Add release v2.0.1

Merge pull request #3045 from weaveworks/fix-v2-plugin-release
net-plugin release: swap publish and plugin_publish
2017-06-29 10:58:27 +01:00 · 2017-06-28 17:32:27 +01:00 · 2017-06-28 17:30:03 +01:00 · 2017-06-28 17:25:45 +01:00 · 2017-06-28 17:16:53 +01:00 · 2017-06-28 13:34:49 +00:00
442 changed files with 27297 additions and 18068 deletions
@@ -0,0 +1,2 @@
+--ignore-dir=vendor
+--ignore-dir=.pkg
@@ -0,0 +1,65 @@
+<!--
+Hi, thank you for opening an issue!
+Before hitting the button...
+
+** Is this a REQUEST FOR HELP? **
+If so, please have a look at:
+- our FAQ: https://www.weave.works/docs/net/latest/faq/
+- our troubleshooting page: https://www.weave.works/docs/net/latest/troubleshooting/
+- our help page, to choose the best channel (Slack, etc.) to reach out: https://www.weave.works/help/
+
+** Is this a FEATURE REQUEST? **
+If so, please search existing feature requests, and if you find a similar one, up-vote it and/or add your comments to it instead.
+If you did not find a similar one, please describe in details:
+- why: your use-case, specific constraints you may have, etc.
+- what: the feature/behaviour/change you would like to see in Weave Net
+Do not hesitate, when appropriate, to share the exact commands or API you would like, and/or to share a diagram (e.g.: asciiflow.com): "a picture is worth a thousand words".
+
+** Is this a BUG REPORT? **
+Please fill in as much of the template below as you can.
+
+Thank you!
+-->
+
+## What you expected to happen?
+
+## What happened?
+<!-- Error message, actual behaviour, etc. -->
+
+## How to reproduce it?
+<!-- Specific steps, as minimally and precisely as possible. -->
+
+## Anything else we need to know?
+<!-- Cloud provider? Hardware? How did you configure your cluster? Kubernetes YAML, KOPS, etc. -->
+
+## Versions:
+<!-- Please paste in the output of these commands; 'kubectl' only if using Kubernetes -->
+```
+$ weave version
+$ docker version
+$ uname -a
+$ kubectl version
+```
+
+## Logs:
+```
+$ docker logs weave
+```
+or, if using Kubernetes:
+```
+$ kubectl logs -n kube-system <weave-net-pod> weave
+```
+<!-- (If output is long, please consider a Gist.) -->
+<!-- Anything interesting or unusual output by the below, potentially relevant, commands?
+$ journalctl -u docker.service --no-pager
+$ journalctl -u kubelet --no-pager
+$ kubectl get events
+-->
+
+## Network:
+<!-- If your problem has anything to do with one network endpoint not being able to contact another, please run the following commands -->
+```
+$ ip route
+$ ip -4 -o addr
+$ sudo iptables-save
+```
@@ -25,6 +25,7 @@ _testmain.go
 *.prof
 profile.cov
 coverage.html
+.pkg

 # Emacs backup files
 *~
@@ -36,29 +37,38 @@ coverage.html
 prog/weaver/weaver
 prog/weavedns/weavedns
 prog/weavehosts/weavehosts
-prog/weaveproxy/weaveproxy
 prog/sigproxy/sigproxy
 prog/weavewait/weavewait
 prog/weavewait/weavewait_noop
-prog/netcheck/netcheck
-prog/docker_tls_args/docker_tls_args
+prog/weavewait/weavewait_nomcast
+prog/weaveutil/weaveutil
+prog/*/Dockerfile.weaveworks
+prog/kube-peers/kube-peers
+prog/weave-kube/kube-peers
+prog/weave-npc/weave-npc
+prog/plugin/plugin
 testing/cover/cover
 testing/runner/runner
 releases
-.weaver.uptodate
-.weavedns.uptodate
-.weaveexec.uptodate
-weave.tar
-prog/weaveexec/weave
+.*.uptodate
+weave*.tar.gz
+prog/weaver/weave
 prog/weaveexec/weavehosts
-prog/weaveexec/weaveproxy
 prog/weaveexec/weavewait
 prog/weaveexec/weavewait_noop
+prog/weaveexec/weavewait_nomcast
 prog/weaveexec/sigproxy
-prog/weaveexec/netcheck
-prog/weaveexec/docker_tls_args
-prog/weaveexec/docker*.tgz
+prog/weaver/weaveutil
+prog/weaveexec/docker*
 Vagrantfile.local
+test/tls/tls
 test/tls/*.pem
 test/coverage
 test/coverage.*
+*qemu-*-static
+terraform.tfstate
+terraform.tfstate.backup
+prog/net-plugin/rootfs/
+
+# Website
+site-build
@@ -1,3 +1,135 @@
 [submodule "tools"]
 	path = tools
 	url = https://github.com/weaveworks/build-tools
+[submodule "vendor/github.com/vishvananda/netlink"]
+	path = vendor/github.com/vishvananda/netlink
+	url = https://github.com/vishvananda/netlink.git
+[submodule "vendor/github.com/gorilla/mux"]
+	path = vendor/github.com/gorilla/mux
+	url = https://github.com/gorilla/mux.git
+[submodule "vendor/github.com/gorilla/context"]
+	path = vendor/github.com/gorilla/context
+	url = https://github.com/gorilla/context.git
+[submodule "vendor/github.com/google/gopacket"]
+	path = vendor/github.com/google/gopacket
+	url = https://github.com/google/gopacket.git
+[submodule "vendor/github.com/weaveworks/go-odp"]
+	path = vendor/github.com/weaveworks/go-odp
+	url = https://github.com/weaveworks/go-odp.git
+[submodule "vendor/github.com/opencontainers/runc"]
+	path = vendor/github.com/opencontainers/runc
+	url = https://github.com/opencontainers/runc.git
+[submodule "vendor/github.com/fsouza/go-dockerclient"]
+	path = vendor/github.com/fsouza/go-dockerclient
+	url = https://github.com/fsouza/go-dockerclient.git
+[submodule "vendor/github.com/docker/libnetwork"]
+	path = vendor/github.com/docker/libnetwork
+	url = https://github.com/docker/libnetwork.git
+[submodule "vendor/github.com/docker/docker"]
+	path = vendor/github.com/docker/docker
+	url = https://github.com/docker/docker.git
+[submodule "vendor/github.com/Sirupsen/logrus"]
+	path = vendor/github.com/Sirupsen/logrus
+	url = https://github.com/Sirupsen/logrus.git
+[submodule "vendor/github.com/andybalholm/go-bit"]
+	path = vendor/github.com/andybalholm/go-bit
+	url = https://github.com/andybalholm/go-bit.git
+[submodule "vendor/github.com/miekg/dns"]
+	path = vendor/github.com/miekg/dns
+	url = https://github.com/miekg/dns.git
+[submodule "vendor/github.com/stretchr/testify"]
+	path = vendor/github.com/stretchr/testify
+	url = https://github.com/stretchr/testify.git
+[submodule "vendor/golang.org/x/crypto"]
+	path = vendor/golang.org/x/crypto
+	url = https://github.com/golang/crypto.git
+[submodule "vendor/github.com/docker/machine"]
+	path = vendor/github.com/docker/machine
+	url = https://github.com/docker/machine.git
+[submodule "vendor/github.com/weaveworks/mesh"]
+	path = vendor/github.com/weaveworks/mesh
+	url = https://github.com/weaveworks/mesh
+[submodule "vendor/github.com/vishvananda/netns"]
+	path = vendor/github.com/vishvananda/netns
+	url = https://github.com/vishvananda/netns.git
+[submodule "vendor/github.com/boltdb/bolt"]
+	path = vendor/github.com/boltdb/bolt
+	url = https://github.com/boltdb/bolt
+[submodule "vendor/github.com/weaveworks/go-checkpoint"]
+	path = vendor/github.com/weaveworks/go-checkpoint
+	url = https://github.com/weaveworks/go-checkpoint
+[submodule "vendor/github.com/hashicorp/go-cleanhttp"]
+	path = vendor/github.com/hashicorp/go-cleanhttp
+	url = http://github.com/hashicorp/go-cleanhttp
+[submodule "vendor/github.com/coreos/go-iptables"]
+	path = vendor/github.com/coreos/go-iptables
+	url = https://github.com/coreos/go-iptables
+[submodule "vendor/github.com/pkg/profile"]
+	path = vendor/github.com/pkg/profile
+	url = https://github.com/pkg/profile
+[submodule "vendor/github.com/j-keck/arping"]
+	path = vendor/github.com/j-keck/arping
+	url = https://github.com/j-keck/arping
+[submodule "vendor/github.com/aws/aws-sdk-go"]
+	path = vendor/github.com/aws/aws-sdk-go
+	url = https://github.com/aws/aws-sdk-go
+[submodule "vendor/golang.org/x/sys"]
+	path = vendor/golang.org/x/sys
+	url = https://go.googlesource.com/sys
+[submodule "vendor/github.com/beorn7/perks"]
+	path = vendor/github.com/beorn7/perks
+	url = https://github.com/beorn7/perks
+[submodule "vendor/github.com/prometheus/procfs"]
+	path = vendor/github.com/prometheus/procfs
+	url = https://github.com/prometheus/procfs
+[submodule "vendor/github.com/prometheus/client_golang"]
+	path = vendor/github.com/prometheus/client_golang
+	url = https://github.com/prometheus/client_golang
+[submodule "vendor/github.com/prometheus/client_model"]
+	path = vendor/github.com/prometheus/client_model
+	url = https://github.com/prometheus/client_model
+[submodule "vendor/github.com/prometheus/common"]
+	path = vendor/github.com/prometheus/common
+	url = https://github.com/prometheus/common
+[submodule "vendor/github.com/golang/protobuf"]
+	path = vendor/github.com/golang/protobuf
+	url = https://github.com/golang/protobuf
+[submodule "vendor/github.com/matttproud/golang_protobuf_extensions"]
+	path = vendor/github.com/matttproud/golang_protobuf_extensions
+	url = https://github.com/matttproud/golang_protobuf_extensions
+[submodule "vendor/k8s.io/client-go"]
+	path = vendor/k8s.io/client-go
+	url = https://github.com/kubernetes/client-go
+[submodule "vendor/github.com/pkg/errors"]
+	path = vendor/github.com/pkg/errors
+	url = https://github.com/pkg/errors
+[submodule "vendor/github.com/spf13/cobra"]
+	path = vendor/github.com/spf13/cobra
+	url = https://github.com/spf13/cobra
+[submodule "vendor/github.com/spf13/plag"]
+	path = vendor/github.com/spf13/plag
+	url = https://github.com/spf13/pflag
+[submodule "vendor/github.com/spf13/pflag"]
+	path = vendor/github.com/spf13/pflag
+	url = https://github.com/spf13/pflag
+[submodule "vendor/github.com/mgutz/ansi"]
+	path = vendor/github.com/mgutz/ansi
+	url = https://github.com/mgutz/ansi
+[submodule "vendor/github.com/estesp/manifest-tool"]
+	path = vendor/github.com/estesp/manifest-tool
+	url = https://github.com/estesp/manifest-tool
+[submodule "vendor/github.com/weaveworks/common"]
+	path = vendor/github.com/weaveworks/common
+	url = https://github.com/weaveworks/common
+[submodule "vendor/github.com/containernetworking/cni"]
+	path = vendor/github.com/containernetworking/cni
+	url = https://github.com/containernetworking/cni
+[submodule "vendor/google.golang.org/grpc"]
+	path = vendor/google.golang.org/grpc
+	url = https://github.com/grpc/grpc-go
+[submodule "vendor/github.com/docker/go-units"]
+	path = vendor/github.com/docker/go-units
+	url = https://github.com/docker/go-units
+[submodule "vendor/golang.org/x/net"]
+	path = vendor/golang.org/x/net
+	url = https://github.com/golang/net
@@ -0,0 +1,27 @@
+# IMPORTANT: Please only use GLOBs to ignore files and directories.
+
+# Do not lint the below bash scripts in Weave Net, in order to 
+# avoid breaking the build on make lint:
+bin/circle-deploy-issues
+bin/circle-deploy-master
+bin/circle-teardown-pre
+bin/circle-test-teardown
+bin/circle-test-unit
+bin/install-wordepress
+bin/multiweave
+bin/publish-site
+bin/release
+build/build.sh
+prog/weave-kube/launch.sh
+prog/weaveexec/symlink
+test/*_test.sh
+test/600_proxy_docker_py.sh
+test/assert.sh
+test/config.sh
+test/gce.sh
+test/gen_coverage_reports.sh
+test/run_all.sh
+test/sanity_check.sh
+test/setup.sh
+vendor/*.sh
+weave
@@ -0,0 +1 @@
+2.3.0
@@ -1,28 +0,0 @@
-language: go
-sudo: false
-go:
-  - 1.4
-
-env:
-  - PATH="${PATH}:${HOME}/.local/bin"
-
-addons:
-  apt:
-    packages:
-    - libpcap-dev
-
-install:
-  - go clean -i net
-  - go install -tags netgo std
-  - go get github.com/golang/lint/golint
-  - pip install --user --upgrade gsutil
-  - bin/setup-ci-secrets $encrypted_5ba036e89377_key $encrypted_5ba036e89377_iv
-  - make travis
-  - go get ./...
-
-script:
-  - make tests
-
-after_success:
-  - go get github.com/mattn/goveralls
-  - test "${TRAVIS_PULL_REQUEST}" = "false" && site/build-and-publish.sh
@@ -1,3 +1,617 @@
+## Release 2.0.1
+
+Bug fixes and minor improvements
+
+* Fall back to slower data path (`sleeve`), rather than crashing, when the machine lacks VXLAN support (required for “fast data path”, `fastdp`)  #3043
+* Fix bug in processing of arguments when Docker has TLS enabled, rather than crashing with invalid peers list, e.g. `lookup --tlsverify: no such host` #3039
+* Add `kube-system` namespace back to `weave-kube`'s YAMLs, preventing omissions leading to errors like `error contacting APIServer: the server does not allow access to the requested resource` #3033,#3042
+* Fix release script to prevent ARM64 binaries to end up in AMD64 `net-plugin`, leading to `Error response from daemon: dial unix /run/docker/plugins/<id>/weave.sock: connect: no such file or directory` when installing `net-plugin` #3045
+* `weave reset` and `weave rmpeer` now only contact Weave Cloud when Weave Net is configured with a Weave Cloud token, preventing unnecessary requests and potentially confusing `401 Unauthorized` errors in Weave Net’s logs #3044
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/58?closed=1).
+
+
+## Release 2.0.0
+
+New Features
+
+Peer Discovery via Weave Cloud
+
+You can now get all your Weave Net peers to find each other via the
+Weave Cloud service, instead of maintaining a list of peers at
+startup. #2799,#2827
+
+See the [docs
+page](https://github.com/weaveworks/weave/blob/master/site/using-weave/weave-cloud.md)
+for more details
+
+New Docker Plugin
+
+Docker has a [new plugin
+system](https://docs.docker.com/engine/extend/) which improves the
+installation UX and solves some issues around startup.  This means
+Weave Net 2.0 can now run with Docker in "swarm mode" and supports the
+`docker service` command. #2396,#2397,#2651,#2727,#2805,#2816,#2905,
+#2906,#2929,#2932,#2945,#2950,#2956,#2963,#2964,#2966,#3019
+
+The previous Docker Plugin is still available and can be installed as before.
+
+All of Weave Net now runs in one container
+
+Previously we had three separate containers for routing, Docker API
+proxy and Docker plugin. Running everything in one simplifies start-up
+and removes the need to detect various error
+conditions. #1642,#2897,#2936,#2945,#2946,#2951,#2960
+
+The individual commands ‘weave launch-router’, ‘weave launch-plugin’,
+etc., have been removed. You can turn off the plugin and proxy with
+new command-line options. In keeping with Semantic Versioning, we have
+changed the major version number for this release.
+
+Other new features
+
+* Kubernetes configuration now comes from our “Launch Generator” that
+  allows different options to be selected via
+  URL. #2754,#2903,#3000,#3001
+* `weave-kube` now stores data about IP allocation in `/var/lib/weave`
+  on the host instead of in a Kubernetes volume.  This means that the
+  data will persist across pod deletion and re-creation, e.g. during
+  an upgrade of Weave Net, which makes restarts more
+  reliable. #2610,#2967
+* `weave-kube` turned on rolling updates, so careful manual handling
+  of updates is no longer required. #3024
+
+Bug fixes
+
+* Kubernetes Network Policies which allowed a specific set of pods to
+  connect would block all pods on other hosts. Revert the change in
+  v1.9.6 which ignored pods on other hosts #3025,#3028
+
+Features removed
+
+* `weave run` has been removed.  This was the original method provided
+  to start containers with Weave Net, but it always required care over
+  timing of start-up, and we now provide three alternative, better,
+  ways. You can replicate the effect by calling `docker run` then
+  `weave attach`.  Similarly `weave start` and `weave restart` were
+  removed. #2353,#2885
+* Everything deprecated more than one release ago has been removed, so
+  if you use it now you get an error rather than a warning. This
+  includes the ‘create-bridge’ command and older command-line
+  arguments, e.g. `--iprange` was replaced by `--ipalloc-range`
+  #2901,#2909,#2913,#2942,#2989,#2991
+
+Functions moved from shell-script to Go code.
+
+This enables more precise error-checking and runs a bit faster. It has
+also enabled us to shrink the size of images downloaded: `weave-kube`
+is 101MB compared to 163MB previously #2953,#2954,#2974
+
+Specific items that moved from shell-script to Go:
+* Setting up the `weave` bridge #1958,#2975,#2977,#2978
+* Container attachment #2947
+* Creation of the ’weave’ default plugin network #2920
+
+Minor improvements
+
+* You can now restart the Weave Net router without requiring the proxy to be enabled #2112
+* Plugin (legacy version) now respects `--ipalloc-default-subnet` option #2919
+* The `weave` script now detects and issues an error message if
+  `weave-kube` is running and you attempt to launch again from the script. #2709/#2966
+* It is now possible to choose the the MAC address of the `weave`
+  bridge using `--name`, in case your hosts have identical unique
+  IDs. #2900
+* Relaxed Kubernetes tolerations for Weave Net's daemonset in order to
+  match any node (previously, only taints directed at master). #3018
+* Kubernetes' `seLinuxOptions` configuration is now empty by default,
+  to reduce spurious failures on hosts not using seLinux. #3001
+* Improved reliability of namespace changes via `nsenter`. #2992
+* `weave ps` now fetches the list container IDs internally, rather
+  than calling out to `docker ps` #2814,#2898
+* at startup, actively remove dead containers’ Weave Net IP addresses from IPAM #3013
+* at startup, only check live containers to see if they have an
+  existing Weave Net IP address #2815,#2829
+* Weave Net CNI plugin now logs but does not raise an error if
+  anything goes wrong during network interface delete, to be more
+  compatible with Kubernetes 1.6. #2928
+* Stop running a shell in “privileged” mode when it’s only writing a file #2838
+* New internal REST endpoint to return all IP address mappings. #1350
+* Changed the wording where we do not log the password #2833
+* Fixed typo in plugin error messages #2894
+
+Build and test
+
+* Weave Net is now built with Go version 1.8, which has better code
+  generation and garbage collection #2914
+* During smoke-tests, use a webserver instead of just `ping` so we get
+  a more realistic test that the Weave network is working #2918
+* When installing dependencies for the build container, use a
+  keyserver port that's better for firewalls #2812
+* Kubernetes test script now scales up to more hosts, and works with
+  Kubernetes 1.6 #2837,#2853,#2923
+* Other minor build improvements and refactoring #2760,#2910
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/49?closed=1).
+
+
+## Release 1.9.8
+
+Bug fixes and minor improvements
+
+* Fix weave-npc blocking NodePort and any other non-local access #3011,#3014
+* Fix bug where IPAM would duplicate a fixed IP address assigned via Docker plugin #3003,#3010
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/56?closed=1).
+
+
+## Release 1.9.7
+
+This is identical to 1.9.6 with one additional bug-fix:
+
+* weave-npc would block everything if `kubelet --hostname-override` was used #2995,#2996
+
+## Release 1.9.6
+
+Bug fixes and minor improvements
+
+* Ensure that Kubernetes pods can contact a service implemented within
+  the same pod, by turning on "hairpin mode". This is required because
+  of a quiet change between Kubernetes 1.5 and 1.6. #2993
+* Network Policy Controller (`weave-npc`) now checks local addresses
+  only, so it doesn't interfere with cross-cluster traffic. It should
+  be more efficient too #2622,#2973,#2979
+* Stop reporting back to Kubernetes any issues encountered when
+  deleting a pod's network interface. This is required because
+  of a quiet change between Kubernetes 1.5 and 1.6. #2921,#2928
+* Fixed an issue whereby `weave-npc` couldn't start because one
+  `ipset` was referring to another one and could not be destroyed #2915,#2949
+* Improved the code which checks whether the kernel supports `ipset` #2934,#2935
+* `weave-npc` now creates ipsets with only valid xml characters in the
+  name #2958,#2959
+
+Build and Testing
+
+* In build container use cross-compilers from debian package
+  repository, so they match other components #2940
+* Pin the version of the linting tool `shfmt` so the set of things it
+  checks is stable #2987
+* Fix lint error in script that runs smoke-tests #2962
+* Moved website publishing from Wordpress to Netlify #2986
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/55?closed=1).
+
+## Release 1.9.5
+
+Bug fixes and minor improvements
+
+* Improve log messages generated if "hairpin" conditions are detected,
+  to make clear which kind is likely to cause problems #2808/#2926
+* Filter out IPv6 peer addresses from Kubernetes; Weave Net currently
+  only supports IPv4 #2904/#2912
+* Fix rare crash during initialization of weave-kube #2893/#2892
+* Include overlay and encryption modes in checkpoint reports, in case
+  this is relevant to a version upgrade #2771/#2907
+
+Build and Testing
+
+* Ensure CI build can run gcloud tools  #2887
+* Prevent kubeadm from upgrading Kubernetes if we are trying to test an older version #2886
+* Upgrade build scripts to support Kubernetes 1.6 #2880
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/54?closed=1).
+
+## Release 1.9.4
+
+Bug fixes and minor improvements
+
+* Support Kubernetes 1.6 by creating a new DaemonSet #2777,#2801
+* Support Kubernetes 1.6 by allowing CNI callers to send a
+  network-delete request for a container that is not running or has
+  never been attached to the network #2850
+* Leave non-weave ipsets alone in Network Policy Controller (e.g. when
+  running Weave Net alonside keepalived-vip) #2751,#2846
+* Fix various small issues revealed by 'staticcheck' tool #2843,#2857
+* Avoid leaving 'defunct' processes when weave-kube container restarts #2836,#2845
+* When using the CNI plugin with a non-standard network configuration
+  file, the weave bridge could get the same IP as a container, if
+  'weave expose' hadn't run at that point #2839,#2856
+
+Build and Testing
+
+* Check that no defunct processes remain after each test #2852
+* Update build and test scripts to work with Kubernetes 1.6 beta #2851
+
+[Full list of changes](https://github.com/weaveworks/weave/milestone/53?closed=1).
+
+## Release 1.9.3
+
+Bug fixes and minor improvements
+
+* Fixed a race condition in Fast Datapath encrypted connections which could lead Weave Net to crash: #2824, #2825.
+
+## Release 1.9.2
+
+Bug fixes and minor improvements
+
+* Fix a weave-kube bug when `br_netfilter` or `xt_set` module is compiled into
+  kernel #2820/#2821
+* Detect the absence of the required `xt_set` kernel module #2821
+
+## Release 1.9.1
+
+Bug fixes and minor improvements
+
+* Fix a race condition when the Weave Net container is restarted
+  which could allow a new container to be allocated the same IP
+  address as an existing one #2784,#2787
+* Handle the message type received when a pod has been deleted during
+  Kubernetes api-server fail-over #2772,#2773
+* Make weave-kube work with `dockerd --iptables=false` #2726
+* Ensure we have the right kernel modules loaded for Network Policy in weave-kube #2819
+* Reference-count addresses in Network Policy Controller, to avoid
+  errors when updates come in an unexpected order #2792,#2795
+* Allow the soft connection limit to be raised in weave-kube, so
+  larger clusters can be created #2781
+* WeaveDNS was incorrectly case-sensitive for reverse DNS lookups #2817,#2818
+
+Build and Testing
+
+* Scripts to create VMs to run automated tests were rewritten to use
+  Terraform and Ansible, to make it much easier to test with different
+  versions of components such as Docker and Kubernetes #2647,#2694,#2775,#2796
+* Upgrade to latest Weaveworks common build-tools #2780
+* Improve encryption tests #2793
+* Update vishvananda/netlink library to bring in changes we had previously forked #2790
+* Slight change to the build container to avoid permission errors and slow builds #2761,#2802
+
+## Release 1.9.0
+
+Highlights:
+
+* Encryption is now available for Fast Datapath connections, which
+  greatly improves the performance. #1644,#2687
+* We now build images for Intel/AMD 64-bit, ARM and ARM 64-bit. #2713
+* Weave Net Docker images are now labelled with description, vendor, etc. #2712
+* `weave status connections` now shows the MTU #2389,#2663
+* CNI plugin is now a stand-alone binary that does not depend on Docker #2594,#2662
+* Embedded docker client updated to version 1.10.3 #2395
+
+**NOTE:** The move to multi-architecture required that we update the
+  embedded Docker client, and this has the effect that this release of
+  Weave Net will not work with Docker installations older than
+  1.10.
+
+[Full list of changes](https://github.com/weaveworks/weave/issues?q=milestone%3A1.9.0).
+
+## Release 1.8.2
+
+Bug fixes and minor improvements
+
+* Fixed a bug where looping flows were installed which caused high CPU
+  usage #2650, #2674
+* Fixed a bug where Kubernetes master could not contact pods #2673, #2683
+* Fixed a bug where weave-kube was crashing in a loop due to invalid
+  Weave bridge state #2657
+* Fixed a bug where iptables NAT rules were not appended due to
+  "temporary unavailable" iptables error #2679
+* Added a detection of enabling the hairpin mode on the Weave bridge port
+  which caused installation of looping flows #2674
+* Added a detection of overlaps between Weave and the host IP address
+  ranges when launching weave-kube #2669, #2672
+* Added logging of connections blocked by weave-npc #2546, #2573
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.8.2).
+
+## Release 1.8.1
+
+Bug fixes and minor improvements
+
+* Fixed weave-npc crash from Succeeded/Failed pods #2632,#2658
+* Fixed occasional failure to create Weave bridge on node reboot
+  #2617,#2637
+* Fixed a bug where weave-kube would fail to install when run with
+  unrelease snapshot builds #2642
+* Improved conformance to CNI spec by not releasing IP addresses when
+  a container dies #2643
+* Improved troubleshooting of install failure by creating CNI config
+  after Weave Net is up #2570
+* "up to date" shown even when the version check was blocked by
+  firewall #2537,#2565,#2645
+* "Unable to claim" message on re-launching Weave after using CNI
+  #2548,#2577
+* Eliminated spurious IP reclaim operations when IPAM was disabled
+  #2567,#2644
+* Include `jq` tool in our build VM configuration #2656
+
+## Release 1.8.0
+
+Highlights:
+
+* Exposed network policy controller Prometheus metrics
+  weaveworks/weave-npc#23, #2595, #2549
+* Exposed router Prometheus metrics #2535, #2547, #2523, #2579, #2578,
+  #2568, #2560, #2561
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.8.0).
+
+## Release 1.7.2
+
+Bug fixes and minor improvements
+
+* Fixed an error where the Docker plugin could fail to attach a
+  container with a `bridge "weave" not present` error #2540/#2541
+* Fixed `cannot connect to itself` panic on weave launch #2527/#2543
+* Fixed inferred initial peer count when target peers includes self
+  #2481/#2543
+* Fixed compilation on Raspberry Pi #2506/#2538
+
+## Release 1.7.1
+
+Bug fixes and minor improvements
+
+* Added utility to recover container addresses without asking Docker #2531
+* Improved CI #2274
+
+## Release 1.7.0
+
+Highlights:
+
+* weave-kube - Deploy Weave Net to Kubernetes with a single command
+* weave-npc - Kubernetes network policy controller implementation
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.7.0).
+
+## Release 1.6.2
+
+Bug fixes and minor improvements
+
+* Fixed hang after stopping and restarting on Docker 1.12 #2469/#2502
+* Avoid an error on Google container images by checking for tx offload support #2504
+* Fixed an issue where the supplied peer list could be ignored when restarting after failure #2503/#2509
+* Check for empty peer name on launch #2495/#2501
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.6.2).
+
+## Release 1.6.1
+
+Bug fixes and minor improvements
+
+* `weave ps` was occasionally failing to list allocated addresses of containers #2388/#2418/#2445
+* `weave launch[-router]` on 4.2 kernel would appear to succeed even if the fast datapath VXLAN UDP port was in use by a different process #2375/#2474
+* Launching the proxy would fail when the Docker daemon could not be detected #2457/#2424
+* The CNI plugin did not work with Apache Mesos #2394/#2442
+* Router stopped working after a restart in the AWSVPC mode #2381/#2409
+* Router crashed when the Docker API endpoint parameter was explicitly set to empty #2421/#2467
+* The CNI plugin did not work on recent versions of Docker for Mac #2434/#2442
+* The CNI plugin assigns an IP to the bridge if necessary, which avoids failures if `weave expose` has not run yet #2471
+* Distinguish peer name collisions from attempts to connect to self in logs #2460
+* Improve host clock skew detection message #2174
+* Improve the error message returned when executing `weave launch-plugin` without the router running #2293/#2416
+* The `create-bridge` subcommand was not enabled in the fast datapath mode #2464/2466
+* Allow users to omit `weave setup[-cni]` by initializing the CNI plugin on the `launch[-router]` subcommand #2435/#2442
+* Reduce verbosity of fast datapath miss event logs #1852/#2417
+* Include the `ipam` option in the help output of the `status` subcommand #2425/#2426
+* Remove a harmless duplication of the `--no-dns` parameter #2430
+* Internal refactoring
+* Improvements to testing and building
+* Improvements to the documentation
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.6.1).
+
+## Release 1.6.0
+
+## Highlights
+
+* A new [AWS VPC](https://weave.works/docs/net/latest/using-weave/awsvpc/) mode
+  that leverages Amazon Virtual Private Cloud for near-native network
+  performance, as an alternative to the Sleeve and Fast Datapath
+  overlays
+* Docker 1.12 introduced some internal changes that made it
+  incompatible with previous version of Weave Net - this version
+  restores compatibility
+* An [operational guide](https://weave.works/docs/net/latest/operational-guide)
+  detailing best practices for deploying and operating Weave Net
+* Changes to the target peer list are remembered across restarts,
+  making it much easier to deploy resilient networks
+* The version checkpoint now transmits network mode (e.g. 'awsvpc')
+  and kernel/docker versions to us to inform and guide our development
+  efforts. See the [installation documentation](https://weave.works/docs/net/latest/installing-weave/)
+  for instructions on disabling the checkpoint feature.
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.6.0).
+
+## Release 1.5.2
+
+Bug fixes and minor improvements
+
+* Weave Proxy did not flush the initial http header in the Docker event stream, which could cause Docker Swarm to show all nodes pending. #2306/#2311
+* When using the CNI plugin, if a container was removed and quickly replaced by another using the same IP address, other containers might be unable to contact it. Send an address resolution protocol message to update them. #2313
+* Avoid Docker hanging for 1 minute in `weave launch` if the plugin had not shut down cleanly #2286/#2292
+* Print an error message when Weave bridge mode is changed without `weave reset` #2304
+* Eliminate spurious warning message from IP allocator on plugin shutdown #2300/#2319
+* Display error message when address requested in a subnet that is too small (/31 or /32) #2282/#2321
+* Add short wait after `weave reset` to allow updates to reach peers #2280
+* Weave was occasionally unable to claim existing IP address immediately after launch #2275/#2281
+* Refactor some integration tests to run faster and more reliably #2291
+
+## Release 1.5.1
+
+Bug fixes and minor improvements
+
+* Persisted data that was rendered invalid by changing peer name or
+  allocation range is detected and removed automatically, preventing
+  crashes and hangs #2246/#2209/#2249
+* `weave rmpeer` persists the range takeover in case the peer on which
+  it was executed dies subsequently #2238
+* Launching a container with an explicit `WEAVE_CIDR` in the
+  allocation range now waits instead of erroring if the allocator
+  hasn't finished initialising #2232/#2265
+* Weave DNS now responds to AAAA queries with an empty answer section,
+  instead of NXDOMAIN which could be cached and block subsequent
+  resolution of A records #2244/#2252
+* Many improvements to the documentation.
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.5.1).
+
+## Release 1.5.0
+
+## Highlights
+
+- A new [Container Network Interface](https://github.com/appc/cni#cni---the-container-network-interface) plugin.
+- This release is much more robust against unscheduled restarts,
+  because it persists key data to disk.
+- New configuration options that are useful when you create or
+  auto-scale larger networks.
+- Weave now periodically checks for updates (can be disabled)
+
+Plus many bug fixes and minor enhancements. More details below and in
+the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.5.0).
+
+## Release 1.4.6
+
+Fixes some issues encountered by our users.
+
+* Restarting a peer could leave stale entries in WeaveDNS
+  #1867/#2023/#2081
+* Weave proxy occasionally failing to attach any new containers
+  #2016/#2049
+
+Other fixes:
+
+* Resolved a crash when a restarting peer re-connected to a peer that
+  had not received the latest IPAM data #2083/#2092
+* Make IP address space available immediately after a dead peer is
+  removed #2068
+
+## Release 1.4.5
+
+Higher performance for multicast and broadcast traffic when using
+Weave's Fast Datapath
+
+* The flow rule to deliver broadcast and multicast packets in-kernel
+  was not created correctly, hence every such packet caused a
+  context-switch to the software router #2003/#2008
+
+Other fixes:
+
+* Remove DNS entries for containers that are being restarted by Docker
+  but are not live yet #1977/#1989
+* Don't let one failing allocation attempt prevent others from
+  succeeding; they could be in different subnets which are more
+  available #1996/#2001
+* Don't complain on second router launch on kernels that lack support
+  for Fast Datapath #1929/#1983
+* Fix build broken by change in libnetwork IPAM API #1984/#1985
+
+## Release 1.4.4
+
+Fixing a rather serious issue that slipped through our preparations
+for Docker 1.10:
+
+* Restarting Docker or rebooting your machine while the Weave Net
+  plugin is running causes Docker 1.10.0 to fail to start #1959/#1963
+
+Also one other small fix:
+
+* Avoid a hang when trying to use plugin and proxy at the same time
+  #1956/#1961
+
+## Release 1.4.3
+
+Preparing for Docker 1.10, plus some bug-fixes.
+
+* Avoid hang in Docker v1.10 on `docker volume ls` after `weave stop`,
+  `weave stop-plugin` or `weave reset` #1934/#1936
+* Fix "unexpected EOF" from Docker 1.10 on `docker exec` with Weave
+  proxy `--rewrite-inspect` enabled #1911/#1917
+* Avoid losing DNS entries and potentially double-allocating IP
+  Addresses allocated via plugin, on router restart; also extend
+  `weave ps` to show IP addresses allocated via plugin #1745/#1921
+* Stop creating lots of copies of `weavewait` program in Docker
+  volumes #1757/#1935
+* Prevent container starting prematurely when proxy in
+`--no-multicast-route` mode #1942/#1943
+* Log error message from plugin rather than crashing when weave not
+  running #1906/#1918
+* Warn, don't error, if unable to remove plugin network in 'weave
+  stop', to avoid breaking the usual upgrade or config change process
+  #1900/#1919
+* Don't crash if network conditions suggest only very small packets
+  will get through #1905/#1926
+* Cope with unexpected errors during route traversal when starting
+  container via proxy #1909/#1910/#1932
+
+## Release 1.4.2
+
+Bug-fixes and minor improvements.
+
+* A race condition in weavewait that would occasionally hang
+  containers at startup #1882/#1884
+* Having the plugin auto-restart prevents successful `weave launch` on
+  reboot #1869
+* Work round weave router failure on CoreOS 4.3 caused by kernel bug
+  #1854
+* `weave launch` would exit with error code on docker <1.9 #1851
+* Running `eval $(weave env)` multiple times would break `eval $(weave
+  env --restore)` #1824/#1825
+* Don't complain in `weave stop` about "Plugin is not running" when
+  plugin is not enabled #1840/#1841
+* `weave --local launch` would fail if utility program
+  `docker_tls_args` could not be found #1844
+* Improved error reporting when TLS arg detection fails #1843
+* Improve error reporting when docker isn't running #1845
+* Add `--trusted-subnets` usage to `weave` script #1842
+* weave run can hang under rare combinations of options #1858
+
+## Release 1.4.1
+
+This is a bug-fix release to cover a few issues that came up since the
+release of 1.4.0.
+
+* Weave would fail to launch when `$DOCKER_HOST` was set to a TCP socket secured with TLS #1820/#1822
+* Weave would fail to stop when run against Docker pre-1.9 #1815/#1817
+* Issue a warning instead of failing on `--with-dns` option to proxy, which was removed #1810/#1812
+* Make `weave version` show the plugin version #1797/#1813
+* Make `weave launch` show when the a container is restarting #1778/#1814
+* Make `weave launch` fail if the plugin is running, for consistency with router and proxy. #1818/#1819
+
+More details in the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.4.1).
+
+## Release 1.4.0
+
+**Highlights**
+
+* The Docker Network plugin can now operate without a cluster store, so it is now run by default.
+* You can now use the fast datapath over trusted links and Weave encryption over untrusted links, in the same network.
+
+Plus many bug fixes and minor enhancements. More details below and in
+the [change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.4.0).
+
+## Release 1.3.1
+
+**Highlights**
+
+* The minimum Docker version has been increased to 1.6 due to the
+  [upcoming deprecation of Dockerhub access](https://blog.docker.com/2015/10/docker-hub-deprecation-1-5/)
+  for old clients. From December 7th onwards previous versions of the
+  `weave` script will fail to pull down images from the hub; if you
+  are unable to upgrade to 1.3.1 immediately you can work around this
+  by running `weave --local setup` in conjunction with a compatible
+  Docker client installation
+* Docker networking plugin now works with older kernels and allows you
+  to configure the MTU
+
+More details in the
+[change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.3.1).
+
+## Release 1.3.0
+
+**Highlights**
+
+This release includes a Docker Plugin, so you have the option to use
+Weave Net that way. More details in the
+[change log](https://github.com/weaveworks/weave/issues?q=milestone%3A1.3.0).
+
 ## Release 1.2.1

