN_SENT
+	TCP_SYN_RECV
+	TCP_FIN_WAIT1
+	TCP_FIN_WAIT2
+	TCP_TIME_WAIT
+	TCP_CLOSE
+	TCP_CLOSE_WAIT
+	TCP_LAST_ACK
+	TCP_LISTEN
+	TCP_CLOSING
+)
+
+const (
+	TCP_ALL_STATES = 0xFFF
+)
+
+var tcpStatesMap = map[uint8]string{
+	TCP_ESTABLISHED: "established",
+	TCP_SYN_SENT:    "syn_sent",
+	TCP_SYN_RECV:    "syn_recv",
+	TCP_FIN_WAIT1:   "fin_wait1",
+	TCP_FIN_WAIT2:   "fin_wait2",
+	TCP_TIME_WAIT:   "time_wait",
+	TCP_CLOSE:       "close",
+	TCP_CLOSE_WAIT:  "close_wait",
+	TCP_LAST_ACK:    "last_ack",
+	TCP_LISTEN:      "listen",
+	TCP_CLOSING:     "closing",
+}
+
+var diagFamilyMap = map[uint8]string{
+	syscall.AF_INET:  "tcp",
+	syscall.AF_INET6: "tcp6",
+}
+
+// InetDiagSockID is the binary linux representation of a socket, as in linux/inet_diag.h
+// Note that netlink messages use host byte ordering, unless NLA_F_NET_BYTEORDER flag is present.
+type InetDiagSockID struct {
+	IDiagSPort  uint16
+	IDiagDPort  uint16
+	IDiagSrc    [16]byte
+	IDiagDst    [16]byte
+	IDiagIf     uint32
+	IDiagCookie [2]uint32 // This cannot be uint64, because of alignment rules.
+}
+
+// SrcIP returns a golang net encoding of source address.
+func (id *InetDiagSockID) SrcIP() net.IP {
+	return ip(id.IDiagSrc)
+}
+
+// DstIP returns a golang net encoding of destination address.
+func (id *InetDiagSockID) DstIP() net.IP {
+	return ip(id.IDiagDst)
+}
+
+// TODO should use more net.IP code instead of custom code.
+func ip(bytes [16]byte) net.IP {
+	if isIpv6(bytes) {
+		return ipv6(bytes)
+	} else {
+		return ipv4(bytes)
+	}
+}
+
+func isIpv6(original [16]byte) bool {
+	for i := 4; i < 16; i++ {
+		if original[i] != 0 {
+			return true
+		}
+	}
+	return false
+}
+
+func ipv4(original [16]byte) net.IP {
+	return net.IPv4(original[0], original[1], original[2], original[3])
+}
+
+func ipv6(original [16]byte) net.IP {
+	return original[:]
+}
+
+func (id *InetDiagSockID) String() string {
+	return fmt.Sprintf("%s:%d -> %s:%d", id.SrcIP().String(), id.IDiagSPort, id.DstIP().String(), id.IDiagDPort)
+}
+
+// InetDiagReqV2 is the Netlink request struct, as in linux/inet_diag.h
+// Note that netlink messages use host byte ordering, unless NLA_F_NET_BYTEORDER flag is present.
+type InetDiagReqV2 struct {
+	SDiagFamily   uint8
+	SDiagProtocol uint8
+	IDiagExt      uint8
+	Pad           uint8
+	IDiagStates   uint32
+	ID            InetDiagSockID
+}
+
+// SizeofInetDiagReqV2 is the size of the struct.
+// TODO should we just make this explicit in the code?
+const SizeofInetDiagReqV2 = int(unsafe.Sizeof(InetDiagReqV2{})) // Should be 0x38
+
+// Serialize is provided for json serialization?
+// TODO - should use binary functions instead?
+func (req *InetDiagReqV2) Serialize() []byte {
+	return (*(*[SizeofInetDiagReqV2]byte)(unsafe.Pointer(req)))[:]
+}
+
+// Len is provided for json serialization?
+func (req *InetDiagReqV2) Len() int {
+	return SizeofInetDiagReqV2
+}
+
+// NewInetDiagReqV2 creates a new request.
+func NewInetDiagReqV2(family, protocol uint8, states uint32) *InetDiagReqV2 {
+	return &InetDiagReqV2{
+		SDiagFamily:   family,
+		SDiagProtocol: protocol,
+		IDiagStates:   states,
+	}
+}
+
+// InetDiagMsg is the linux binary representation of a InetDiag message header, as in linus/inet_diag.h
+// Note that netlink messages use host byte ordering, unless NLA_F_NET_BYTEORDER flag is present.
+type InetDiagMsg struct {
+	IDiagFamily  uint8
+	IDiagState   uint8
+	IDiagTimer   uint8
+	IDiagRetrans uint8
+	ID           InetDiagSockID
+	IDiagExpires uint32
+	IDiagRqueue  uint32
+	IDiagWqueue  uint32
+	IDiagUID     uint32
+	IDiagInode   uint32
+}
+
+func (msg *InetDiagMsg) String() string {
+	return fmt.Sprintf("%s, %s, %s", diagFamilyMap[msg.IDiagFamily], tcpStatesMap[msg.IDiagState], msg.ID.String())
+}
+
+// rtaAlignOf round the length of a netlink route attribute up to align it
+// properly.
+func rtaAlignOf(attrlen int) int {
+	return (attrlen + syscall.RTA_ALIGNTO - 1) & ^(syscall.RTA_ALIGNTO - 1)
+}
+
+// ParseInetDiagMsg returns the InetDiagMsg itself, and the aligned byte array containing the message content.