 This release contains a number of bug fixes and minor
@@ -0,0 +1,32 @@
+Thanks for your interest in contributing to this project and for
+taking the time to read this guide.
+
+# Code of conduct
+
+**Be respectful:** disagreements between people are to be expected and are usually the sign of healthy debate and engagement. Disagreements can lead to frustration and even anger for some members. Turning to personal insults, intimidation or threatening behaviour does not improve the situation. Participants should thus take care to ensure all communications / interactions stay professional at all times.
+
+**Be considerate:** remember that the community has members with a diverse background, many of whom have English as a second language. What might appear impolite, may simply be a result of a lack of knowledge of the English language. Bear in mind that actions will have an impact on other community members and the project as a whole, so take potential consequences into account before pursuing a course of action.
+
+**Be forgiving:** humans are fallible and as such prone to make mistakes and inexplicably change their positions at times. Don't assume that other members are acting with malicious intent. Be prepared to forgive people who make mistakes and assist each other in learning from them. Playing a blame game doesn't help anyone.
+
+# Issues
+* Before reporting an issue make sure you search first if anybody has already reported a similar issue and whether or not it has been fixed.
+* Make sure your issue report sufficiently details the problem:
+ * If you set up weave in some way other than that described by our [main instructions](https://github.com/zettio/weave#installation), please mention that.
+ * Include step-by-step instructions to reproduce the issue.
+ * State the OS platform you are using, i.e. `uname -a`
+ * `iptables` rules and firewalls can have a great impact on any Linux network; please note whether you have set up any rules and/or are running a firewall on your machine(s).
+* If you want to tackle any open issue, make sure you let people know you are working on it.
+
+## Pull Requests
+* Please be generous describing your changes.
+* Although it is highly suggested to include tests, they are not a hard requirement in order to get your contributions accepted.
+* Keep pull requests small so core developers can review them quickly.
+* Keep each pull request focused on a specific topic.  If you have two things to change, create two pull requests. This helps reviewers to understand the meat of your contribution.
+
+## Workflow
+See [WorkingOnWeave](https://github.com/weaveworks/weave/wiki/WorkingOnWeave)
+
+# Acknowledgements
+
+Much of this taken from https://github.com/cloudescape/gowsdl/blob/master/CONTRIBUTING.md
@@ -0,0 +1,3 @@
+DOCKER_VERSION=1.13.1
+KUBERNETES_VERSION=1.6.2
+KUBERNETES_CNI_VERSION=0.5.1
@@ -176,7 +176,7 @@

   END OF TERMS AND CONDITIONS

-   Copyright 2014-2015 Weaveworks Ltd.
+   Copyright 2014-2016 Weaveworks Ltd.

   Licensed under the Apache License, Version 2.0 (the "License");
   you may not use this file except in compliance with the License.
@@ -1,40 +1,143 @@
-PUBLISH=publish_weave publish_weaveexec
-
 .DEFAULT: all
-.PHONY: all update tests lint publish $(PUBLISH) clean clean-bin prerequisites build travis run-smoketests
+.PHONY: all exes testrunner update tests lint publish-one-arch $(PUBLISH) clean clean-bin clean-work-dir prerequisites build run-smoketests

 # If you can use docker without being root, you can do "make SUDO="
-SUDO=sudo
+SUDO=$(shell docker info >/dev/null 2>&1 || echo "sudo -E")
+BUILD_IN_CONTAINER=true
+RM=--rm
+RUN_FLAGS=-ti
+COVERAGE=

-DOCKERHUB_USER=weaveworks
-WEAVE_VERSION=git-$(shell git rev-parse --short=12 HEAD)
+# This specifies the architecture we're building for
+ARCH?=amd64

+# We're using QEMU to be able to build Docker images for other architectures
+# from amd64 machines.
+QEMU_VERSION=v2.7.0
+
+# A list of all supported architectures here. Should be named as Go is naming platforms
+# All supported architectures must have an "ifeq" block below that customizes the parameters
+ALL_ARCHITECTURES=amd64 arm arm64
+ML_PLATFORMS=linux/amd64,linux/arm,linux/arm64
+
+ifeq ($(ARCH),amd64)
+# The architecture to use when downloading the docker binary
+	WEAVEEXEC_DOCKER_ARCH?=x86_64
+
+# The name of the alpine baseimage to use as the base for weave images
+	ALPINE_BASEIMAGE?=alpine:3.4
+
+# The extension for the made images
+# Specifying none means for example weaveworks/weave:latest
+	ARCH_EXT?=
+
+# The name of the gcc we're using for compiling C code
+	CC=gcc
+
+# Optional parameters that can be passed when linking C code into the Go binary
+	CGO_LDFLAGS=
+endif
+ifeq ($(ARCH),arm)
+# The architecture to use when downloading the docker binary
+	WEAVEEXEC_DOCKER_ARCH?=armel
+
+# Using the (semi-)official alpine image
+	ALPINE_BASEIMAGE?=armhf/alpine:3.4
+
+# arm images have the -arm suffix, for instance weaveworks/weave-arm:latest
+	ARCH_EXT?=-arm
+
+# The name of the gcc binary
+	CC=arm-linux-gnueabihf-gcc
+
+# The architecture name to use when downloading a prebuilt QEMU binary
+	QEMUARCH=arm
+
+# In the weaveworks/build image; libpcap libraries for arm are placed here
+# Tell the gcc linker to search for libpcap here
+	CGO_LDFLAGS="-L/usr/local/lib/$(CC)"
+endif
+ifeq ($(ARCH),arm64)
+# Use the arm 32-bit docker client variant
+	WEAVEEXEC_DOCKER_ARCH?=armel
+
+# Using the (semi-)official alpine image
+	ALPINE_BASEIMAGE?=aarch64/alpine:3.5
+
+# arm64 images have the -arm64 suffix, for instance weaveworks/weave-arm64:latest
+	ARCH_EXT?=-arm64
+
+# The name of the gcc binary
+	CC=aarch64-linux-gnu-gcc
+
+# The architecture name to use when downloading a prebuilt QEMU binary
+	QEMUARCH=aarch64
+
+# In the weaveworks/build image; libpcap libraries for arm64 are placed here
+# Tell the gcc linker to search for libpcap here
+	CGO_LDFLAGS="-L/usr/local/lib/$(CC)"
+endif
+
+# The name of the user that this Makefile should produce image artifacts for. Can/should be overridden
+DOCKERHUB_USER?=weaveworks
+# The default version that's chosen when pushing the images. Can/should be overridden
+WEAVE_VERSION?=git-$(shell git rev-parse --short=12 HEAD)
+# Docker Store does not allow the "latest" tag.
+NET_PLUGIN_LATEST=latest_release
+
+# Paths to all relevant binaries that should be compiled
 WEAVER_EXE=prog/weaver/weaver
-WEAVEPROXY_EXE=prog/weaveproxy/weaveproxy
 SIGPROXY_EXE=prog/sigproxy/sigproxy
+KUBEPEERS_EXE=prog/kube-peers/kube-peers
+WEAVENPC_EXE=prog/weave-npc/weave-npc
 WEAVEWAIT_EXE=prog/weavewait/weavewait
 WEAVEWAIT_NOOP_EXE=prog/weavewait/weavewait_noop
-NETCHECK_EXE=prog/netcheck/netcheck
-DOCKERTLSARGS_EXE=prog/docker_tls_args/docker_tls_args
-RUNNER_EXE=testing/runner/runner
+WEAVEWAIT_NOMCAST_EXE=prog/weavewait/weavewait_nomcast
+WEAVEUTIL_EXE=prog/weaveutil/weaveutil
+RUNNER_EXE=tools/runner/runner
+MANIFEST_TOOL_DIR=vendor/github.com/estesp/manifest-tool
+MANIFEST_TOOL_EXE=$(MANIFEST_TOOL_DIR)/manifest-tool
+TEST_TLS_EXE=test/tls/tls
+NETWORKTESTER_EXE=test/images/network-tester/webserver

-EXES=$(WEAVER_EXE) $(SIGPROXY_EXE) $(WEAVEPROXY_EXE) $(WEAVEWAIT_EXE) $(WEAVEWAIT_NOOP_EXE) $(NETCHECK_EXE) $(DOCKERTLSARGS_EXE) $(RUNNER_EXE)
+# All binaries together in a list
+EXES=$(WEAVER_EXE) $(SIGPROXY_EXE) $(KUBEPEERS_EXE) $(WEAVENPC_EXE) $(WEAVEWAIT_EXE) $(WEAVEWAIT_NOOP_EXE) $(WEAVEWAIT_NOMCAST_EXE) $(WEAVEUTIL_EXE) $(RUNNER_EXE) $(TEST_TLS_EXE) $(MANIFEST_TOOL_EXE) $(NETWORKTESTER_EXE)

-WEAVER_UPTODATE=.weaver.uptodate
-WEAVEEXEC_UPTODATE=.weaveexec.uptodate
+# These stamp files are used to mark the current state of the build; whether an image has been built or not
+BUILD_UPTODATE=.build.uptodate
+WEAVER_UPTODATE=.weaver$(ARCH_EXT).uptodate
+WEAVEEXEC_UPTODATE=.weaveexec$(ARCH_EXT).uptodate
+PLUGIN_UPTODATE=.net-plugin$(ARCH_EXT).uptodate
+WEAVEKUBE_UPTODATE=.weavekube$(ARCH_EXT).uptodate
+WEAVENPC_UPTODATE=.weavenpc$(ARCH_EXT).uptodate
+WEAVEDB_UPTODATE=.weavedb.uptodate
+NETWORKTESTER_UPTODATE=.networktester$(ARCH_EXT).uptodate

-IMAGES_UPTODATE=$(WEAVER_UPTODATE) $(WEAVEEXEC_UPTODATE)
+IMAGES_UPTODATE=$(WEAVER_UPTODATE) $(WEAVEEXEC_UPTODATE) $(WEAVEKUBE_UPTODATE) $(WEAVENPC_UPTODATE) $(NETWORKTESTER_UPTODATE)

-WEAVER_IMAGE=$(DOCKERHUB_USER)/weave
-WEAVEEXEC_IMAGE=$(DOCKERHUB_USER)/weaveexec
+# The names of the images. Note that the images for other architectures than amd64 have a suffix in the image name.
+WEAVER_IMAGE=$(DOCKERHUB_USER)/weave$(ARCH_EXT)
+WEAVEEXEC_IMAGE=$(DOCKERHUB_USER)/weaveexec$(ARCH_EXT)
+WEAVEKUBE_IMAGE=$(DOCKERHUB_USER)/weave-kube$(ARCH_EXT)
+WEAVENPC_IMAGE=$(DOCKERHUB_USER)/weave-npc$(ARCH_EXT)
+BUILD_IMAGE=weaveworks/weavebuild
+WEAVEDB_IMAGE=$(DOCKERHUB_USER)/weavedb
+PLUGIN_IMAGE=$(DOCKERHUB_USER)/net-plugin
+NETWORKTESTER_IMAGE=$(DOCKERHUB_USER)/network-tester

-IMAGES=$(WEAVER_IMAGE) $(WEAVEEXEC_IMAGE)
+IMAGES=$(WEAVER_IMAGE) $(WEAVEEXEC_IMAGE) $(WEAVEKUBE_IMAGE) $(WEAVENPC_IMAGE) $(WEAVEDB_IMAGE) $(NETWORKTESTER_IMAGE)

-WEAVE_EXPORT=weave.tar
+PLUGIN_WORK_DIR="prog/net-plugin/rootfs"
+PLUGIN_BUILD_IMG="plugin-builder"

-WEAVEEXEC_DOCKER_VERSION=1.3.1
-DOCKER_DISTRIB=prog/weaveexec/docker-$(WEAVEEXEC_DOCKER_VERSION).tgz
-DOCKER_DISTRIB_URL=https://get.docker.com/builds/Linux/x86_64/docker-$(WEAVEEXEC_DOCKER_VERSION).tgz
+PUBLISH=publish_weave publish_weaveexec publish_weave-kube publish_weave-npc
+PUSH_ML=push_ml_weave push_ml_weaveexec push_ml_weave-kube push_ml_weave-npc
+
+WEAVE_EXPORT=weave$(ARCH_EXT).tar.gz
+
+DOCKER_VERSION=1.10.3
+DOCKER_DISTRIB=prog/weaveexec/docker-$(DOCKER_VERSION).tgz
+DOCKER_DISTRIB_URL=https://get.docker.com/builds/Linux/$(WEAVEEXEC_DOCKER_ARCH)/docker-$(DOCKER_VERSION).tgz
 NETGO_CHECK=@strings $@ | grep cgo_stub\\\.go >/dev/null || { \
 	rm $@; \
 	echo "\nYour go standard library was built without the 'netgo' build tag."; \
@@ -43,104 +146,281 @@ NETGO_CHECK=@strings $@ | grep cgo_stub\\\.go >/dev/null || { \
 	echo "    sudo go install -tags netgo std"; \
 	false; \
 }
-BUILD_FLAGS=-ldflags "-extldflags \"-static\" -X main.version $(WEAVE_VERSION)" -tags netgo
+# The flags we are passing to go build. -extldflags -static for making a static binary, 
+# -linkmode external for linking external C libraries into the binary, -X main.version for telling the
+# Go binary which version it is, -tags netgo for enforcing the native Go DNS resolver
+BUILD_FLAGS=-i -ldflags "-linkmode external -extldflags -static -X main.version=$(WEAVE_VERSION)" -tags netgo

 PACKAGE_BASE=$(shell go list -e ./)

-all: $(WEAVE_EXPORT) $(RUNNER_EXE)
+all: $(WEAVE_EXPORT)
+testrunner: $(RUNNER_EXE) $(TEST_TLS_EXE)

-travis: $(EXES)
+$(WEAVER_EXE) $(WEAVEUTIL_EXE): common/*.go common/*/*.go net/*.go net/*/*.go
+$(WEAVER_EXE): router/*.go ipam/*.go ipam/*/*.go db/*.go nameserver/*.go prog/weaver/*.go
+$(WEAVER_EXE): api/*.go plugin/*.go plugin/*/*
+$(WEAVER_EXE):  proxy/*.go
+$(WEAVEUTIL_EXE): prog/weaveutil/*.go net/*.go plugin/net/*.go plugin/ipam/*.go db/*.go
+$(SIGPROXY_EXE): prog/sigproxy/*.go
+$(KUBEPEERS_EXE): prog/kube-peers/*.go
+$(WEAVENPC_EXE): prog/weave-npc/*.go npc/*.go npc/*/*.go
+$(TEST_TLS_EXE): test/tls/*.go
+$(RUNNER_EXE): tools/runner/*.go
+$(MANIFEST_TOOL_EXE): $(MANIFEST_TOOL_DIR)/*.go
+$(WEAVEWAIT_NOOP_EXE): prog/weavewait/*.go
+$(WEAVEWAIT_EXE): prog/weavewait/*.go net/*.go
+$(WEAVEWAIT_NOMCAST_EXE): prog/weavewait/*.go net/*.go
+$(NETWORKTESTER_EXE): test/images/network-tester/*.go
+tests: tools/.git
+lint: tools/.git

-update:
-	go get -u -f -v -tags netgo $(addprefix ./,$(dir $(EXES)))
+ifeq ($(BUILD_IN_CONTAINER),true)

-$(WEAVER_EXE) $(WEAVEPROXY_EXE): common/*.go common/*/*.go net/*.go
+# This make target compiles all binaries inside of the weaveworks/build container
+# It bind-mounts the source into the container and passes all important variables
+exes $(EXES) tests lint: $(BUILD_UPTODATE)
+	git submodule update --init
+# Containernetworking has another copy of vishvananda/netlink which leads to duplicate definitions
+	-@rm -r vendor/github.com/containernetworking/cni/vendor
+	@mkdir -p $(shell pwd)/.pkg
+	$(SUDO) docker run $(RM) $(RUN_FLAGS) \
+	    -v $(shell pwd):/go/src/github.com/weaveworks/weave \
+		-v $(shell pwd)/.pkg:/go/pkg \
+		-e GOARCH=$(ARCH) -e CGO_ENABLED=1 -e GOOS=linux -e CIRCLECI -e CIRCLE_BUILD_NUM -e CIRCLE_NODE_TOTAL -e CIRCLE_NODE_INDEX -e COVERDIR -e SLOW -e DEBUG \
+		$(BUILD_IMAGE) COVERAGE=$(COVERAGE) WEAVE_VERSION=$(WEAVE_VERSION) CC=$(CC) QEMUARCH=$(QEMUARCH) CGO_LDFLAGS=$(CGO_LDFLAGS) $@
+	touch $@
+
+else
+
+exes: $(EXES)
+
+$(WEAVER_EXE):
 ifeq ($(COVERAGE),true)
-	$(eval COVERAGE_MODULES := $(shell (go list ./$(@D); go list -f '{{join .Deps "\n"}}' ./$(@D) | grep "^$(PACKAGE_BASE)/") | paste -s -d,))
-	go get -t -tags netgo ./$(@D)
+	$(eval COVERAGE_MODULES := $(shell (go list ./$(@D); go list -f '{{join .Deps "\n"}}' ./$(@D) | grep "^$(PACKAGE_BASE)/") | grep -v "^$(PACKAGE_BASE)/vendor/" | paste -s -d,))
 	go test -c -o ./$@ $(BUILD_FLAGS) -v -covermode=atomic -coverpkg $(COVERAGE_MODULES) ./$(@D)/
 else
-	go get -tags netgo ./$(@D)
 	go build $(BUILD_FLAGS) -o $@ ./$(@D)
 endif
 	$(NETGO_CHECK)

-$(NETCHECK_EXE): common/*.go common/*/*.go net/*.go
-	go get -tags netgo ./$(@D)
+$(WEAVEUTIL_EXE) $(KUBEPEERS_EXE) $(WEAVENPC_EXE) $(NETWORKTESTER_EXE):
 	go build $(BUILD_FLAGS) -o $@ ./$(@D)
 	$(NETGO_CHECK)

-$(WEAVER_EXE): router/*.go ipam/*.go ipam/*/*.go nameserver/*.go prog/weaver/*.go
-$(WEAVEPROXY_EXE): proxy/*.go prog/weaveproxy/main.go
-$(NETCHECK_EXE): prog/netcheck/netcheck.go
+$(WEAVEWAIT_EXE):
+	go build $(BUILD_FLAGS) -tags "netgo iface mcast" -o $@ ./$(@D)
+	$(NETGO_CHECK)

-# Sigproxy and weavewait need separate rules as they fail the netgo check in
+$(WEAVEWAIT_NOMCAST_EXE):
+	go build $(BUILD_FLAGS) -tags "netgo iface" -o $@ ./$(@D)
+	$(NETGO_CHECK)
+
+# These programs need a separate rule as they fail the netgo check in
 # the main build stanza due to not importing net package
-$(SIGPROXY_EXE): prog/sigproxy/main.go
-$(WEAVEWAIT_EXE): prog/weavewait/*.go net/*.go
-$(RUNNER_EXE): testing/runner/runner.go
-$(DOCKERTLSARGS_EXE): prog/docker_tls_args/*.go
-
-$(WEAVEWAIT_EXE) $(SIGPROXY_EXE) $(RUNNER_EXE) $(DOCKERTLSARGS_EXE):
-	go get -tags netgo ./$(@D)
+$(SIGPROXY_EXE) $(TEST_TLS_EXE) $(WEAVEWAIT_NOOP_EXE) $(RUNNER_EXE) $(MANIFEST_TOOL_EXE):
 	go build $(BUILD_FLAGS) -o $@ ./$(@D)

-$(WEAVEWAIT_NOOP_EXE): prog/weavewait/*.go
-	go get -tags netgo ./$(@D)
-	go build $(BUILD_FLAGS) -tags noop -o $@ ./$(@D)
+tests:
+	./tools/test -no-go-get -netgo -timeout 8m

-$(WEAVER_UPTODATE): prog/weaver/Dockerfile $(WEAVER_EXE)
-	$(SUDO) docker build -t $(WEAVER_IMAGE) prog/weaver
+lint:
+	./tools/lint -nocomment -notestpackage
+
+endif
+
+# This rule makes sure the build image is up-to-date.
+# It also makes sure the multiarch hooks are reqistered in the kernel so the QEMU emulation works
+$(BUILD_UPTODATE): build/*
+	$(SUDO) docker build -t $(BUILD_IMAGE) build/
+	$(SUDO) docker run --rm --privileged multiarch/qemu-user-static:register --reset
 	touch $@

-$(WEAVEEXEC_UPTODATE): prog/weaveexec/Dockerfile $(DOCKER_DISTRIB) weave $(SIGPROXY_EXE) $(WEAVEPROXY_EXE) $(WEAVEWAIT_EXE) $(WEAVEWAIT_NOOP_EXE) $(NETCHECK_EXE) $(DOCKERTLSARGS_EXE)
-	cp weave prog/weaveexec/weave
+# Creates the Dockerfile.your-user-here file from the template
+# Also replaces all placeholders with real values
+# If the architecture is amd64, it deletes all CROSS_BUILD lines
+# but otherwise, it only removes the "CROSS_BUILD_" placeholder and downloads QEMU
+%/Dockerfile.$(DOCKERHUB_USER): %/Dockerfile.template
+	echo "DOCKERHUB_USER|$(DOCKERHUB_USER)|g;s|ARCH_EXT|$(ARCH_EXT)|g;s|ALPINE_BASEIMAGE|$(ALPINE_BASEIMAGE)|g;s|QEMUARCH|$(QEMUARCH)"
+	sed -e "s|DOCKERHUB_USER|$(DOCKERHUB_USER)|g;s|ARCH_EXT|$(ARCH_EXT)|g;s|ALPINE_BASEIMAGE|$(ALPINE_BASEIMAGE)|g;s|QEMUARCH|$(QEMUARCH)|g" $^ > $@
+ifeq ($(ARCH),amd64)
+# When building "normally" for amd64, remove the whole line, it has no part in the amd64 image
+	sed -i "/CROSS_BUILD_/d" $@
+else
+# When cross-building, only the placeholder "CROSS_BUILD_" should be removed
+# Register /usr/bin/qemu-ARCH-static as the handler for ARM binaries in the kernel
+	curl -sSL https://github.com/multiarch/qemu-user-static/releases/download/$(QEMU_VERSION)/x86_64_qemu-$(QEMUARCH)-static.tar.gz | tar -xz -C $(shell dirname $@)
+	cd $(shell dirname $@) && sha256sum -c $(shell pwd)/build/shasums/qemu-$(QEMUARCH)-static.sha256sum
+	sed -i "s/CROSS_BUILD_//g" $@
+endif
+
+# The targets below builds the weave images
+$(WEAVER_UPTODATE): prog/weaver/Dockerfile.$(DOCKERHUB_USER) $(WEAVER_EXE) weave $(WEAVEUTIL_EXE)
+	cp $(WEAVEUTIL_EXE) prog/weaver/weaveutil
+	cp weave prog/weaver/weave
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker build -f prog/weaver/Dockerfile.$(DOCKERHUB_USER) -t $(WEAVER_IMAGE) prog/weaver
+	touch $@
+
+$(WEAVEEXEC_UPTODATE): prog/weaveexec/Dockerfile.$(DOCKERHUB_USER) prog/weaveexec/symlink $(DOCKER_DISTRIB) $(SIGPROXY_EXE) $(WEAVEWAIT_EXE) $(WEAVEWAIT_NOOP_EXE) $(WEAVEWAIT_NOMCAST_EXE) $(WEAVER_UPTODATE)
 	cp $(SIGPROXY_EXE) prog/weaveexec/sigproxy
-	cp $(WEAVEPROXY_EXE) prog/weaveexec/weaveproxy
 	cp $(WEAVEWAIT_EXE) prog/weaveexec/weavewait
 	cp $(WEAVEWAIT_NOOP_EXE) prog/weaveexec/weavewait_noop
-	cp $(NETCHECK_EXE) prog/weaveexec/netcheck
-	cp $(DOCKERTLSARGS_EXE) prog/weaveexec/docker_tls_args
-	cp $(DOCKER_DISTRIB) prog/weaveexec/docker.tgz
-	$(SUDO) docker build -t $(WEAVEEXEC_IMAGE) prog/weaveexec
+	cp $(WEAVEWAIT_NOMCAST_EXE) prog/weaveexec/weavewait_nomcast
+	tar -xf $(DOCKER_DISTRIB) usr/local/bin/docker -O > prog/weaveexec/docker
+	chmod +x prog/weaveexec/docker
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker build -f prog/weaveexec/Dockerfile.$(DOCKERHUB_USER) -t $(WEAVEEXEC_IMAGE) prog/weaveexec
 	touch $@

-$(WEAVE_EXPORT): $(IMAGES_UPTODATE)
-	$(SUDO) docker save $(addsuffix :latest,$(IMAGES)) > $@
+# Builds Docker plugin.
+$(PLUGIN_UPTODATE): prog/net-plugin/launch.sh prog/net-plugin/config.json $(WEAVER_UPTODATE)
+	-$(SUDO) docker rm -f $(PLUGIN_BUILD_IMG) 2>/dev/null
+	$(SUDO) docker create --name=$(PLUGIN_BUILD_IMG) $(WEAVER_IMAGE) true
+	rm -rf $(PLUGIN_WORK_DIR)
+	mkdir $(PLUGIN_WORK_DIR)
+	$(SUDO) docker export $(PLUGIN_BUILD_IMG) | tar -x -C $(PLUGIN_WORK_DIR)
+	$(SUDO) docker rm -f $(PLUGIN_BUILD_IMG)
+	cp prog/net-plugin/launch.sh $(PLUGIN_WORK_DIR)/home/weave/launch.sh
+	-$(SUDO) docker plugin disable $(PLUGIN_IMAGE):$(WEAVE_VERSION) 2>/dev/null
+	-$(SUDO) docker plugin rm $(PLUGIN_IMAGE):$(WEAVE_VERSION) 2>/dev/null
+	$(SUDO) docker plugin create $(PLUGIN_IMAGE):$(WEAVE_VERSION) prog/net-plugin
+	-$(SUDO) docker plugin disable $(PLUGIN_IMAGE):$(NET_PLUGIN_LATEST) 2>/dev/null
+	-$(SUDO) docker plugin rm $(PLUGIN_IMAGE):$(NET_PLUGIN_LATEST) 2>/dev/null
+	$(SUDO) docker plugin create $(PLUGIN_IMAGE):$(NET_PLUGIN_LATEST) prog/net-plugin
+	touch $@
+
+$(WEAVEKUBE_UPTODATE): prog/weave-kube/Dockerfile.$(DOCKERHUB_USER) prog/weave-kube/launch.sh $(KUBEPEERS_EXE) $(WEAVER_UPTODATE)
+	cp $(KUBEPEERS_EXE) prog/weave-kube/
+	$(SUDO) docker build -f prog/weave-kube/Dockerfile.$(DOCKERHUB_USER) -t $(WEAVEKUBE_IMAGE) prog/weave-kube
+	touch $@
+
+$(WEAVENPC_UPTODATE): prog/weave-npc/Dockerfile.$(DOCKERHUB_USER) $(WEAVENPC_EXE) prog/weave-npc/ulogd.conf
+	$(SUDO) docker build -f prog/weave-npc/Dockerfile.$(DOCKERHUB_USER) -t $(WEAVENPC_IMAGE) prog/weave-npc
+	touch $@
+
+$(WEAVEDB_UPTODATE): prog/weavedb/Dockerfile
+	$(SUDO) docker build -t $(WEAVEDB_IMAGE) prog/weavedb
+	touch $@
+
+$(NETWORKTESTER_UPTODATE): test/images/network-tester/Dockerfile $(NETWORKTESTER_EXE)
+	$(SUDO) docker build -t $(NETWORKTESTER_IMAGE) test/images/network-tester
+	touch $@
+
+$(WEAVE_EXPORT): $(IMAGES_UPTODATE) $(WEAVEDB_UPTODATE)
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker save $(addsuffix :latest,$(IMAGES)) | gzip > $@