+// Modified from original to also return attribute data array.
+func ParseInetDiagMsg(data []byte) (*InetDiagMsg, []byte) {
+	align := rtaAlignOf(int(unsafe.Sizeof(InetDiagMsg{})))
+	if len(data) < align {
+		log.Println("Wrong length", len(data), "<", align)
+		log.Println(data)
+		return nil, nil
+	}
+	return (*InetDiagMsg)(unsafe.Pointer(&data[0])), data[rtaAlignOf(int(unsafe.Sizeof(InetDiagMsg{}))):]
+}
diff --git a/inetdiag/inetdiag_test.go b/inetdiag/inetdiag_test.go
new file mode 100644
index 0000000..61a1998
--- /dev/null
+++ b/inetdiag/inetdiag_test.go
@@ -0,0 +1,107 @@
+package inetdiag_test
+
+import (
+	"log"
+	"syscall"
+	"testing"
+	"unsafe"
+
+	"github.com/m-lab/tcp-info/inetdiag"
+)
+
+// This is not exhaustive, but covers the basics.  Integration tests will expose any more subtle
+// problems.
+
+func TestSizes(t *testing.T) {
+	if unsafe.Sizeof(inetdiag.InetDiagSockID{}) != 48 {
+		t.Error("SockID wrong size", unsafe.Sizeof(inetdiag.InetDiagSockID{}))
+	}
+
+	hdr := inetdiag.InetDiagMsg{}
+	if unsafe.Sizeof(hdr) != 4*6+unsafe.Sizeof(inetdiag.InetDiagSockID{}) {
+		t.Error("Header is wrong size", unsafe.Sizeof(hdr))
+	}
+}
+
+func TestParseInetDiagMsg(t *testing.T) {
+	var data [100]byte
+	for i := range data {
+		data[i] = byte(i + 2)
+	}
+	hdr, value := inetdiag.ParseInetDiagMsg(data[:])
+	if hdr.IDiagFamily != syscall.AF_INET {
+		t.Errorf("Failed %+v\n", hdr)
+	}
+	if hdr.IDiagState != inetdiag.TCP_SYN_RECV {
+		t.Errorf("Failed %+v\n", hdr)
+	}
+
+	if len(value) != 28 {
+		t.Error("Len", len(value))
+	}
+}
+
+func TestSerialize(t *testing.T) {
+	v2 := inetdiag.NewInetDiagReqV2(syscall.AF_INET, 23, 0x0E)
+	data := v2.Serialize()
+	if v2.Len() != len(data) {
+		t.Error("That's odd")
+	}
+}
+
+func TestID4(t *testing.T) {
+	var data [unsafe.Sizeof(inetdiag.InetDiagMsg{})]byte
+	srcIPOffset := unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID) + unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID.IDiagSrc)
+	data[srcIPOffset] = 127
+	data[srcIPOffset+1] = 0
+	data[srcIPOffset+2] = 0
+	data[srcIPOffset+3] = 1
+
+	srcPortOffset := unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID) + unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID.IDiagSPort)
+	// netlink uses host byte ordering, which may or may not be network byte ordering.  So no swapping should be
+	// done.
+	*(*uint16)(unsafe.Pointer(&data[srcPortOffset])) = 0x1234
+
+	dstIPOffset := unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID) + unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID.IDiagDst)
+	data[dstIPOffset] = 1
+	data[dstIPOffset+1] = 0
+	data[dstIPOffset+2] = 0
+	data[dstIPOffset+3] = 127 // Looks like localhost, but its reversed.
+
+	hdr, _ := inetdiag.ParseInetDiagMsg(data[:])
+	if !hdr.ID.SrcIP().IsLoopback() {
+		log.Println(hdr.ID.SrcIP().IsLoopback())
+	}
+	if hdr.ID.IDiagSPort != 0x1234 {
+		t.Errorf("SPort should be 0x1234 %+v\n", hdr.ID)
+	}
+
+	if !hdr.ID.SrcIP().IsLoopback() {
+		t.Errorf("Should be identified as loopback")
+	}
+	if hdr.ID.DstIP().IsLoopback() {
+		t.Errorf("Should not be identified as loopback") // Yeah I know this is not self-consistent. :P
+	}
+}
+
+func TestID6(t *testing.T) {
+	var data [unsafe.Sizeof(inetdiag.InetDiagMsg{})]byte
+	srcIPOffset := unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID) + unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID.IDiagSrc)
+	for i := 0; i < 8; i++ {
+		data[srcIPOffset] = byte(0x0A + i)
+	}
+
+	dstIPOffset := unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID) + unsafe.Offsetof(inetdiag.InetDiagMsg{}.ID.IDiagDst)
+	for i := 0; i < 8; i++ {
+		data[dstIPOffset] = byte(i + 1)
+	}
+
+	hdr, _ := inetdiag.ParseInetDiagMsg(data[:])
+
+	if hdr.ID.SrcIP().IsLoopback() {
+		t.Errorf("Should not be identified as loopback")
+	}
+	if hdr.ID.DstIP().IsLoopback() {
+		t.Errorf("Should not be identified as loopback")
+	}
+}
diff --git a/travis b/travis
new file mode 160000
index 0000000..5f139a8
--- /dev/null
+++ b/travis
@@ -0,0 +1 @@
+Subproject commit 5f139a87a39e67c3e18a13965169de38f3134c51