 $(DOCKER_DISTRIB):
 	curl -o $(DOCKER_DISTRIB) $(DOCKER_DISTRIB_URL)
+	cd $(shell dirname $@) && sha256sum -c $(shell pwd)/build/shasums/docker-tgz-$(WEAVEEXEC_DOCKER_ARCH).sha256sum

-tests: tools/.git
-	tools/test
-
-lint: tools/.git
-	tools/lint -nocomment -notestpackage .
-
-tools/.git:
+tools/.git $(MANIFEST_TOOL_DIR)/.git:
 	git submodule update --init

-$(PUBLISH): publish_%: $(IMAGES_UPTODATE)
-	$(SUDO) docker tag -f $(DOCKERHUB_USER)/$* $(DOCKERHUB_USER)/$*:$(WEAVE_VERSION)
-	$(SUDO) docker push   $(DOCKERHUB_USER)/$*:$(WEAVE_VERSION)
-ifneq ($(UPDATE_LATEST),false)
-	$(SUDO) docker push   $(DOCKERHUB_USER)/$*:latest
+# CODE FOR PUBLISHING THE IMAGES
+
+# Push plugin
+plugin_publish: $(PLUGIN_UPTODATE)
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker plugin push $(PLUGIN_IMAGE):$(WEAVE_VERSION)
+# "$(NET_PLUGIN_LATEST)" means "stable release" here, so only push that if explicitly told to
+ifeq ($(UPDATE_LATEST),true)
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker plugin push $(PLUGIN_IMAGE):$(NET_PLUGIN_LATEST)
 endif

-publish: $(PUBLISH)
+# This target first runs "make publish" for each architecture
+# Then it pushes the manifest lists
+publish: $(addprefix sub-publish-,$(ALL_ARCHITECTURES)) $(WEAVEDB_UPTODATE)
+	$(MAKE) DOCKER_HOST=$(DOCKER_HOST) DOCKERHUB_USER=$(DOCKERHUB_USER) WEAVE_VERSION=$(WEAVE_VERSION) UPDATE_LATEST=$(UPDATE_LATEST) $(PUSH_ML)
+ifeq ($(PUBLISH_WEAVEDB),true)
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker push $(DOCKERHUB_USER)/weavedb:latest
+endif
+
+sub-publish-%:
+	$(MAKE) ARCH=$* DOCKER_HOST=$(DOCKER_HOST) clean-bin
+	$(MAKE) ARCH=$* DOCKER_HOST=$(DOCKER_HOST) DOCKERHUB_USER=$(DOCKERHUB_USER) WEAVE_VERSION=$(WEAVE_VERSION) UPDATE_LATEST=$(UPDATE_LATEST) $(IMAGES_UPTODATE)
+	$(MAKE) ARCH=$* DOCKER_HOST=$(DOCKER_HOST) DOCKERHUB_USER=$(DOCKERHUB_USER) WEAVE_VERSION=$(WEAVE_VERSION) UPDATE_LATEST=$(UPDATE_LATEST) publish-one-arch
+
+publish-one-arch: $(PUBLISH)
+
+# This rule handles the pushing of the built images
+$(PUBLISH): publish_%: $(IMAGES_UPTODATE)
+# Tag :latest with the real version
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker tag  $(DOCKERHUB_USER)/$*$(ARCH_EXT) $(DOCKERHUB_USER)/$*$(ARCH_EXT):$(WEAVE_VERSION)
+# Push the image with the arch suffix for arm and arm64, and without suffix for amd64
+	$(MAKE) DOCKER_HOST=$(DOCKER_HOST) DOCKERHUB_USER=$(DOCKERHUB_USER) WEAVE_VERSION=$(WEAVE_VERSION) UPDATE_LATEST=$(UPDATE_LATEST) push_$*$(ARCH_EXT)
+
+ifeq ($(ARCH),amd64)
+# If the architecture is amd64, add the -amd64 suffix.
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker tag  $(DOCKERHUB_USER)/$*:$(WEAVE_VERSION) $(DOCKERHUB_USER)/$*-amd64:$(WEAVE_VERSION)
+# If the version is latest, tag the -amd64 with latest as well
+ifneq ($(UPDATE_LATEST),false)
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker tag  $(DOCKERHUB_USER)/$*-amd64:$(WEAVE_VERSION) $(DOCKERHUB_USER)/$*-amd64:latest
+endif
+# Push the image with the -amd64-suffix so BINARY-ARCH-named images exist for all arches, so manifest lists may be made and ARCH is replaceable in scripts
+	$(MAKE) DOCKER_HOST=$(DOCKER_HOST) DOCKERHUB_USER=$(DOCKERHUB_USER) WEAVE_VERSION=$(WEAVE_VERSION) UPDATE_LATEST=$(UPDATE_LATEST) push_$*-amd64
+endif
+
+# This target pushes an image, and if UPDATE_LATEST is anything but false, also updates the latest tag
+# It takes one parameter, the image name
+push_%:
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker push $(DOCKERHUB_USER)/$*:$(WEAVE_VERSION)
+ifneq ($(UPDATE_LATEST),false)
+	$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker push $(DOCKERHUB_USER)/$*:latest
+endif
+
+# This target pushes a manifest list; it takes one parameter, the image name.
+# The variable UPDATE_LATEST controls whether it updates the versioned tag and/or the latest tag
+$(PUSH_ML): push_ml_%: $(MANIFEST_TOOL_EXE)
+ifneq ($(UPDATE_LATEST),latest-only)
+	$(MANIFEST_TOOL_EXE) push from-args --platforms $(ML_PLATFORMS) --template $(DOCKERHUB_USER)/$*-ARCH:$(WEAVE_VERSION) --target $(DOCKERHUB_USER)/$*:$(WEAVE_VERSION)
+endif
+ifneq ($(UPDATE_LATEST),false)
+# Push the manifest list to :latest as well
+	$(MANIFEST_TOOL_EXE) push from-args --platforms $(ML_PLATFORMS) --template $(DOCKERHUB_USER)/$*-ARCH:latest --target $(DOCKERHUB_USER)/$*:latest
+endif
+
+clean-work-dir:
+	rm -rf $(PLUGIN_WORK_DIR)

 clean-bin:
-	-$(SUDO) docker rmi $(IMAGES)
-	go clean -r $(addprefix ./,$(dir $(EXES)))
-	rm -f $(EXES) $(IMAGES_UPTODATE) $(WEAVE_EXPORT)
+	-$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker rmi $(IMAGES)
+	find prog -type f -name "Dockerfile.*" -not -name "Dockerfile.template" -print | xargs rm -f
+	find prog -type f -name "*qemu-*" -print | xargs rm -f
+	rm -rf $(EXES) $(IMAGES_UPTODATE) $(WEAVEDB_UPTODATE) weave*.tar.gz $(DOCKER_DISTRIB) prog/weaveexec/docker .pkg

-clean: clean-bin
-	rm -rf test/tls/*.pem test/coverage.* test/coverage
+clean: clean-bin clean-work-dir
+	-$(SUDO) DOCKER_HOST=$(DOCKER_HOST) docker rmi $(BUILD_IMAGE)
+	rm -rf test/tls/*.pem test/coverage.* test/coverage $(BUILD_UPTODATE) $(MANIFEST_TOOL_EXE)

 build:
 	$(SUDO) go clean -i net
 	$(SUDO) go install -tags netgo std
 	$(MAKE)

-run-smoketests: all
+run-smoketests: all testrunner
 	cd test && ./setup.sh && ./run_all.sh
+
+integration-tests: all testrunner
+# Usage:
+#   $ make \
+#     NAME="<prefix used to name VMs and other resources>" \
+#     PROVIDER="<provider among {vagrant|gcp|aws|do}>" \
+#     NUM_HOSTS="<# test machines>" \
+#     PLAYBOOK="<filename>" \
+#     RUNNER_ARGS="<...>" \
+#     TESTS="<...>" \  # Can be set to only run one or a few tests instead of the full test suite.
+#     DOCKER_VERSION=<...> \
+#     KUBERNETES_VERSION=<...> \
+#     KUBERNETES_CNI_VERSION=<...> \
+#     <...> # See also run-integration-test.sh for all variables and individual functions.
+#     integration-tests
+#
+	RUNNER_ARGS="-parallel" ./test/run-integration-tests.sh
+
+# This target is only intended for use in Netlify CI environment for generating preview pages on feature branches and pull requests.
+# We need to obtain website repository (checked out under `site-build`) and place `site` directory into the context (`site-build/_weave_net_docs`).
+# We then run make in `site-build` and Netlify will publish the output (`site-build/_site`).
+netlify-site-preview:
+	@mkdir -p site-build
+	@curl --user $(WEBSITE_GITHUB_USER) --silent 'https://codeload.github.com/weaveworks/website-next/tar.gz/$(WEBSITE_BRANCH)' \
+	  | tar --extract --gunzip --directory site-build --strip 1
+	@cp -r site site-build/_weave_net_docs
+	@$(MAKE) -C site-build netlify_ensure_install
+	@$(MAKE) -C site-build BUILD_ENV=netlify
@@ -1,3 +1,3 @@
 Weave
-Copyright 2014-2015 Weaveworks Ltd.
+Copyright 2014-2016 Weaveworks Ltd.
 This product includes software developed at Weaveworks Ltd.
@@ -1,174 +1,43 @@
-# Weave - weaving containers into applications
+# Weave Net - Weaving Containers into Applications

-[![Build Status](https://travis-ci.org/weaveworks/weave.svg?branch=master)](https://travis-ci.org/weaveworks/weave) [![Integration Tests](https://circleci.com/gh/weaveworks/weave/tree/master.svg?style=shield&circle-token=4933c7dabb3d0383e62117565cb9d16df7b1a811)](https://circleci.com/gh/weaveworks/weave) [![Coverage Status](https://coveralls.io/repos/weaveworks/weave/badge.svg)](https://coveralls.io/r/weaveworks/weave)
+[![Integration Tests](https://circleci.com/gh/weaveworks/weave/tree/master.svg?style=shield)](https://circleci.com/gh/weaveworks/weave)
+[![Coverage Status](https://coveralls.io/repos/weaveworks/weave/badge.svg)](https://coveralls.io/r/weaveworks/weave)
+[![Go Report Card](https://goreportcard.com/badge/github.com/weaveworks/weave)](https://goreportcard.com/report/github.com/weaveworks/weave)
+[![Docker Pulls](https://img.shields.io/docker/pulls/weaveworks/weave.svg?maxAge=604800)](https://hub.docker.com/r/weaveworks/weave/)

 # About Weaveworks

-[Weaveworks](http://weave.works) is the company that develops Weave -
-the most productive way for developers to connect, observe and control
-Docker containers. To learn about our products, including getting
-started tutorials, visit our [website](http://weave.works) and
-[documentation](http://docs.weave.works) or continue to read about
-some of the more technical aspects of Weave in this readme.
+[Weaveworks](https://www.weave.works) is the company that delivers the most productive way for developers to connect, observe and control
+Docker containers.

-# Weave
+This repository contains [Weave Net](https://www.weave.works/products/weave-net/), the first product developed by Weaveworks, and with over 8 million downloads to date, it enables you to get started with Docker clusters and portable apps in a fraction of the time compared with other solutions.

-Weave creates a virtual network that connects Docker containers
-deployed across multiple hosts and enables their automatic discovery.
+## Weave Cloud: 

-![Weave Virtual Network](/docs/virtual-network.png?raw=true "Weave Virtual Network")
+[Weave Cloud](https://www.weave.works/product/cloud/) is software-as-a-service that simplifies deployment, monitoring and management for containers and microservices. It extends and complements popular orchestrators, and enables developers and DevOps to realize faster deployments, insightful monitoring, visualization and networking.

-Applications use the network just as if the containers were all
-plugged into the same network switch, with no need to configure port
-mappings, links, etc. Services provided by application containers on
-the weave network can be made accessible to the outside world,
-regardless of where those containers are running. Similarly, existing
-internal systems can be exposed to application containers irrespective
-of their location.
+Weave Cloud is built using these Open Source projects: [Weave Scope](https://www.weave.works/products/weave-scope/), a powerful container monitoring tool that automatically maps Docker containers and their interactions, [Weave Cortex](https://github.com/weaveworks/cortex), a horizontally-scalable version of Prometheus, and [Weave Flux](https://www.weave.works/products/weave-flux/), a continuous deployment tool that works with Kubernetes.

-![Weave Deployment](/docs/deployment.png?raw=true "Weave Deployment")
+# Weave Net

-Weave can traverse firewalls and operate in partially connected
-networks. Traffic can be encrypted, allowing hosts to be connected
-across an untrusted network.
+Weave Net creates a virtual network that connects Docker containers across multiple hosts and enables their automatic discovery. With Weave Net, portable microservices-based applications consisting of multiple containers can run anywhere: on one host, multiple hosts or even across cloud providers and data centers. Applications use the network just as if the containers were all plugged into the same network switch, without having to configure port mappings, ambassadors or links.

-With weave you can easily construct applications consisting of
-multiple containers, running anywhere.
+Services provided by application containers on the Weave network can be exposed to the outside world, regardless of where they are running. Similarly, existing internal systems can be opened to accept connections from application containers irrespective of their location.

-Weave works alongside Docker's existing (single host) networking
-capabilities, so these can continue to be used by containers.
+## <a name="help"></a>Getting help

-## Installation
+If you have any questions about, feedback for or a problem with Weave Net:

-Ensure you are running Linux (kernel 3.8 or later) and have Docker
-(version 1.3.1 or later) installed. Then install weave with
+- Read [the Weave Net docs](https://www.weave.works/docs/net/latest/introducing-weave/).
+- Invite yourself to the <a href="https://weaveworks.github.io/community-slack/" target="_blank"> #weave-community </a> slack channel.
+- Ask a question on the <a href="https://weave-community.slack.com/messages/general/"> #weave-community</a> slack channel.
+- Join the <a href="https://www.meetup.com/pro/Weave/"> Weave User Group </a> and get invited to online talks, hands-on training and meetups in your area.
+- Send an email to <a href="mailto:weave-users@weave.works">weave-users@weave.works</a>
+- <a href="https://github.com/weaveworks/weave/issues/new">File an issue.</a>

-    sudo curl -L git.io/weave -o /usr/local/bin/weave
-    sudo chmod a+x /usr/local/bin/weave
+Your feedback is always welcome!

-For usage on OSX (with Docker Machine) you first need to make sure
-that a VM is running and configured, as shown in [the Docker Machine
-documentation](https://docs.docker.com/installation/mac/#from-your-shell).
-Then you can launch weave directly from the OSX host.
+## Further Information

-CoreOS users see [here](https://github.com/fintanr/weave-gs/blob/master/coreos-simple/user-data) for an example of installing weave using cloud-config.
-
-Amazon ECS users see
-[here](https://github.com/weaveworks/guides/blob/master/aws-ecs/LATESTAMIs.md)
-for the latest Weave AMIs and
-[here](http://weave.works/guides/service-discovery-with-weave-aws-ecs.html) to get started with Weave on ECS.
-
-Weave respects the environment variable `DOCKER_HOST`, so you can run
-it locally to control a weave network on a remote host.
-
-## Quick Start Screencast
-
-<a href="https://youtu.be/kihQCCT1ykE" alt="Click to watch the screencast" target="_blank">
-  <img src="/docs/hello-screencast.png" />
-</a>
-
-## Example
-
-Say you have docker running on two hosts, accessible to each other as
-`$HOST1` and `$HOST2`, and want to deploy an application consisting of
-two containers, one on each host.
-
-On $HOST1 we run:
-
-    host1$ weave launch
-    host1$ eval $(weave env)
-    host1$ docker run --name a1 -ti ubuntu
-
-> NB: If the first command results in an error like
-> `http:///var/run/docker.sock/v1.19/containers/create: dial unix
-> /var/run/docker.sock: permission denied. Are you trying to connect
-> to a TLS-enabled daemon without TLS?` then you likely need to be
-> 'root' in order to connect to the Docker daemon. If so, run the
-> above and all subsequent commands in a *single* root shell (e.g. one
-> created with `sudo -s`). Do *not* prefix individual commands with
-> `sudo`, since some commands modify environment entries and hence
-> they all need to be executed from the same shell.
-
-The first line runs weave. The second line configures our environment
-so that containers launched via the docker command line are
-automatically attached to the weave network. Finally, we run our
-application container.
-
-That's it! If our application consists of more than one container on
-this host we simply launch them with `docker run` as appropriate.
-
-Next we repeat similar steps on `$HOST2`...
-
-    host2$ weave launch $HOST1
-    host2$ eval $(weave env)
-    host2$ docker run --name a2 -ti ubuntu
-
-The only difference, apart from the name of the application container,
-is that we tell our weave that it should peer with the weave on
-`$HOST1` (specified as the IP address or hostname, and optional
-`:port`, by which `$HOST2` can reach it). NB: if there is a firewall
-between `$HOST1` and `$HOST2`, you must permit traffic to the weave
-control port (TCP 6783) and data ports (UDP 6783/6784).
-
-Note that we could instead have told the weave on `$HOST1` to connect to
-`$HOST2`, or told both about each other. Order does not matter here;
-weave automatically (re)connects to peers when they become
-available. Also, we can tell weave to connect to multiple peers by
-supplying multiple addresses, separated by spaces. And we can
-[add peers dynamically](http://docs.weave.works/weave/latest_release/features.html#dynamic-topologies).
-
-Weave must be started once per host. The relevant container images are
-pulled down on demand, but if you wish you can preload them by running
-`weave setup` - this is particularly useful for automated deployments,
-and ensures that there are no delays during later operations.
-
-Now that we've got everything set up, let's see whether our containers
-can talk to each other...
-
-In the container started on `$HOST1`...
-
-    root@a1:/# ping -c 1 -q a2
-    PING a2.weave.local (10.40.0.2) 56(84) bytes of data.
-    --- a2.weave.local ping statistics ---
-    1 packets transmitted, 1 received, 0% packet loss, time 0ms
-    rtt min/avg/max/mdev = 0.341/0.341/0.341/0.000 ms
-
-Similarly, in the container started on `$HOST2`...
-
-    root@a2:/# ping -c 1 -q a1
-    PING a1.weave.local (10.32.0.2) 56(84) bytes of data.
-    --- a1.weave.local ping statistics ---
-    1 packets transmitted, 1 received, 0% packet loss, time 0ms
-    rtt min/avg/max/mdev = 0.366/0.366/0.366/0.000 ms
-
-So there we have it, two containers on separate hosts happily talking
-to each other.
-
-## Find out more
-
- * [Documentation homepage](http://docs.weave.works/weave/latest_release/)
- * [Getting Started Guides](http://weave.works/guides/)
- * [Features](http://docs.weave.works/weave/latest_release/features.html)
- * [Troubleshooting](http://docs.weave.works/weave/latest_release/troubleshooting.html)
- * [Building](http://docs.weave.works/weave/latest_release/building.html)
- * [How it works](http://docs.weave.works/weave/latest_release/how-it-works.html)
-
-## Contact Us
-
-Found a bug, want to suggest a feature, or have a question?  You can
-[File an issue](https://github.com/weaveworks/weave/issues).  And
-there is a [Weave Users Google
-Group](https://groups.google.com/a/weave.works/forum/#!forum/weave-users),
-which you can post to by email at weave-users@weave.works.  Or you can
-contact Weaveworks directly at help@weave.works. When reporting a
-bug, please include which version of weave you are running, as shown
-by `weave version`.
-
-Follow on Twitter:
-[@weaveworks](https://twitter.com/weaveworks).
-
-Read the Weave blog:
-[Weaveblog](http://weaveblog.com/).
-
-IRC:
-[#weavenetwork](https://botbot.me/freenode/weavenetwork/)
+* [Installing Weave Net](https://www.weave.works/docs/net/latest/installing-weave/)
+* [Weave Net Product Overview](https://www.weave.works/products/weave-net/)
@@ -1,41 +1,36 @@
-VAGRANTFILE_API_VERSION = "2"
-
-require './vagrant-common.rb'
+require File.expand_path(File.join(File.dirname(__FILE__), 'vagrant-common.rb'))

 vm_ip = "172.16.0.3" # arbitrary private IP

-pkgs = %w(
-  docker-engine
-  aufs-tools
-  build-essential
-  ethtool
-  iputils-arping
-  libpcap-dev
-  git
-  mercurial
-  bc
-)
-
 Vagrant.configure(VAGRANTFILE_API_VERSION) do |config|

-  config.vm.box = "ubuntu/ubuntu-15.04-amd64"
-  config.vm.box_url = "https://cloud-images.ubuntu.com/vagrant/vivid/current/vivid-server-cloudimg-amd64-vagrant-disk1.box"
+  config.vm.box = VAGRANT_IMAGE

  config.vm.network "private_network", ip: vm_ip
  config.vm.provider :virtualbox do |vb|
-    vb.customize ["modifyvm", :id, "--natdnshostresolver1", "off"]
-    vb.customize ["modifyvm", :id, "--natdnsproxy1", "off"]
+    vb.memory = 2048
+    configure_nat_dns(vb)
  end

+  # Disable default Vagrant shared folder, which we don't need:
  config.vm.synced_folder ".", "/vagrant", disabled: true
+  # Keep Weave Net sources' in sync:
  config.vm.synced_folder ".", "/home/vagrant/src/github.com/weaveworks/weave"
+  # Create a convenience symlink to $HOME/src/github.com/weaveworks/weave
+  config.vm.provision :shell, :inline => 'ln -sf ~vagrant/src/github.com/weaveworks/weave ~vagrant/'

-  install_build_deps config.vm, pkgs
-  install_go_toochain config.vm
-  tweak_user_env config.vm
-  tweak_docker_daemon config.vm
+  # Set SSH keys up to be able to run smoke tests straightaway:
+  config.vm.provision "file", source: "~/.vagrant.d/insecure_private_key", destination: "/home/vagrant/src/github.com/weaveworks/weave/test/insecure_private_key"
+  # Grant permissions on sources:
+  config.vm.provision :shell, :inline => 'sudo chown -R vagrant:vagrant ~vagrant/src', :privileged => false
  cleanup config.vm

+  config.vm.provision 'ansible' do |ansible|
+    ansible.playbook = 'tools/config_management/setup_weave-net_dev.yml'
+    ansible.extra_vars = {
+      go_version: GO_VERSION
+    }.merge(ansibleize(get_dependencies_version_from_file_and_env()))
+  end
 end

 begin
@@ -0,0 +1,22 @@
+package api
+
+import (
+	"fmt"
+	"net/url"
+)
+
+func (client *Client) DNSDomain() (string, error) {
+	return client.httpVerb("GET", "/domain", nil)
+}
+
+func (client *Client) RegisterWithDNS(ID string, fqdn string, ip string) error {
+	data := url.Values{}
+	data.Add("fqdn", fqdn)
+	_, err := client.httpVerb("PUT", fmt.Sprintf("/name/%s/%s", ID, ip), data)
+	return err
+}
+
+func (client *Client) DeregisterWithDNS(ID string, ip string) error {
+	_, err := client.httpVerb("DELETE", fmt.Sprintf("/name/%s/%s", ID, ip), nil)
+	return err
+}
@@ -0,0 +1,81 @@
+package api
+
+import (
+	"fmt"
+	"net"
+	"net/url"
+)
+
+// Special token used in place of a container identifier when:
+// - the caller does not know the container ID, or
+// - the caller is unsure whether IPAM was previously told the container ID.
+//
+// While addresses are typically stored under their container IDs, when this
+// token is used, the address will be stored under its own string, something
+// which should be kept in mind if the entry needs to be removed at any point.
+const NoContainerID string = "_"
+
+func (client *Client) ipamOp(ID string, op string, values url.Values) (*net.IPNet, error) {
+	ip, err := client.httpVerb(op, fmt.Sprintf("/ip/%s", ID), values)
+	if err != nil {
+		return nil, err
+	}
+	return parseIP(ip)
+}
+
+func ipamValues(checkAlive bool) url.Values {
+	values := make(url.Values)
+	if checkAlive {
+		values.Set("check-alive", "true")
+	}
+	return values
+}
+
+// returns an IP for the ID given, allocating a fresh one if necessary
+func (client *Client) AllocateIP(ID string, checkAlive bool) (*net.IPNet, error) {
+	return client.ipamOp(ID, "POST", ipamValues(checkAlive))
+}
+
+func (client *Client) AllocateIPInSubnet(ID string, subnet *net.IPNet, checkAlive bool) (*net.IPNet, error) {
+	ip, err := client.httpVerb("POST", fmt.Sprintf("/ip/%s/%s", ID, subnet), ipamValues(checkAlive))
+	if err != nil {
+		return nil, err
+	}
+	return parseIP(ip)
+}
+
+// returns an IP for the ID given, or nil if one has not been
+// allocated
+func (client *Client) LookupIP(ID string) (*net.IPNet, error) {
+	return client.ipamOp(ID, "GET", nil)
+}
+
+// Claim a specific IP on behalf of the ID
+func (client *Client) ClaimIP(ID string, cidr *net.IPNet, checkAlive bool) error {
+	_, err := client.httpVerb("PUT", fmt.Sprintf("/ip/%s/%s", ID, cidr), ipamValues(checkAlive))
+	return err
+}
+
+// release all IPs owned by an ID
+func (client *Client) ReleaseIPsFor(ID string) error {
+	_, err := client.httpVerb("DELETE", fmt.Sprintf("/ip/%s", ID), nil)
+	return err
+}
+
+func (client *Client) DefaultSubnet() (*net.IPNet, error) {
+	cidr, err := client.httpVerb("GET", fmt.Sprintf("/ipinfo/defaultsubnet"), nil)
+	if err != nil {
+		return nil, err
+	}
+	_, ipnet, err := net.ParseCIDR(cidr)
+	return ipnet, err
+}
+
+func parseIP(body string) (*net.IPNet, error) {
+	ip, ipnet, err := net.ParseCIDR(string(body))
+	if err != nil {
+		return nil, err
+	}
+	ipnet.IP = ip
+	return ipnet, nil
+}
@@ -0,0 +1,99 @@
+package api
+
+import (
+	"errors"
+	"fmt"
+	"io"
+	"io/ioutil"
+	"net/http"
+	"net/url"
+	"strings"
+	"time"
+)
+
+const (
+	WeaveHTTPHost = "127.0.0.1"
+	WeaveHTTPPort = 6784
+)
+
+type Client struct {
+	baseURL string
+	log     Logger
+}
+
+func (client *Client) httpVerb(verb string, url string, values url.Values) (string, error) {
+	url = client.baseURL + url
+	client.log.Debugf("weave %s to %s with %v", verb, url, values)
+	var body io.Reader
+	if values != nil {
+		body = strings.NewReader(values.Encode())
+	}
+	req, err := http.NewRequest(verb, url, body)
+	if err != nil {
+		return "", err
+	}
+	if values != nil {
+		req.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	}
+	resp, err := http.DefaultClient.Do(req)
+	if err != nil {
+		return "", err
+	}
+	defer resp.Body.Close()
+	rbody, err := ioutil.ReadAll(resp.Body)
+	if err != nil {
+		return "", err
+	}
+	if resp.StatusCode == http.StatusOK || resp.StatusCode == http.StatusNoContent {
+		return string(rbody), nil
+	}
+	return "", errors.New(resp.Status + ": " + string(rbody))
+}
+
+func NewClient(addr string, log Logger) *Client {
+	host := WeaveHTTPHost
+	port := fmt.Sprintf("%d", WeaveHTTPPort)
+	switch parts := strings.Split(addr, ":"); len(parts) {
+	case 0:
+	case 1:
+		if parts[0] != "" {
+			host = parts[0]
+		}
+	case 2:
+		if parts[0] != "" {
+			host = parts[0]
+		}
+		if parts[1] != "" {
+			port = parts[1]
+		}
+	default:
+		return &Client{baseURL: fmt.Sprintf("http://%s", addr), log: log}
+	}
+	return &Client{baseURL: fmt.Sprintf("http://%s:%s", host, port), log: log}
+}
+
+func (client *Client) Connect(remote string) error {
+	_, err := client.httpVerb("POST", "/connect", url.Values{"peer": {remote}})
+	return err
+}
+
+// IsReady returns true if the API server is up and running
+// (note it returns StatusServiceUnavailable until all parts are ready,
+// but callers of this function are expected to be ok with a partial service)
+func (client *Client) IsReady() bool {
+	resp, err := http.Get(client.baseURL + "/status")
+	return err == nil &&
+		(resp.StatusCode == http.StatusOK || resp.StatusCode == http.StatusServiceUnavailable)
+}
+
+// WaitAPIServer waits until the API server is ready to serve.
+func (client *Client) WaitAPIServer(n int) {
+	for i := 0; i < n && !client.IsReady(); i++ {
+		time.Sleep(2 * time.Second)
+	}
+}
+
+type Logger interface {
+	Infof(string, ...interface{})
+	Debugf(string, ...interface{})
+}
@@ -0,0 +1,17 @@
+#!/bin/bash
+
+set -e
+
+source "$STATE"
+
+if [ -n "$TEST_AND_PUBLISH" ]; then
+    cd "$SRCDIR"
+    git submodule update --init
+    cd "$SRCDIR/build"
+    ../tools/rebuild-image weaveworks/weavebuild . Dockerfile build.sh
+    touch "$SRCDIR/.build.uptodate"
+    sudo chown -R ubuntu /usr/local/go
+    cd "$SRCDIR"
+    make testrunner
+    make COVERAGE=true RM= exes all
+fi
@@ -0,0 +1,36 @@
+#!/bin/bash
+
+set -e
+
+echo "TEST_AND_PUBLISH=1" >"$STATE"
+
+# Avoid doing a full build if we can.  Note if the PR doesn't exist
+# when the build is kicked off, we can't run this check so we do a
+# full build.  Subsequent pushes to the same branch will have the desired behaviour
+if [ -n "$CI_PULL_REQUEST" -a "$CI_PULL_REQUEST" = "$CI_PULL_REQUESTS" ]; then
+    if PR=$(echo "$CI_PULL_REQUEST" | grep -oP '(?<=^https://github.com/weaveworks/weave/pull/)[0-9]+$'); then
+        echo "GH_PR=$PR" >>"$STATE"
+        if FILES=$(curl -s "https://api.github.com/repos/weaveworks/weave/pulls/$PR/files" | jq -r '.[]|.filename'); then
+            echo "PR touches [$FILES]"
+            if ! echo "$FILES" | grep -v ^site/; then
+                echo "PR contains site changes only - skipping tests"
+                echo "TEST_AND_PUBLISH=" >>"$STATE"
+            else
+                echo "PR contains non-site changes - running tests"
+            fi
+        else
+            echo "Unable to get files for PR $PR" >&2
+        fi
+    else
+        echo "Unable to extract PR number from $CI_PULL_REQUEST" >&2
+    fi
+fi
+
+source "$STATE"
+
+mkdir -p "$(dirname "$SRCDIR")" && cp -r "$(pwd)/" "$SRCDIR"
+
+if [ -n "$TEST_AND_PUBLISH" ]; then
+    cd "$SRCDIR"
+    git submodule update --init
+fi
@@ -0,0 +1,13 @@
+#!/bin/bash
+
+set -e
+
+source "$STATE"
+
+if [ -n "$TEST_AND_PUBLISH" ] ; then
+    docker login -e $DOCKER_EMAIL -u $DOCKER_USER -p $DOCKER_PASS
+    cd $SRCDIR
+    make clean-bin
+    # CircleCI doesn't let us install the qemu module to do ARM docker builds
+    make publish ALL_ARCHITECTURES=amd64 UPDATE_LATEST=latest-only
+fi
@@ -0,0 +1,13 @@
+#!/bin/bash
+
+set -e
+
+source "$STATE"
+
+if [ -n "$TEST_AND_PUBLISH" ] ; then
+    if [ "$CIRCLE_NODE_INDEX" = "0" ] ; then
+        go get github.com/mattn/goveralls
+        goveralls -repotoken $COVERALLS_REPO_TOKEN -coverprofile=$SRCDIR/test/profile.cov -service=circleci
+        cd $SRCDIR/test; cp coverage.* $CIRCLE_ARTIFACTS
+    fi
+fi
@@ -0,0 +1,30 @@
+#!/bin/bash
+
+set -e
+
+# Signal failures in lock file, in order to fail fast:
+function signal_failure() {
+    echo "KO" >"$TEST_VMS_READY_LOCK_FILE"
+    exit 1
+}
+trap signal_failure ERR
+
+source "$STATE"
+source "$(dirname "$0")/wait_for.sh"
+
+if [ -n "$TEST_AND_PUBLISH" ]; then
+    [ -n "$SECRET_KEY" ] || {
+        echo "Cannot run smoke tests: no secret key"
+        exit 1
+    }
+
+    # Provisioning from image typically take 90 seconds, 
+    # but if we are creating a GCP image, it can take more than 10 minutes:
+    wait_for 900 "$TEST_VMS_PROV_AND_CONF_LOCK_FILE"
+
+    # Upload built binaries to testing VMs:
+    cd "$SRCDIR/test" # Ensures we generate Terraform state files in the right folder, for later use by integration tests.
+    ./run-integration-tests.sh setup >>"$TEST_VMS_SETUP_OUTPUT_FILE" 2>&1
+    echo "OK" >"$TEST_VMS_READY_LOCK_FILE"
+    echo "Test VMs now ready. $(date)." >>"$TEST_VMS_SETUP_OUTPUT_FILE"
+fi
@@ -0,0 +1,23 @@
+#!/bin/bash
+
+set -e
+
+source "$STATE"
+source "$(dirname "$0")/wait_for.sh"
+
+if [ -n "$TEST_AND_PUBLISH" ]; then
+    [ -n "$SECRET_KEY" ] || {
+        echo "Cannot run smoke tests: no secret key"
+        exit 1
+    }
+
+    # Provisioning from image typically take 90 seconds, 
+    # but if we are creating a GCP image, it can take more than 10 minutes:
+    wait_for 900 "$TEST_VMS_READY_LOCK_FILE"
+
+    # Run integration tests:
+    export COVERAGE=true
+    export WEAVE_NET_SANITY_CHECKS_FILES="$CIRCLE_ARTIFACTS/weave_net_sanity_check_*.log"
+    cd "$SRCDIR/test" # Ensures we generate code coverage files in the right folder, and use the Terraform state of previously provisionned VMs.
+    ./run-integration-tests.sh test
+fi
@@ -0,0 +1,10 @@
+#!/bin/bash
+
+set -e
+
+source "$STATE"
+
+if [ -n "$TEST_AND_PUBLISH" ] ; then
+    cd $SRCDIR/test
+    test "$CIRCLE_NODE_INDEX" != "0" || (cd $SRCDIR/test; ./gen_coverage_reports.sh)
+fi
@@ -0,0 +1,11 @@
+#!/bin/bash
+
+set -e
+
+source "$STATE"
+
+if [ -n "$TEST_AND_PUBLISH" ] ; then
+    cd $SRCDIR
+    make DEBUG=1 lint
+    COVERDIR=test/coverage make RM= tests
+fi
@@ -14,7 +14,8 @@ set -e
 START=${FIRST_WEAVE:-1}
 COUNT=${NUM_WEAVES:-28}
 FINISH=$((START+COUNT-1))
-PORT=6783
+
+SUBNET=192.168.7

 weavedir=$(dirname $0)/..

@@ -26,29 +27,42 @@ random_mac() {
    od -txC -An -N6 /dev/urandom | ( read a b c d e f && printf "%02x:$b:$c:$d:$e:$f" $((0x$a & ~1 | 2)) )
 }

+weave() {
+    i=$1
+    shift
+    WEAVE_CONTAINER_NAME=weave$i WEAVE_HTTP_ADDR=$SUBNET.$i:6784 $weavedir/weave "$@"
+}
+
 case "$1" in
    launch)
        shift 1
+        echo $(date -R) launching
+        ip link show multiweave >/dev/null 2>&1 || sudo ip link add name multiweave type bridge
        for i in $(seq $START $FINISH); do
-            WEAVE_CONTAINER_NAME=weave$i WEAVE_PORT=$((5000+i)) $weavedir/weave launch-router --iface '' --name $(random_mac) --nickname weave$i --port $PORT --no-dns "$@"
+            IP=$SUBNET.$i
+            sudo ip addr add $IP/24 dev multiweave
+            weave $i launch --datapath '' --iface '' --no-dns --host $IP \
+                --name $(random_mac) --nickname weave$i "$@"
        done
-        if [ $START -ne 1 ] ; then
-            status=$(docker inspect --format='{{.State.Running}} {{.NetworkSettings.IPAddress}}' weave$((START-1)))
-            PREV="${status#true }"
-        fi
+        echo $(date -R) connecting
+        [ $START -gt 1 ] || START=2
        for i in $(seq $START $FINISH); do
-            if [ -n "$PREV" ] ; then
-                echo connecting weave$i to $PREV
-                WEAVE_CONTAINER_NAME=weave$i $weavedir/weave connect $PREV
-            fi
-            status=$(docker inspect --format='{{.State.Running}} {{.NetworkSettings.IPAddress}}' weave$i)
-            PREV="${status#true }"
+            TO=$SUBNET.$((i-1))
+            echo connecting weave$i to $TO
+            weave $i connect $TO
        done
+        echo $(date -R) done
        ;;
    stop)
-        for i in $(seq $START $FINISH); do
+        for i in $(seq $FINISH -1 $START); do
+            IP=$SUBNET.$i
            echo Stopping weave$i
-            WEAVE_CONTAINER_NAME=weave$i $weavedir/weave stop-router
+            weave $i stop
+            sudo ip addr del $IP/24 dev multiweave || true
        done
+        [ -z "$(ip -4 -o addr show multiweave)" ] && sudo ip link del multiweave
+        ;;
+    *)
+            weave "$@"
        ;;
 esac
@@ -0,0 +1,41 @@
+#!/bin/bash
+
+set -e
+
+# Signal failures in lock file, in order to fail fast:
+function signal_failure() {
+    echo "KO" >"$TEST_VMS_PROV_AND_CONF_LOCK_FILE"
+    exit 1
+}
+trap signal_failure ERR
+
+source "$STATE"
+
+function install_terraform() {
+    curl -fsS https://releases.hashicorp.com/terraform/0.8.5/terraform_0.8.5_linux_amd64.zip | gunzip >terraform && chmod +x terraform && sudo mv terraform /usr/bin
+}
+
+function install_ansible() {
+    sudo apt-get update || true
+    sudo apt-get install -qq -y python-pip python-dev libffi-dev libssl-dev && pip install --user -U cffi && pip install --user ansible
+    export PATH="$PATH:$HOME/.local/bin"
+}
+
+if [ -n "$TEST_AND_PUBLISH" ]; then
+    [ -n "$SECRET_KEY" ] || {
+        echo "Cannot run smoke tests: no secret key"
+        exit 1
+    }
+
+    install_terraform >>"$TEST_VMS_SETUP_OUTPUT_FILE" 2>&1
+    install_ansible >>"$TEST_VMS_SETUP_OUTPUT_FILE" 2>&1
+
+    # Only attempt to create GCP image in first container, wait for it to be created otherwise:
+    [ "$CIRCLE_NODE_INDEX" != "0" ] && export CREATE_IMAGE=0
+
+    # Provision and configure testing VMs:
+    cd "$SRCDIR/test" # Ensures we generate Terraform state files in the right folder, for later use by integration tests.
+    ./run-integration-tests.sh configure >>"$TEST_VMS_SETUP_OUTPUT_FILE" 2>&1
+    echo "OK" >"$TEST_VMS_PROV_AND_CONF_LOCK_FILE"
+    echo "Test VMs now provisioned and configured. $(date)." >>"$TEST_VMS_SETUP_OUTPUT_FILE"
+fi
@@ -89,6 +89,8 @@ build() {
    ## Inject the version numbers and build the distributables
    ## (library versions?)
    sed -i "/SCRIPT_VERSION=/ c\SCRIPT_VERSION=\"$VERSION\"" ./weave
+    sed -i -e "s/:latest/:$VERSION/" -e "/imagePullPolicy: Always/d" ./prog/weave-kube/weave-daemonset.yaml
+    sed -i -e "s/:latest/:$VERSION/" -e "/imagePullPolicy: Always/d" ./prog/weave-kube/weave-daemonset-k8s-1.6.yaml
    make SUDO=$SUDO WEAVE_VERSION=$VERSION DOCKERHUB_USER=$DOCKERHUB_USER

    if make tests; then
@@ -157,6 +159,20 @@ draft() {
        --name "weave" \
        --file "./weave"

+    github-release upload \
+        --user $GITHUB_USER \
+        --repo weave \
+        --tag $LATEST_TAG \
+        --name "weave-daemonset.yaml" \
+        --file "./prog/weave-kube/weave-daemonset.yaml"
+
+    github-release upload \
+        --user $GITHUB_USER \
+        --repo weave \
+        --tag $LATEST_TAG \
+        --name "weave-daemonset-k8s-1.6.yaml" \
+        --file "./prog/weave-kube/weave-daemonset-k8s-1.6.yaml"
+
    echo "** Draft $TYPE $RELEASE_NAME $VERSION created at"
    echo -e "\thttps://github.com/$GITHUB_USER/weave/releases/$LATEST_TAG"
 }
@@ -165,14 +181,18 @@ publish() {
    setup
    cd $PWD/$RELEASE_DIR

-    UPDATE_LATEST=false
+    IS_MAINLINE=false
    if [ "$RELEASE_TYPE" = 'MAINLINE' ] ; then
-        UPDATE_LATEST=true
+        IS_MAINLINE=true
    fi

    if [ "$RELEASE_TYPE" = 'PRERELEASE' ] ; then
+        echo "== Building and pushing plugin"
+        make UPDATE_LATEST=$IS_MAINLINE SUDO=$SUDO WEAVE_VERSION=$VERSION DOCKERHUB_USER=$DOCKERHUB_USER plugin_publish
+        echo "** Plugin built and pushed"
+
        echo "== Tagging and pushing images on docker hub as user $DOCKERHUB_USER"
-        make UPDATE_LATEST=$UPDATE_LATEST SUDO=$SUDO WEAVE_VERSION=$VERSION DOCKERHUB_USER=$DOCKERHUB_USER publish
+        make UPDATE_LATEST=$IS_MAINLINE SUDO=$SUDO WEAVE_VERSION=$VERSION DOCKERHUB_USER=$DOCKERHUB_USER publish
        echo "** Docker images tagged and pushed"

        echo "== Publishing pre-release on GitHub"
@@ -203,8 +223,12 @@ publish() {
        fi
        echo '** Sanity checks OK for publishing tag' $LATEST_TAG as $DOCKERHUB_USER/weave:$VERSION

+        echo "== Building and pushing plugin"
+        make UPDATE_LATEST=$IS_MAINLINE SUDO=$SUDO WEAVE_VERSION=$VERSION DOCKERHUB_USER=$DOCKERHUB_USER plugin_publish
+        echo "** Plugin built and pushed"
+
        echo "== Tagging and pushing images on docker hub as user $DOCKERHUB_USER"
-        make UPDATE_LATEST=$UPDATE_LATEST SUDO=$SUDO WEAVE_VERSION=$VERSION DOCKERHUB_USER=$DOCKERHUB_USER publish
+        make PUBLISH_WEAVEDB=$IS_MAINLINE UPDATE_LATEST=$IS_MAINLINE SUDO=$SUDO WEAVE_VERSION=$VERSION DOCKERHUB_USER=$DOCKERHUB_USER publish
        echo "** Docker images tagged and pushed"

        echo "== Publishing release on GitHub"
@@ -236,6 +260,20 @@ publish() {
            --name "weave" \
            --file "./weave"

+        github-release upload \
+            --user $GITHUB_USER \
+            --repo weave \
+            --tag latest_release \
+            --name "weave-daemonset.yaml" \
+            --file "./prog/weave-kube/weave-daemonset.yaml"
+
+        github-release upload \
+            --user $GITHUB_USER \
+            --repo weave \
+            --tag latest_release \
+            --name "weave-daemonset-k8s-1.6.yaml" \
+            --file "./prog/weave-kube/weave-daemonset-k8s-1.6.yaml"
+
        echo "** Release $RELEASE_NAME $VERSION published at"
        echo -e "\thttps://github.com/$GITHUB_USER/weave/releases/$LATEST_TAG"
        echo -e "\thttps://github.com/$GITHUB_USER/weave/releases/latest_release"
@@ -1,57 +0,0 @@
-#!/bin/sh
-
-## This is the shape of unencrypted `do-setup-ci-secrets`
-## 
-## cat > /tmp/boto_gsutil_cert.pem << EOF
-## -----BEGIN PRIVATE KEY-----
-## <...>
-## -----END PRIVATE KEY-----
-## EOF
-## 
-## cat > ~/.boto << EOF
-## [Credentials]
-## gs_service_client_id = <...>
-## gs_service_key_file = /tmp/boto_gsutil_cert.pem
-## 
-## [Boto]
-## https_validate_certificates = True
-## 
-## [GSUtil]
-## content_language = en
-## default_api_version = 2
-## default_project_id = <...>
-## EOF
-
-openssl base64 -d << EOF \
-  | openssl aes-256-cbc -d -K $1 -iv $2 \
-    -out bin/do-setup-ci-secrets
-mOYNzNsAdwva0LbB2YFMY9VgL0SITFu1bPrbOujyLtOrBNjmOYbzR5d/sG+ARMFr
-7NZ5OKcvMbItllz0n7TtNkzkCOMztMC870Wcx3nGJzbBephdcYwos4tBlp5bihTU
-AXgxLUiDT1broYAFKnbE1aZC/XyNxuLTN+nXJWIWeijaXTgvhkEEm3tab+z7ioCb
-tlg71W1BrUuocctAJAuVwH/uTZAaXaJGJj1s2S7F2DrFnG6sFlrp8rnCucPLWKnC
-4IwJFC5saR5gxK6wECffwmwbpeXVOoQ58OgzxoZTEyAr9nhVbaBt5UZz3tfHlx8W
-Rn5jHehACw5kg5eMtGPqMEksJ63LX4h3mfSTLJXJHEmOrzCRpFrigVrsbA408U2u
-8jJTHJmKrGfwBh5ZFzBg6NuV4VplOp+ZquJLgegv2WH/iu/ciopXqB2YDCdx5ZBi
-nHV2Puth7kjS4EMCKUGfeqpesjL9Djb8cygzv/J3ZBmLjNH9dqlSXWy2DcFZA1Iy
-pFTTHkVU3w5sNz3nYOuT2UPQvULBW0FUndDKX6olyVa0mlxPwGBP9N3yGg918/rw
-7rNV/r54lA334F5QMhDrVxar6sl9wfLA+Z+8qemBbNFlLog1uQGhuwWK2nFu1tRM
-9uyvQIq0oq5NHKpogWc7NEuk7CPe1JyfRIcbMVj8+ilmKjUPB2HPC43IDmKt9/56
-rY3IXAcxZ/ClGfGpeNHXqop9T/WhfaowponOZFcdGSUddG0OswqkTE4KIv71mgTF
-jXooMkTQBOG+gBnCe+8bCRdRCG3anxHX74za4dGD58cOJtl6DYWnjWuR91NWnxdT
-1Zr3iUe41txg9hCNY9LHo2QHGLPxe6pmNnSClQ1mcFhuY6NF8RmCX3HAGKJQkxJb
-vpLGbXTXjAIFgIYgLNgKP2RLvCif/LmSWgkvE3ZU9GXtCUbm6xgJodwOqXiI9uuO
-4H4+/vpGHnSFENiclkgMFrLG4yiMu8ArgmnCBzg9E2AoOX5zCrh/2ZRTU8Dk4mt7
-jUd5SKQLsjB9QtB+y3tRO7zY9RDvR6iBOPRsjx6t8EJMCfPb06KchfMJw+VGxvMM
-48jtmepREFnhd/8XDRGnw5bznjRuwohGik8BToltjIviMkkpifKWaLwlaOiDXaHh
-VBFNXnUKvOBwRMU+PZqv4+DP2qff1gSXnGKqA5GIURYrK9P7TqHHNTnmjBdfJLzc
-0x+W1DTiYhLT2vrLrkEpZ99o6qrg+lwvjat6IsJF8xoQEo4pLD/ZRPJ0WzoDyctT
-oWZ5rMsN1TzTvJrBrTAWIJyhIzZcuNwChMrU6qogMr0fd6TSxWhWn0sOJ5y8AaTR
-ej4JHx0qroS/9JPmgFYU61pX3c0ypdef/sSHWG23mcFzLTgwe2d4ty9VakypZy3T
-sXCl6EHVVXlUFWhY4zQCjdZpYJv/rtoKL1BL4fPE2NV5WTjCL2GNvaO5SfHR7pA0
-yQJR/iU+cS2n8UXsWNjyV+/qEn6+QXuyVHGLbxyH8EeqlliAbyHsc2rzVI8Xctzo
-W5DuIqxwHAkNNurhmx27ktIIQ6HhaCp028SWwvLttkbBBgF4v1mmd4u/2bKbKjUd
-dOaetaIUx1hU9OAeaePvVkeTDcc3mlFtPARCS9Fqj16hgJQ43onZyC0H1K2WAxmU
-GsmzjCHlTNquzFTuuQihYGZKfsWIh9z6HEOkvNJ3VEQ=
-EOF
-
-exec sh bin/do-setup-ci-secrets
@@ -1,115 +0,0 @@
-#!/bin/bash
-
-set -eu
-
-# openssl enc -in do-setup-circleci-secrets.orig -out setup-circleci-secrets.orig -e -aes256 -pass stdin
-# openssl base64 < setup-circleci-secrets.orig
-
-openssl base64 -d << EOF \
-  | openssl enc \
-    -out bin/do-setup-circleci-secrets \
-		-d -aes256 -pass pass:$1
-U2FsdGVkX1/3QHnDgpZn0srp+GiEQBTlag0CBdsw6Bp6iaIKy6Jy8gFshrsmvl3+
-c+7hn1nqsBVpzVH0S5QA3Ex5nx9rl7/dcGoogsySFcwMmBFcauT0UTMK1bvX0Hqx
-QrshYSndUkww72KA7EsSnjR1BBJWm3AhCDOTIcnO6nQhig1Fc+SBEj5FGS5xTQ5Z
-/8hDsKf54Ms49vOO6QeE5w5ndZ+k7IvVJfq9Evhb7htD20RntDhR86foAwMnLV7R
-Wh9qAqI3gnRq0GmbeRKaHmB7pcLq9f0OdScqM/Tn1D+q0556fqJjBziXpZiGbxVs
-DxEp8zDJ8yNDAi0Wb2NXjoA79SXJ8Q7iTieaST4y2OmXQpY+Vw8RVDXxi6ERyVnH
-41OeMzyLm0qvCoMl32pe1sSG5DRCUiAm7wOtDTVvNDRqmGJYUgwwexyayecJI1wQ
-LlhYLJavUvw3UGDZOld8BaMQhfOKBVCdFrjmA4HuVxmOf3029qkgzOCSZiQVNqs+
-fvM6J05zBuk+GPy8YInvRkzfxum/x00vmUyGjMs/lHDFfrmYDFK3ux41EoVqvWLG
-S5uYxdZJFdR3Xsrs+sMr51XNBQGMmz6KoEEzFYyHs5fZgtdpRdPAuQS1m6FqgZyi
-wSz6+EKsALtmGorEp+CF8JN5YzKJCrKzNkg0fILBqsqmJZGP64lmX69950lRLNGM
-/l3KVR2zH31b4AnnzTcvldVSCKDXjaMw3JA2Ce/MK6XhOWOcpEv4oDt/62lbhUKP
-QUQTPbfpQRvVfBcoIrfAm3ghi/CZc2TDVEuD7DhasLTZD3r7LoK261BWC9O5Wubi
-vlXuFIF+uyrpTDf6EJKE/+XywtArppffMP6RkHu1Iqnd8JtsqPaIZsmMmZbjgb6j
-BX4ZLb5U9R7jYoSgIke4+u6geFcrMQ+Lny4r7MxXaS0Y1UJ6Jo1bwpYw1TBIobZc
-eQ30+xhbSeRC9vEw1NXI6qM2Ai2gIHtAZvEiBkBBXTG5Ld7N/mBAmDIJEzVRxh6F
-jRkB3dzESF6eeTBm61OQepwV3aW6lslMaf26drPGv2W2q+K/WOpQOejfggtHUgVs
-9SWqtn7hHYSpIASe1Ihi14HAWv3kPhQien0zY1DLKJWf4ilbRaICn1QSJjNuKwtl
-A2ZwEw7FFddFcsF4okHPkxEoCre3shq0DEw9UA1gnwEaEF1Q+i9cvLtQg56zK3+R
-MfTzFUv2K1z3j1s5tC6Z6LzL8j682EFOll7MkihqiIWh+po1zcNs6Ln0SkjB3v9T
-OFRYVJis9GqXYO9PUW1mI69udR4nXbKlBl5tWNyAxn9hufJHLSAmbGMwActKLRM5
-kROapKrN1Ra/sHUmhoSSDUK9AXeYKkqPXVMbhly9xC1Kya4hFaFokjRQIMLPmXdQ
-lP46RL3HOq0v1BkXg05poSwDm+BtitZkcIhOzcmeucUqn1whsz39y3VoOhJqQu2j
-0dUcj19Fue1CDZfwzACb9mLZr643h0yCIHQaVT7is/yiN2vEt0S0zynvOA1D+Xgy
-4bFdjqed9DgBShLwF/lh4/hur5iZPpVr9A8rt3Vq0k5A1tdjsZpXP3qv5jLc7lJv
-cAQxdVLEuv8WPwwTV/EQsoH7lpoEdjktw/gu0v5PkUgR/WBxEqJq4liTekViQ+8X
-BAmdm/FGWeIxSSEm0a8I+muiBpIuKjuQoJit1f0k+lLUohoVDqpDDfeYHh1TxXHE
-xSKAKNmFSc2XOr5Zg2mOInxeH7WoxH6iylXKADBgEgAQ4C55DoAPRjFNtGGz2iJl
-WkzhQOfgsDss1VxpPZ4SWDQO+xxuuEnYcCYdJ6Tyi5hAwZ0+RZmXx0g9BCK+5qzc
-Ffei066kCItBcayekwCs0i86r8JvDtsMUUkzEKDQTSWHpP8ypIZoz3P+Skdqcs6p
-IcZHy2fODIsClctRAsHOjZ9BnWSY2rqf1lysqvRfOkNTViTOUVx2hf+LuL+8rT6g
-K4GvS39Q+c9FYSMjCNxv1m37NfHJv2W7AlqVV95gMBEFB97M3388T7PfvmSLsyPv
-qAKDE7genTOycQvEX9oeEACxPo08ai2Mgs/YXCig0g0btZoScoKPelQKtD0HLpZ+
-j7ruaKFy4/vySDMostB3AKSkCE/SkGu/W9oI37n1guSYC/9yJjtTcwZULXAClfwt
-LyHvLAw4+7BdGJqDe9Tf8z4zqVLax+BcBHkW7pbP+pvxC6fhIA4PO8k5HwwAVlRw
-68lG6e3z9+Vst4zrFLewHJvkDRSOTSecs9h9SPOjReTK0ovlDzMHVakfUvqpk9mv
-nCj8SWaMPF70z138AuRuKN7X4exNXPBFnOcseKp/NpT1DyvgoHm/oIO65mq9vvo+
-42uSbyyegYiTdVK4SaXBc2hixkhaTteqdEy6GxfWqKCvHtYuRD2LyUSjZn/0Y6CG
-3wZapylKL1sxfH+b1gogv2wAA8KtZAfiTlMISuymwQr5RbOFma7UVMVecCa7XrNG
-fZwO8R+6++ru0efDzeC/gtLOItjn+yVfsbk2gfu0FikHIV1sogOjnsv3dl6IpH/s
-w6WbuK/yZhyrrmhIIueHF30ES512Peogf/Gxp8U/ZAmqBa7p4qsv93XWDR52IPYz
-yU1MtboLD02J39daBNrYZuO4Yfo/xAvU7Ijg2KIR9HS0bSWNuSFAZaZXlJ2bSag3
-8l0dZMx2fZa0FALxWvY2+nd1tWcH+Wo99RSxX+4TH3guCpe1x/58yKOKL4d1xRpu
-02nGaqerV9bHU6WjCUYTrrZurTp6yqivpSFCALDMhAU5Ihd/Qjz69zg4uZ7iqbNd
-WqIYJ3SAOei2tSfoOJbdh8mohotnMFFbupck3Fv+lXAXYDp0V/sSuK32MSeJxan1
-q3t0eenAh6zoBP0Tj53N8B9YmrgZ1i1AljQqYsAdb8N5Z3vecuzQa946U8H4axEc
-BL5fgmqSfsbj1EjJVUrw3cE9hgqeqAYBBvZsyW5zSAsLq/RJKYtBxDGADqr0TOOu
-TjVqfPObb1u9EAQj15ALBhhPORa2gBAK5gnhJWW5deNgA0UOYoWARFYfATuQvr1T
-9xTVbqzZ0VJR4n7X5JyxJo2tZ44QYVzA0fqX0lt+uRkarrkbZMgplKp4MJTFCp2N
-tLciWS0YSHPbKTm7svw5mROazUvEoj1N3SYrJT+Ai9UgSL1cA4FmQ5Jsi+GOeQbH
-7E4pzjeST4b1aATY9ZZSy4TzK+UNYBcCVqY1dN5Xnk0e5yPMnrkkcYI41s/6oOaW
-lQkp/yrHkvpzQWZa7vF2yWg54j3eX8YWx/TsLvmuJmSQqiIHfu9TXcAIPDispCpF
-gp4ZbYPZ8RDQLw3ROc+bZ9E0J8e+FHwgXEELsMLW9n4d3PTj9pLFej1vmBKa03o3
-g+JViNxb1gUPNmC6ADk4o7TQzBrjV2pmmuCza16kOmq6oAbJPOeCLtVWA5HJTX8d
-AuOzBLO37tWP457dNqoNDaCMik3LCy8cslZjAz02D6VG/QGOAXP9x8bJ3zg9N7vW
-EOn1BMtIeMI4yIxpC6HgotNWDNbUBWremAFWYQRgYQqo9L1SdW4hz6X8FqdlBFvK
-ml7sdUwjEy2tNW+aiZI4cgAlI/VvMWYr5kPfhvaUu1wGYeS1TD1B+DPcjcXpOGWo
-vB1f/uvGc61zZWD0ph9f3S9OnvEko5qV3VhxOnxYUZTrqlM7LVgy/sHrSaPdR7Eo
-nb4QGdi5wEGMxVgQlDGwOmEmS5nCaA0RcfHbkQV4SebRleocLB9qcSwj5JzPICAB
-sn1qeSv45LnGdTzUnCQErw1EzgrK1FXOknujjZ3GX5dM7a6Gag0E6cAxoQeC8Mh0
-aUWolyBo9aSXltn7Vnywoe4X3v96eVH6tUQtN3wt51LBlfSDTWOQDeXLFyW57oYL
-/ErbmJLaSHpANBcC9InI+rxNYePTTba7K8rdPDppmMEmdOXjaL+5aPay2Bay9K8T
-FUyZdoyrbzxFZYT9/YyVJV/ukVcT9ru3HpzLTg9Vgc/CyRfIvCZY/Gzh+pPUbRsR
-pHYDT09BFS+mjszTkcdbceweYXJvknXpajMiISrPUUkb9rwCTmLS4QR4y7I9zAgJ
-9U72mCZB9yhhzS17kyKAANihXvOFrMXCO7QTWZbtSkobMBS2SybDxYP3rM5b6824
-HQaY4MZW5qCndydfz4fa908XbPTDka9Nu3CZfI3fhGiiGDFWViotGEYE4i9BuGZJ
-vmtzII7zeIoHwM4t5fb5mYDd9H+Rkct1yATGEIFEx7cu1fj4Ht2177EzhdkF3Q6I
-Fccn43yGDW06nFvEjnjb+Lz4EvtGp6dDefUhNBohdzd4Rm2o+etKobm0lZ1xOX74
-2gOGBLfXS6PF+msNkWP76mqKE6hG0+R1iPrN06Arti+ZRBhNGJDp1pnOERl7IyB0
-hiLUqae7xVSMVRDeq4O04EUpMXh5ixs3RJXjthYILt7vX+6qskJC3rMuyThrsvMi
-Kk0wErDpzR5U7KFnM2fk4jY5by/0b/jHZ/PWFGnaFrXNOiwK/mV1RIJtg5FCrvMv
-xJN5Qo7o2ZYzUL+IBf5mP5O49FCnz3e++9r4ySY/K57ybhQft1l+cQPWS7xuvVAg
-wZGM9DPl9AhBus0I3V3XzfkcWWHkMynMRcfTnkBHrmKjMopIOggsq05JPalM9b0V
-NfwZngVvTFbQejbGjgQQQ+G7spav/icCIA8Iu8JPHaKZt2IX77L1xMcQOouqMrx8
-JCXnJ7zsuU+8W3GQGm70H2kcen4bf4beA8TSkFmbVsaqDNtGaG+ZopWO8ctZX3CP
-jFgIYgEGi137IWvs1KD5QNFHZI6l0JMzglmvcfnViRPDHLi+EQXAoL/98j/YzA9t
-a9L4xMO0oucovKkqYOS2cdFaQKVIUM9L5ibcbW7Tol4ugMo5ibIsTgjZoyFohG6t
-acg+h2wTTyxtrIapvOOBeAo9V/iQowTAPQf38LKQZ5tApksiCG1qvJerKALTMp1J
-yUHjTCPCPsaiXVVO0mfa9OPtzVvi/icLmlz7umk82mew13djedgMS3/ghq5aEHbg
-e7rtI5snFlm/GQnV6IaCDytb+jYxqQ0DZ9Mb0CHcMpnLyiBO6nzHfadpZYlrxrpx
-dPx9MVqfLavj1UVjKuVj4Mgec9J38IBBRGMJZ4MGhde2hrw/s2jVzG1BzaIzcvFa
-bWiJOkKKRC4BY9DBDYfae5c/obwzkxLmvkbWAVzoFXc+46K1IK6bbC2a4EerOAxc
-C7B3X8BucD8OtiPORwXgjctd9AJ7cXq0EEVGoZZ8jkILnklsK6RXsxrT4gSOGFpx
-cBU7TF4kHpH+o4KGQmjlFA2VYP1tXirjEep57LOHfL+zy+HiSXdsbexf6odkqLYA
-1RR0TzSkmqJYvP0GfHPMvp2xredx1kP934fhTIUcWvErrG3NPd/Dod4BzQye1nNK
-8Y8aFYF6fACcD4SFZ11ecK9DJmB+WXGlsxtO3w1LeEysny2lVkkH6HuMRIzPyXAM
-/sEpwh3hwPJosAh6vGM6JxuL7CXP8hC1oUvK91VR0hxeDcO/LgdcHtDnlNMqt1wC
-jvYt9Hg+3aY5GehBnQPRppnIMaGhiBrRC6HjBg8Umt6UTUr5PXNY2aia5wRcx4M6
-Tj0O0dAt8WWZ9ySgzOoq4SRoaAp6m4E7CpRNMgXH031OWEeHqtt8rsYqAcvcQLuW
-vIwts5hiXZOfgXyR6SBY3ANE1SjDY46na2G4oiG8FI7eAqYgehjHrXqCnGrzjefE
-V1eVcIr7DBYuM8wq7C9x+qdMir6RNV4MrVD+7rO/drivnd3MaqDHlG0LfMZo6GIK
-QZGIy0z1W5lOXm9x3RAz7/Ea11ynkZyPDDxBE5NRcV6JfukdjDrvsaQA/Qk2Wo7h
-PdUZjv17H4syMdap6tUDTc8Cohut6n1UxnDnhaAYEDGR3TgPqI5vEaJ+yoaM9itv
-b2I/K+0uK6GfWwca7PeTtYn92/5etO4czG99xEBFzw7tiEmBTA8KfseZIUe/i3fr
-9LO+rKEGalFB4mVTYyCKPYYx4hOIvjdCox/jEBTFyYUEhgGnZvArgw9XabvII7eM
-c1FmwnN2CM/MHUzM0Kw7cS/oaoRFcG/PqioO/krL2Bd+xThwSH4Zn+Y1UVO7aXNy
-90MOehHSc2nhN984Y2re3ffVsjPhC3KdWxr8jbho/Az65rUkYg64roiO2ZqdqRvf
-aYtX9fCe8N8SzgLTlWZt+l8sjY8HP2k9NoajphV1nG4iC/pX0ZJiVQ+VBRNaGhSl
-6IhM9gM1ZsI86cULJpOTAhNu1fZBv+Mq8IXYD8SqlNVwVtjvshtlzBN4InSGEKiO
-HM7coo7aDuu+Go5AEiIVMbR09jAa0VINsCVScfSbTFhlGRTrZr2DSO/s8WxbV+Mb
-EQAfZmyrLF71Z3MfIukg+ps5iuqFhZ+gJSXXSFwumpg=
-EOF
-
-exec sh bin/do-setup-circleci-secrets
@@ -0,0 +1,21 @@
+#!/bin/bash
+
+# shellcheck disable=SC2155
+wait_for() {
+    local timeout=$1
+    local lock_file=$2
+    for i in $(seq 1 "$timeout"); do
+        if [ -f "$lock_file" ]; then
+            local status="$(cat "$lock_file")"
+            if [ "$status" == "OK" ]; then
+                echo "[$i seconds]: $lock_file found and status: $status." && return
+            else
+                echo "[$i seconds]: $lock_file found and status: $status." && return 1
+            fi
+        fi
+        if ! ((i % 10)); then echo "[$i seconds]: Waiting for $lock_file to be created..."; fi
+        sleep 1
+    done
+    echo "Timed out waiting for test VMs to be ready. See details in: $TEST_VMS_SETUP_OUTPUT_FILE" >&2
+    return 1
+}
@@ -1,24 +1,87 @@
-FROM ubuntu
+FROM golang:1.8.1

-RUN apt-get -y update && apt-get -y install --no-install-recommends ca-certificates apt-transport-https
-RUN apt-key adv --keyserver hkp://keyserver.ubuntu.com:80 --recv-keys 36A1D7869245C8950F966E92D8576A8BA88D21E9
-RUN echo deb https://get.docker.io/ubuntu docker main > /etc/apt/sources.list.d/docker.list
-RUN apt-get -y update && apt-get -y install --no-install-recommends build-essential git lxc-docker-1.3.1 mercurial libpcap-dev curl make pkg-config gcc bison flex python-requests
+# Support Raspberry Pi 2 and newer
+ENV GOARM 7

-# When doing a build in a container, "apt-get update" happens twice,
-# which can be a very significant overhead for incremental builds.
-# And it's unnecessary, because if any of the prerequisites do change,
-# the right thing to do is to rebuild the container image.  Rather
-# than suppressing the "apt-get update" commands explicitly, we clear
-# out sources.list so that 'apt-get update' doesn't do anything.
-RUN echo >/etc/apt/sources.list
+# The names of the architectures that we're fetching cross-compilers for from the Debian packages
+ENV DEB_CROSSPLATFORMS \
+	armhf \
+	arm64

-ENV GO_VERSION 1.4.2
-RUN curl -sSL https://golang.org/dl/go${GO_VERSION}.linux-amd64.tar.gz | tar -C /usr/local -xz
-ENV PATH /usr/local/go/bin:$PATH
+# The go platforms we are supporting to cross-compile
+ENV GO_CROSSPLATFORMS \
+	linux/arm \
+	linux/arm64

-# So that sudo does not trample on PATH
-RUN sed -ie '/secure_path/d' /etc/sudoers
+# The names of the gcc cross-compilers we have installed
+ENV GCC_CROSSCOMPILERS \
+	arm-linux-gnueabihf \
+	aarch64-linux-gnu
+
+# Temporarily add the Debian repositories, because we're going to install gcc cross-compilers from there
+# Install the build-essential and crossbuild-essential-ARCH packages
+RUN echo "deb http://emdebian.org/tools/debian/ jessie main" > /etc/apt/sources.list.d/cgocrosscompiling.list \
+  && curl http://emdebian.org/tools/debian/emdebian-toolchain-archive.key | apt-key add - \
+  && for platform in ${DEB_CROSSPLATFORMS}; do dpkg --add-architecture $platform; done \
+  && apt-get update \
+  && apt-get install -y build-essential \
+  && for platform in ${DEB_CROSSPLATFORMS}; do apt-get install -y crossbuild-essential-${platform}; done \
+  && apt-get clean && rm -rf /var/lib/apt/lists/*
+
+# Install some required packages
+# libpcap is required because we're linking against its C libraries from the prog/weaver binary
+# flex and bison are required packages for compiling libpcap manually later
+RUN apt-get update \
+	&& apt-get install -y \
+		libpcap-dev \
+		python-requests \
+		time \
+		file \
+		flex \
+		bison
+
+RUN curl -fsSLo shfmt https://github.com/mvdan/sh/releases/download/v1.3.0/shfmt_v1.3.0_linux_amd64 && \
+	echo "b1925c2c405458811f0c227266402cf1868b4de529f114722c2e3a5af4ac7bb2  shfmt" | sha256sum -c && \
+	chmod +x shfmt && \
+	mv shfmt /usr/bin
+
+# Install common Go tools
+RUN go get \
+	github.com/golang/lint/golint \
+	github.com/fzipp/gocyclo \
+	github.com/fatih/hclfmt \
+	github.com/client9/misspell/cmd/misspell
+
+# First clean, because we're gonna rebuild the std both with -race enabled and disabled
+# Running go install std twice is intentional and required, as content is written in two different directories
+# (/usr/local/go/pkg/linux_amd64 vs /usr/local/go/pkg/linux_amd64_race)
+# The other architectures do not have support for -race
+RUN go clean -i net \
+	&& go install -tags netgo std \
+	&& go install -race -tags netgo std
+
+# Prebuild std for the other architectures
+RUN for platform in ${GO_CROSSPLATFORMS}; do \
+		GOOS=${platform%/*} GOARCH=${platform##*/} \
+			go install -tags netgo std; done
+
+# Allow full write access to the Go folders for anyone
+RUN chmod -R a+w /usr/local/go
+
+# The libpcap version from Debian packages is 1.6.2, matching that version here although newer versions of libpcap have been released
+# We have to cross-compile the libpcap library for the new architectures, that's what we're doing here
+ENV LIBPCAP_CROSS_VERSION=1.6.2
+RUN curl -sSL http://www.tcpdump.org/release/libpcap-${LIBPCAP_CROSS_VERSION}.tar.gz | tar -xz \
+	&& cd libpcap-${LIBPCAP_CROSS_VERSION} \
+	&& for crosscompiler in ${GCC_CROSSCOMPILERS}; do \
+		CC=${crosscompiler}-gcc ac_cv_linux_vers=2 ./configure --host=${crosscompiler} --with-pcap=linux \
+		&& make \
+		&& export LIB_DIR="/usr/local/lib/${crosscompiler}-gcc" VER="$(cat ./VERSION)" MAJOR_VER="$(sed 's/\([0-9][0-9]*\)\..*/\1/' ./VERSION)" \
+		&& mkdir -p ${LIB_DIR} \
+		&& cp -f libpcap.a libpcap.so.${VER} ${LIB_DIR} \
+		&& ln -sf libpcap.so.${VER} /usr/local/lib/libpcap.so.${MAJOR_VER} \
+		&& ln -sf libpcap.so.${MAJOR_VER} /usr/local/lib/libpcap.so \
+		&& make clean; done

 COPY build.sh /
 ENTRYPOINT ["sh", "/build.sh"]
@@ -1,45 +1,28 @@
 #!/bin/sh
-set -e

-GOPATH=/home/go
-export GOPATH
+set -eu

 WEAVE_SRC=$GOPATH/src/github.com/weaveworks/weave

-if [ $# -eq 0 -o "$1" = "tests" ] ; then
-    # No arguments.  Expect that the weave repo will be bind-mounted
-    # into $GOPATH
-    if ! [ -e $WEAVE_SRC ] ; then
-        cat 2>&1 <<EOF
-No container arguments supplied, and nothing at ${WEAVE_SRC}.  Please
-either bind-mount the golang workspace containing weave with the
+if ! [ -e $WEAVE_SRC ] ; then
+    cat 2>&1 <<EOF
+Nothing at ${WEAVE_SRC}.  Please bind-mount the weave repo with the
 docker run -v option, e.g.:

-    $ docker run -v <host gopath>:${GOPATH} \\
-          -v /var/run/docker.sock:/var/run/docker.sock weaveworks/weave-build
-
-Or supply git clone arguments to retrieve it, e.g.:
-
-    $ docker run -v /var/run/docker.sock:/var/run/docker.sock \\
-          weaveworks/weave-build https://github.com/weaveworks/weave.git
+    $ cd weave
+	$ docker run -v $(pwd):/go/src/github.com/weaveworks/weave \\
+		weaveworks/weave-build
 EOF
-        exit 1
-    fi
-
-    # If we run make directly, any files created on the bind mount
-    # will have awkward ownership.  So we switch to a user with the
-    # same user and group IDs as source directory.  We have to set a
-    # few things up so that sudo works without complaining later on.
-    uid=$(stat --format="%u" $WEAVE_SRC)
-    gid=$(stat --format="%g" $WEAVE_SRC)
-    echo "weave:x:$uid:$gid::$WEAVE_SRC:/bin/sh" >>/etc/passwd
-    echo "weave:*:::::::" >>/etc/shadow
-    echo "weave	ALL=(ALL)	NOPASSWD: ALL" >>/etc/sudoers
-
-    su weave -c "PATH=$PATH make -C $WEAVE_SRC build"
-else
-    # There are arguments to pass to git-clone
-    mkdir -p ${WEAVE_SRC%/*}
-    git clone "$@" $WEAVE_SRC
-    make -C $WEAVE_SRC build
 fi
+
+# If we run make directly, any files created on the bind mount
+# will have awkward ownership.  So we switch to a user with the
+# same user and group IDs as source directory.  We have to set a
+# few things up so that sudo works without complaining later on.
+uid=$(stat --format="%u" $WEAVE_SRC)
+gid=$(stat --format="%g" $WEAVE_SRC)
+echo "weave:x:$uid:$gid::$WEAVE_SRC:/bin/sh" >>/etc/passwd
+echo "weave:*:::::::" >>/etc/shadow
+echo "weave	ALL=(ALL)	NOPASSWD: ALL" >>/etc/sudoers
+
+su weave -c "PATH=$PATH make -C $WEAVE_SRC BUILD_IN_CONTAINER=false $*"
@@ -0,0 +1 @@
+8f313a9df867c207ae9ceac6e0935d1d2a0360aafce780dd5d1e0687ee98e0c3 docker-1.10.3.tgz
@@ -0,0 +1 @@
+7541b5490cee1183d911f909c0e91d3d9d57b2b775e40f14a1a7a6de216ca4a8 docker-1.10.3.tgz
@@ -0,0 +1 @@
+c02d861383d035ba020b361ef93bf14beb3964a42a6ae645b505b6148493e7ec qemu-aarch64-static
@@ -0,0 +1 @@
+0ff63682b8868620d771aa812d3053dc3c579dc8633ce62d3d7ae7eee83beb06 qemu-arm-static
@@ -2,77 +2,65 @@ general:
  branches:
    ignore:
      - gh\-pages
-      - /pull\/.*/

 machine:
+  pre:
+    - curl -sSL https://s3.amazonaws.com/circle-downloads/install-circleci-docker.sh | bash -s -- 1.10.0
  services:
    - docker
  environment:
    GOPATH: /home/ubuntu:$GOPATH
    SRCDIR: /home/ubuntu/src/github.com/weaveworks/weave
-    PATH: $PATH:$HOME/.local/bin:$HOME/google-cloud-sdk/bin/
+    PATH: $PATH:$HOME/.local/bin:/opt/google-cloud-sdk/bin
    CLOUDSDK_CORE_DISABLE_PROMPTS: 1
-    WEAVE_TEST: $HOME/docker/weave-test.tar
+    STATE: /home/ubuntu/state.env
+    # Base name of VMs for integration tests:
+    NAME: test-$CIRCLE_BUILD_NUM-$CIRCLE_NODE_INDEX
+    TEST_VMS_SETUP_OUTPUT_FILE: $CIRCLE_ARTIFACTS/test_vms_setup_output.txt
+    TEST_VMS_PROV_AND_CONF_LOCK_FILE: $CIRCLE_ARTIFACTS/test_vms_provisioned_and_configured.txt
+    TEST_VMS_READY_LOCK_FILE: $CIRCLE_ARTIFACTS/test_vms_ready.txt
+    # Google Cloud Platorm images for integration tests:
+    USE_IMAGE: 1
+    DISK_NAME_PREFIX: test-$CIRCLE_BUILD_NUM-0

 dependencies:
  cache_directories:
    - "~/docker"
-  override:
-    - if [[ -e "$WEAVE_TEST" ]]; then
-        docker load -i $WEAVE_TEST;
-      else
-        docker build -t weaveworks/weave-build build;
-        docker build -t weaveworks/weave-test test/image;
-        mkdir -p $(dirname "$WEAVE_TEST");
-        docker save weaveworks/weave-test >$WEAVE_TEST;
-      fi
  post:
-    - sudo apt-get update
-    - sudo apt-get install bc jq
-    - pip install requests
-    - curl https://sdk.cloud.google.com | bash
-    - bin/setup-circleci-secrets "$SECRET_PASSWORD"
-    - mkdir -p $(dirname $SRCDIR)
-    - cp -r $(pwd)/ $SRCDIR
-    - cd $SRCDIR/test; ./gce.sh make_template:
-        parallel: false
+    - bin/circle-dependencies-post-sources
+    # Provision & configure testing VMs:
+    - "while [ ! -d $SRCDIR ]; do sleep 1; done; $SRCDIR/bin/provision_test_vms.sh":
+        parallel: true
+        background: true
+    - bin/circle-dependencies-post

 test:
+  pre:
+    - "$SRCDIR/bin/circle-test-pre":
+        parallel: true
+        background: true
  override:
-    - cd $SRCDIR; make lint:
+    - "$SRCDIR/bin/circle-test-unit":
        parallel: true
-    - docker run -v /home/ubuntu:/home/go -e COVERDIR=test/coverage -e SLOW=true -e CIRCLECI -e CIRCLE_BUILD_NUM -e CIRCLE_NODE_TOTAL -e CIRCLE_NODE_INDEX weaveworks/weave-test:
-        parallel: true
-    - docker run -v /var/run/docker.sock:/run/docker.sock -v /home/ubuntu:/home/go -e COVERAGE=true weaveworks/weave-build:
-        parallel: true
-    - cd $SRCDIR/test; ./gce.sh setup:
-        parallel: true
-    - cd $SRCDIR/test; eval $(./gce.sh hosts); ./setup.sh:
-        parallel: true
-    - cd $SRCDIR/test; eval $(./gce.sh hosts); export COVERAGE=true; ./run_all.sh:
+    - "$SRCDIR/bin/circle-test-smoke":
        parallel: true
        timeout: 300
-    - cd $SRCDIR; make clean-bin:
-        parallel: true
-    - docker run -v /var/run/docker.sock:/run/docker.sock -v /home/ubuntu:/home/go weaveworks/weave-build:
-        parallel: true
  post:
-    - cd $SRCDIR/test; ./gce.sh destroy:
+    - "$SRCDIR/bin/circle-test-teardown":
        parallel: true

 teardown:
  pre:
-    - test "$CIRCLE_NODE_INDEX" != "0" || (cd $SRCDIR/test; ./gen_coverage_reports.sh):
+    # Destroy testing VMs:
+    - "cd $SRCDIR/test && ./run-integration-tests.sh destroy":
        parallel: true
-    - test "$CIRCLE_NODE_INDEX" != "0" || (go get github.com/mattn/goveralls && goveralls -repotoken $COVERALLS_REPO_TOKEN -coverprofile=$SRCDIR/test/profile.cov -service=circleci || true):
-        parallel: true
-    - test "$CIRCLE_NODE_INDEX" != "0" || (cd $SRCDIR/test; cp coverage.* $CIRCLE_ARTIFACTS):
+        background: true
+    - "$SRCDIR/bin/circle-teardown-pre":
        parallel: true

 deployment:
-  hub:
+  master:
    branch: master
    owner: weaveworks
    commands:
-      - docker login -e $DOCKER_EMAIL -u $DOCKER_USER -p $DOCKER_PASS
-      - cd $SRCDIR; make publish
+      - "$SRCDIR/bin/circle-deploy-master"
@@ -0,0 +1,7 @@
+## Community Code of Conduct
+
+Weaveworks follows the [CNCF Code of Conduct](https://github.com/cncf/foundation/blob/master/code-of-conduct.md).
+
+Instances of abusive, harassing, or otherwise unacceptable behavior
+may be reported by contacting a Weaveworks project maintainer, or
+Alexis Richardson alexis@weave.works.
@@ -1,82 +1,256 @@
 package docker

 import (
-	"github.com/fsouza/go-dockerclient"
-	. "github.com/weaveworks/weave/common"
+	"errors"
+	"fmt"
+	"strings"
+	"time"
+
+	docker "github.com/fsouza/go-dockerclient"
+
+	"github.com/weaveworks/weave/common"
+)
+
+const (
+	InitialInterval = 1 * time.Second
+	MaxInterval     = 20 * time.Second
 )

 // An observer for container events
 type ContainerObserver interface {
 	ContainerStarted(ident string)
 	ContainerDied(ident string)
+	ContainerDestroyed(ident string)
 }

 type Client struct {
 	*docker.Client
 }

+type syncPair struct {
+	stop chan struct{}
+	done chan struct{}
+}
+
+type pendingStarts map[string]*syncPair
+
 // NewClient creates a new Docker client and checks we can talk to Docker
 func NewClient(apiPath string) (*Client, error) {
+	if apiPath != "" && !strings.Contains(apiPath, "://") {
+		apiPath = "tcp://" + apiPath
+	}
 	dc, err := docker.NewClient(apiPath)
 	if err != nil {
 		return nil, err
 	}
 	client := &Client{dc}

-	return client, client.checkWorking(apiPath)
+	return client, client.checkWorking()
 }

 func NewVersionedClient(apiPath string, apiVersionString string) (*Client, error) {
+	if !strings.Contains(apiPath, "://") {
+		apiPath = "tcp://" + apiPath
+	}
 	dc, err := docker.NewVersionedClient(apiPath, apiVersionString)
 	if err != nil {
 		return nil, err
 	}
 	client := &Client{dc}

-	return client, client.checkWorking(apiPath)
+	return client, client.checkWorking()
 }

-func (c *Client) checkWorking(apiPath string) error {
+func NewVersionedClientFromEnv(apiVersionString string) (*Client, error) {
+	dc, err := docker.NewVersionedClientFromEnv(apiVersionString)
+	if err != nil {
+		return nil, err
+	}
+	client := &Client{dc}
+
+	return client, client.checkWorking()
+}
+
+func (c *Client) checkWorking() error {
+	_, err := c.Version()
+	return err
+}
+
+func (c *Client) Info() string {
 	env, err := c.Version()
 	if err != nil {
-		return err
+		return fmt.Sprintf("Docker API error: %s", err)
 	}
-	Log.Infof("[docker] Using Docker API on %s: %v", apiPath, env)
-	return nil
+	return fmt.Sprintf("Docker API on %s: %v", c.Endpoint(), env)
+}
+
+func (c *Client) DockerVersion() string {
+	if env, err := c.Version(); err == nil {
+		if v, found := env.Map()["Version"]; found {
+			return v
+		}
+	}
+	return "unknown"
 }

 // AddObserver adds an observer for docker events
 func (c *Client) AddObserver(ob ContainerObserver) error {
-	events := make(chan *docker.APIEvents)
-	if err := c.AddEventListener(events); err != nil {
-		Log.Errorf("[docker] Unable to add listener to Docker API: %s", err)
-		return err
-	}
-
 	go func() {
-		for event := range events {
-			switch event.Status {
-			case "start":
-				id := event.ID
-				ob.ContainerStarted(id)
-			case "die":
-				id := event.ID
-				ob.ContainerDied(id)
+		pending := make(pendingStarts)
+		retryInterval := InitialInterval
+		for {
+			events := make(chan *docker.APIEvents)
+			if err := c.AddEventListener(events); err != nil {
+				c.errorf("Unable to add listener to Docker API: %s - retrying in %ds", err, retryInterval/time.Second)
+			} else {
+				start := time.Now()
+				for event := range events {
+					switch event.Status {
+					case "start":
+						pending.finish(event.ID)
+						pending.start(event.ID, c, ob)
+					case "die":
+						pending.finish(event.ID)
+						ob.ContainerDied(event.ID)
+					case "destroy":
+						pending.finish(event.ID)
+						ob.ContainerDestroyed(event.ID)
+					}
+				}
+				if time.Since(start) > retryInterval {
+					retryInterval = InitialInterval
+				}
+				c.errorf("Event listener channel closed - retrying subscription in %ds", retryInterval/time.Second)
+			}
+			time.Sleep(retryInterval)
+			retryInterval = retryInterval * 3 / 2
+			if retryInterval > MaxInterval {
+				retryInterval = MaxInterval
 			}
 		}
 	}()
 	return nil
 }

+// Docker sends a 'start' event before it has attempted to start the
+// container.  Delay notifying the observer until the container has a
+// pid, or we are told to stop when a 'die' event arrives.
+//
+// Note we always deliver the event, even if the container seems to
+// have gone away.
+func (pending pendingStarts) start(id string, c *Client, ob ContainerObserver) {
+	sync := syncPair{make(chan struct{}), make(chan struct{})}
+	pending[id] = &sync
+	go func() {
+		defer close(sync.done)
+		defer ob.ContainerStarted(id)
+		for {
+			if container, err := c.InspectContainer(id); err != nil || container.State.Pid != 0 {
+				return
+			}
+			select {
+			case <-sync.stop:
+				return
+			case <-time.After(200 * time.Millisecond):
+			}
+		}
+	}()
+}
+
+func (pending pendingStarts) finish(id string) {
+	if sync, found := pending[id]; found {
+		close(sync.stop)
+		<-sync.done
+		delete(pending, id)
+	}
+}
+
+// AllContainerIDs returns all the IDs of Docker containers,
+// whether they are running or not.
+func (c *Client) AllContainerIDs() ([]string, error) {
+	return c.containerIDs(true)
+}
+
+// RunningContainerIDs returns all the IDs of the running
+// Docker containers.
+func (c *Client) RunningContainerIDs() ([]string, error) {
+	return c.containerIDs(false)
+}
+
+func (c *Client) containerIDs(all bool) ([]string, error) {
+	containers, err := c.ListContainers(docker.ListContainersOptions{All: all})
+	if err != nil {
+		return nil, err
+	}
+	var ids []string
+	for _, c := range containers {
+		ids = append(ids, c.ID)
+	}
+	return ids, nil
+}
+
 // IsContainerNotRunning returns true if we have checked with Docker that the ID is not running
 func (c *Client) IsContainerNotRunning(idStr string) bool {
 	container, err := c.InspectContainer(idStr)
 	if err == nil {
-		return !container.State.Running
+		return !container.State.Running || container.State.Restarting
 	}
 	if _, notThere := err.(*docker.NoSuchContainer); notThere {
 		return true
 	}
-	Log.Errorf("[docker] Could not check container status: %s", err)
+	c.errorf("Could not check container status: %s", err)
 	return false
 }
+
+// This is intended to find an IP address that we can reach the container on;
+// if it is on the Docker bridge network then that address; if on the host network
+// then localhost
+func (c *Client) GetContainerIP(nameOrID string) (string, error) {
+	info, err := c.InspectContainer(nameOrID)
+	if err != nil {
+		return "", err
+	}
+	if info.NetworkSettings.Networks != nil {
+		if bridgeNetwork, ok := info.NetworkSettings.Networks["bridge"]; ok {
+			return bridgeNetwork.IPAddress, nil
+		} else if _, ok := info.NetworkSettings.Networks["host"]; ok {
+			return "127.0.0.1", nil
+		}
+	} else if info.HostConfig.NetworkMode == "host" {
+		return "127.0.0.1", nil
+	}
+	if info.NetworkSettings.IPAddress == "" {
+		return "", errors.New("No IP address found for container " + nameOrID)
+	}
+	return info.NetworkSettings.IPAddress, nil
+}
+
+func (c *Client) EnsureNetwork(networkName, driver, subnet string, options map[string]interface{}) error {
+	_, err := c.CreateNetwork(
+		docker.CreateNetworkOptions{
+			Name:           networkName,
+			CheckDuplicate: true,
+			Driver:         driver,
+			IPAM: docker.IPAMOptions{
+				Driver: driver,
+				Config: []docker.IPAMConfig{{Subnet: subnet}},
+			},
+			Options: options,
+		})
+	if err != docker.ErrNetworkAlreadyExists && err != nil {
+		// Despite appearances to the contrary, CreateNetwork does
+		// sometimes(always?) *not* return ErrNetworkAlreadyExists
+		// when the network already exists. Hence we need to check for
+		// this explicitly.
+		if _, err2 := c.NetworkInfo(networkName); err2 != nil {
+			return fmt.Errorf("unable to create network: %s", err)
+		}
+	}
+	return nil
+}
+
+// logging
+
+func (c *Client) errorf(fmt string, args ...interface{}) {
+	common.Log.Errorf("[docker] "+fmt, args...)
+}
@@ -0,0 +1,18 @@
+package common
+
+import (
+	"net/http"
+)
+
+type loggingHandler struct {
+	next http.Handler
+}
+
+func (l *loggingHandler) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	Log.Debugf("[http] %s %s", r.Method, r.URL.RequestURI())
+	l.next.ServeHTTP(w, r)
+}
+
+func LoggingHTTPHandler(h http.Handler) http.Handler {
+	return &loggingHandler{next: h}
+}
@@ -3,6 +3,7 @@ package common
 import (
 	"bytes"
 	"fmt"
+	"log"
 	"strings"

 	"github.com/Sirupsen/logrus"
@@ -52,3 +53,35 @@ func SetLogLevel(levelname string) {
 	}
 	Log.Level = level
 }
+
+func CheckFatal(e error) {
+	if e != nil {
+		Log.Fatal(e)
+	}
+}
+
+func CheckWarn(e error) {
+	if e != nil {
+		Log.Warnln(e)
+	}
+}
+
+// Adaptor type to go from Logrus to something expecting a standard library log.Logger
+type logLogger struct {
+	logger *logrus.Logger
+}
+
+// io.Writer interface
+func (l logLogger) Write(p []byte) (n int, err error) {
+	ln := len(p)
+	// Strip a trailing newline, because logrus formatter adds one
+	if ln > 0 && p[ln-1] == '\n' {
+		p = p[:ln-1]
+	}
+	l.logger.Info(string(p))
+	return ln, nil
+}
+
+func LogLogger() *log.Logger {
+	return log.New(logLogger{Log}, "", 0)
+}
@@ -1,31 +0,0 @@
-package mflagext
-
-import (
-	"fmt"
-
-	"github.com/docker/docker/pkg/mflag"
-)
-
-type listOpts struct {
-	value      *[]string
-	hasBeenSet bool
-}
-
-func ListVar(p *[]string, names []string, value []string, usage string) {
-	*p = value
-	mflag.Var(&listOpts{p, false}, names, usage)
-}
-
-func (opts *listOpts) Set(value string) error {
-	if opts.hasBeenSet {
-		(*opts.value) = append((*opts.value), value)
-	} else {
-		(*opts.value) = []string{value}
-		opts.hasBeenSet = true
-	}
-	return nil
-}
-
-func (opts *listOpts) String() string {
-	return fmt.Sprintf("%v", []string(*opts.value))
-}
@@ -1,4 +1,4 @@
-package router
+package odp

 import (
 	"fmt"
@@ -10,49 +10,50 @@ import (

 // ODP admin functionality

-func CreateDatapath(dpname string) (err error, supported bool) {
+func CreateDatapath(dpname string) (supported bool, err error) {
 	dpif, err := odp.NewDpif()
 	if err != nil {
 		if odp.IsKernelLacksODPError(err) {
-			return nil, false
+			return false, nil
 		}
-
-		return err, true
+		return true, err
 	}
-
 	defer dpif.Close()

 	dp, err := dpif.CreateDatapath(dpname)
-	if err != nil && !odp.IsDatapathNameAlreadyExistsError(err) {
-		return err, true
+	if err != nil {
+		if !odp.IsDatapathNameAlreadyExistsError(err) {
+			return true, err
+		}
+		if dp, err = dpif.LookupDatapath(dpname); err != nil {
+			return true, err
+		}
 	}

 	// Pick an ephemeral port number to use in probing for vxlan
 	// support.
 	udpconn, err := net.ListenUDP("udp4", nil)
 	if err != nil {
-		return err, true
+		return true, err
 	}
+	defer udpconn.Close()

 	// we leave the UDP socket open, so creating a vxlan vport on
 	// the same port number should fail.  But that's fine: It's
 	// still sufficient to probe for support.
 	portno := uint16(udpconn.LocalAddr().(*net.UDPAddr).Port)
-	vpid, err := dp.CreateVport(odp.NewVxlanVportSpec(
-		fmt.Sprintf("vxlan-%d", portno), portno))
+	vpid, err := dp.CreateVport(odp.NewVxlanVportSpec(fmt.Sprintf("vxlan-%d", portno), portno))
 	if nlerr, ok := err.(odp.NetlinkError); ok {
 		if syscall.Errno(nlerr) == syscall.EAFNOSUPPORT {
 			dp.Delete()
-			return fmt.Errorf("kernel does not have Open vSwitch VXLAN support"), false
+			return false, fmt.Errorf("kernel does not have Open vSwitch VXLAN support")
 		}
 	}
-
 	if err == nil {
 		dp.DeleteVport(vpid)
 	}

-	udpconn.Close()
-	return nil, true
+	return true, nil
 }

 func DeleteDatapath(dpname string) error {
@@ -60,7 +61,6 @@ func DeleteDatapath(dpname string) error {
 	if err != nil {
 		return err
 	}
-
 	defer dpif.Close()

 	dp, err := dpif.LookupDatapath(dpname)
@@ -68,7 +68,6 @@ func DeleteDatapath(dpname string) error {
 		if odp.IsNoSuchDatapathError(err) {
 			return nil
 		}
-
 		return err
 	}

@@ -80,7 +79,6 @@ func AddDatapathInterface(dpname string, ifname string) error {
 	if err != nil {
 		return err
 	}
-
 	defer dpif.Close()

 	dp, err := dpif.LookupDatapath(dpname)
@@ -0,0 +1,27 @@
+package common
+
+import (
+	"os"
+	"strconv"
+)
+
+func AllPids(procRoot string) ([]int, error) {
+	fh, err := os.Open(procRoot)
+	if err != nil {
+		return nil, err
+	}
+	defer fh.Close()
+	dirNames, err := fh.Readdirnames(-1)
+	if err != nil {
+		return nil, err
+	}
+	pids := make([]int, len(dirNames))
+	for _, dirName := range dirNames {
+		pid, err := strconv.Atoi(dirName)
+		if err != nil { // Only interested in numeric entries - skip /proc/stat, etc.
+			continue
+		}
+		pids = append(pids, pid)
+	}
+	return pids, err
+}
@@ -1,32 +0,0 @@
-package common
-
-import (
-	"os"
-	"os/signal"
-	"runtime"
-	"syscall"
-)
-
-// A subsystem/server/... that can be stopped or queried about the status with a signal
-type SignalReceiver interface {
-	Stop() error
-}
-
-func SignalHandlerLoop(ss ...SignalReceiver) {
-	sigs := make(chan os.Signal, 1)
-	signal.Notify(sigs, syscall.SIGINT, syscall.SIGQUIT, syscall.SIGTERM)
-	buf := make([]byte, 1<<20)
-	for {
-		switch <-sigs {
-		case syscall.SIGINT, syscall.SIGTERM:
-			Log.Infof("=== received SIGINT/SIGTERM ===\n*** exiting")
-			for _, subsystem := range ss {
-				subsystem.Stop()
-			}
-			return
-		case syscall.SIGQUIT:
-			stacklen := runtime.Stack(buf, true)
-			Log.Infof("=== received SIGQUIT ===\n*** goroutine dump...\n%s\n*** end", buf[:stacklen])
-		}
-	}
-}
@@ -1,8 +1,6 @@
 package common

-import (
-	"strings"
-)
+import "strings"

 // Assert test is true, panic otherwise
 func Assert(test bool) {
@@ -0,0 +1,25 @@
+package common
+
+import (
+	"sync"
+	"sync/atomic"
+)
+
+type WaitGroup struct {
+	wg    sync.WaitGroup
+	count int32
+}
+
+// Increase the count of things waiting and return a closure which calls Done
+func (w *WaitGroup) Add() func() {
+	atomic.AddInt32(&w.count, 1)
+	w.wg.Add(1)
+	return func() {
+		w.wg.Done()
+		atomic.AddInt32(&w.count, -1)
+	}
+}
+
+func (w *WaitGroup) IsDone() bool {
+	return atomic.LoadInt32(&w.count) == 0
+}
@@ -0,0 +1,115 @@
+package db
+
+import (
+	"bytes"
+	"encoding/gob"
+	"fmt"
+	"time"
+
+	"github.com/boltdb/bolt"
+)
+
+type DB interface {
+	Load(string, interface{}) (bool, error)
+	Save(string, interface{}) error
+}
+
+type BoltDB struct {
+	db *bolt.DB
+}
+
+var (
+	versionIdent       = []byte("version")
+	persistenceVersion = []byte{1, 0} // major.minor
+	topBucket          = []byte("top")
+)
+
+const (
+	NameIdent = "peername"
+	fileName  = "data.db"
+)
+
+func Pathname(dbPrefix string) string {
+	return dbPrefix + fileName
+}
+
+func NewBoltDB(dbPrefix string) (*BoltDB, error) {
+	dbPathname := Pathname(dbPrefix)
+	db, err := bolt.Open(dbPathname, 0660, nil)
+	if err != nil {
+		return nil, fmt.Errorf("[boltDB] Unable to open %s: %s", dbPathname, err)
+	}
+	err = db.Update(checkVersion(false))
+	return &BoltDB{db: db}, err
+}
+
+func NewBoltDBReadOnly(dbPrefix string) (*BoltDB, error) {
+	options := bolt.Options{Timeout: time.Millisecond * 50, ReadOnly: true}
+	dbPathname := Pathname(dbPrefix)
+	db, err := bolt.Open(dbPathname, 0660, &options)
+	if err != nil {
+		return nil, fmt.Errorf("[boltDB] Unable to open %s: %s", dbPathname, err)
+	}
+	err = db.View(checkVersion(true))
+	if err != nil {
+		return nil, fmt.Errorf("[boltDB] Cannot use persistence file %s: %s", dbPathname, err)
+	}
+	return &BoltDB{db: db}, nil
+}
+
+func checkVersion(readOnly bool) func(tx *bolt.Tx) error {
+	return func(tx *bolt.Tx) error {
+		if top := tx.Bucket(topBucket); top == nil {
+			if readOnly {
+				return fmt.Errorf("no top bucket")
+			}
+			top, err := tx.CreateBucket(topBucket)
+			if err != nil {
+				return err
+			}
+			if err := top.Put(versionIdent, persistenceVersion); err != nil {
+				return err
+			}
+		} else {
+			if checkVersion := top.Get(versionIdent); checkVersion != nil {
+				if checkVersion[0] != persistenceVersion[0] {
+					return fmt.Errorf("mismatched version %x", checkVersion)
+				}
+			}
+		}
+		return nil
+	}
+}
+
+func (d *BoltDB) Load(ident string, data interface{}) (bool, error) {
+	found := true
+	err := d.db.View(func(tx *bolt.Tx) error {
+		b := tx.Bucket(topBucket)
+		v := b.Get([]byte(ident))
+		if v == nil {
+			found = false
+			return nil
+		}
+		reader := bytes.NewReader(v)
+		decoder := gob.NewDecoder(reader)
+		err := decoder.Decode(data)
+		return err
+	})
+	return found, err
+}
+
+func (d *BoltDB) Save(ident string, data interface{}) error {
+	return d.db.Update(func(tx *bolt.Tx) error {
+		b := tx.Bucket(topBucket)
+		buf := new(bytes.Buffer)
+		enc := gob.NewEncoder(buf)
+		if err := enc.Encode(data); err != nil {
+			return err
+		}
+		return b.Put([]byte(ident), buf.Bytes())
+	})
+}
+
+func (d *BoltDB) Close() error {
+	return d.db.Close()
+}
@@ -0,0 +1,81 @@
+# Weave Net Benchmarks
+
+## 2017-02-16: Fast Datapath Encryption
+
+Results: https://www.weave.works/weave-net-performance-fast-datapath/
+
+* Two `c3.8xlarge` instances running Ubuntu 16.04 LTS (`ami-d8f4deab`) in
+`eu-west-1` region of AWS.
+* 10 Gigabit [Enhanced Networking](http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/enhanced-networking.html).
+* `iperf 3.0.11`.
+
+### fastdp
+
+* Encrypted:
+
+```
+host1$ WEAVE_MTU=8916 weave launch --password=foobar
+host1$ docker $(weave config) run --rm -ti --name server networkstatic/iperf3 -s
+
+host2$ WEAVE_MTU=8916 weave launch --password=foobar host1
+host2$ docker $(weave config) run --rm -ti networkstatic/iperf3 -c server
+```
+
+* Non-encrypted:
+
+```
+host1$ WEAVE_MTU=8950 weave launch
+host1$ docker $(weave config) run --rm -ti --name server networkstatic/iperf3 -s
+
+host2$ WEAVE_MTU=8950 weave launch --password=foobar host1
+host2$ docker $(weave config) run --rm -ti networkstatic/iperf3 -c server
+```
+
+### sleeve
+
+* Encrypted:
+
+```
+host1$ WEAVE_NO_FASTDP=1 weave launch --password=foobar
+host1$ docker $(weave config) run --rm -ti --name server networkstatic/iperf3 -s
+
+host2$ WEAVE_NO_FASTDP=1 weave launch --password=foobar host1
+host2$ docker $(weave config) run --rm -ti networkstatic/iperf3 -c server
+```
+
+* Non-encrypted:
+
+```
+host1$ WEAVE_NO_FASTDP=1 weave launch
+host1$ docker $(weave config) run --rm -ti --name server networkstatic/iperf3 -s
+
+host2$ WEAVE_NO_FASTDP=1 weave launch --password=foobar host1
+host2$ docker $(weave config) run --rm -ti networkstatic/iperf3 -c server
+```
+
+### host
+
+* Non-encrypted:
+
+```
+host1$ iperf -s
+host2$ iperf -c host1
+```
+
+* Encrypted:
+
+```
+host1$ export KEY1="0x466454f2b1a770f8f872f9afbc35ebeac57e00fc11ac86ed1f82716f010b20f0cf532274"
+host1$ export KEY2="0x3531151241a770f8f872f9afbc35ebeac57e00fc11ac86ed1f82716f010b20f0cf532274"
+host1# ip xfrm state add src ${IP1} dst ${IP2} proto esp spi 0x4c856ffc replay-window 256 flag esn reqid 0 mode transport aead 'rfc4106(gcm(aes))' ${KEY1} 128
+host1# ip xfrm state add src ${IP2} dst ${IP1} proto esp spi 0x9b0830bc replay-window 256 flag esn reqid 0 mode transport aead 'rfc4106(gcm(aes))' ${KEY2} 128
+host1# ip xfrm policy add src ${IP1}/32 dst ${IP2}/32 dir out tmpl src ${IP1} dst ${IP2} proto esp spi 0x4c856ffc reqid 0 mode transport
+host1$ iperf -s
+
+host2$ export KEY1="0x466454f2b1a770f8f872f9afbc35ebeac57e00fc11ac86ed1f82716f010b20f0cf532274"
+host2$ export KEY2="0x3531151241a770f8f872f9afbc35ebeac57e00fc11ac86ed1f82716f010b20f0cf532274"
+host2# ip xfrm state add src ${IP2} dst ${IP1} proto esp spi 0x9b0830bc reqid 0 replay-window 256 flag esn mode transport aead 'rfc4106(gcm(aes))' ${KEY2} 128
+host2# ip xfrm state add src ${IP1} dst ${IP2} proto esp spi 0x4c856ffc reqid 0 replay-window 256 flag esn mode transport aead 'rfc4106(gcm(aes))' ${KEY1} 128
+host2# ip xfrm policy add src ${IP2}/32 dst ${IP1}/32 dir out tmpl src ${IP1} dst ${IP2} proto esp spi 0x9b0830bc reqid 0 mode transport
+host2$ iperf -c host1
+```
@@ -0,0 +1,70 @@
+# Entrypoints
+
+This document describes the various ways that weave as a binary gets launched, and where and how it processes options.
+
+The goal is to enable maintainers to modify functionality without having to reverse engineer the code to find all of the places you can start a weave binary and what it depends upon.
+
+## Components
+### weaver
+
+`weaver` is the primary binary: it sets up the weave bridge to which containers are attached, manages IP address allocation, serves DNS requests and does many more things.
+
+Since version 2.0, `weaver` also bundles the Docker Plugin (legacy version) and Docker API Proxy which used to be shipped as separate containers.
+
+### weaveutil
+weaveutil is a binary which provides a number of functions for managing an existing weave network. It gets information about the network, attaches and detaches containers, etc.
+
+For the majority of interactions with an existing weave network, you will launch `weaveutil` in some manner or another.
+
+Almost all options can be passed as `--option` to `weaveutil`. These options are created in `prog/weaveutil/main.go` with a table mapping each command to a dedicated golang function.
+
+In addition to operating in normal mode, `weaveutil` has several additional operating modes. Before processing any commands, `weaveutil` checks the filename with which it was called.
+
+* If it was `weave-ipam` it delegates responsibility to the `cniIPAM()` function in `prog/weaveutil/cni.go`, which, in turn, calls the standard CNI plugin function `cni.PluginMain()`, passing it weave's IPAM implementation from `plugin/ipam`.
+* If it was `weave-net` it delegates responsibility to the `cniNet()` function in `prog/weaveutil/cni.go`, which, in turn, calls the standard CNI plugin function `cni.PluginMain()`, passing it weave's net plugin implementation from `plugin/net`.
+
+Finally, `weaveutil` is called by `weaver` when it needs to operate in a different network namespace.
+Go programs cannot safely switch namespaces; see this [issue](http://github.com/vishvananda/netns/issues/17).
+The call chain is `weaver`->`nsenter`->`weaveutil`.
+
+### weave
+Wrapping `weaver` and `weaveutil` is `weave`, a `sh` script that provides help information and calls `weaveutil` as relevant.
+
+### weave-kube
+weave-kube is an image with the weave binaries and a wrapper script installed. It is responsible for:
+
+1. Setting up the weave network
+2. Connecting to peers
+3. Copy the weave CNI-compatible binary plugin `weaveutil` to `/opt/cni/bin/weave-net` and weave config in `/etc/cni/net.d/10-weave.conf`
+4. Running the weave network policy controller (weave-npc) to implement kubernetes' `NetworkPolicy`
+
+Once installation is complete, each network-relevant change in a container leads `kubelet` to:
+
+1. Set certain environment variables to configure the CNI plugin, mostly related to the container ID
+2. Launch `/opt/cni/bin/weave-net`
+3. Pass the CNI config - the contents of `/etc/cni/net.d/10-weave.conf` to `weave-net`
+
+Thus, when each container is changed, and `kubelet` calls weave as a CNI plugin, it really just is launching `weaveutil` as `weave-net`.
+
+The installation and setup of all of the above - and therefore the entrypoint to the weave-kube image - is the script `prog/weave-kube/launch.sh`. `launch.sh` does the following:
+
+1. Read configuration variables from the environment. When the documentation for `weave-kube` describes configuring the weave network by changing the environment variables in the daemonset in the `.yml` file, `launch.sh` reads these environment variables.
+2. Set up the config file at `/etc/cni/net.d/10-weave.conf`
+3. Run the `weave` initialization script
+4. Run `weaver` with the correct configuration passed as command-line options
+
+
+## Adding Options
+To add new options to how weave should run with each invocation, you would do the following:
+
+1. determine to which `weave` command you want to add the option(s). `weave` normally is launched as `weave <command> <options> ...`.
+2. Add the option to `weave` script help for each `weave` command you wish to make it available. As of this writing, all of the commands and their options are listed in the function `usage_no_exist()`.
+3. Add the option to `weave` script option processing for the `weave` command, under the `case $COMMAND in` in the main `weave` script.
+4. In the function for the `weave` command, determine if the command should be passed on to `weaveutil` or `weaver`. Pass the option on to `weaveutil` or `weaver`, as appropriate, in the format `--option`, in the function for the `weave` command.
+5. Add a command-line option `--option` to `weaveutil` or `weaver` as appropriate.
+6. If the option can be configured for CNI:
+    * have `prog/weave-kube/launch.sh` read it as an environment variable and set inside
+    * set a default for the environment variable in `prog/weave-kube/weave-daemonset-k8s-1.6.yaml` and `weave-daemonset.yaml`
+    * if it should be set via `weaver` globally on its one-time initialization invocation, pass it on in `launch.sh`
+    * if it needs to be set via `weaveutil` on each invocation of `weave-net`, have `launch.sh` save it as an option in the CNI config file `/etc/cni/net.d/10-weave.conf` and then have the CNI code in `plugin/net/cni.go` in `weaveutil` read it and use where appropriate
+7. Document it!
@@ -0,0 +1,155 @@
+This document describes implementation details of the fast datapath encryption.
+
+# Overview
+
+At the high level, we use the ESP protocol ([RFC 2406][esp]) in the Transport
+mode. Each packet is encrypted with AES in GCM mode ([RFC 4106][aesgcm]), with
+32 byte key and 4 byte salt. This combo provides the following security
+properties:
+
+* Data confidentiality.
+* Data origin authentication.
+* Integrity.
+* Anti-replay.
+* Limited traffic flow confidentiality as fast datapath VXLAN packets are fully
+  encrypted.
+
+## Notation
+
+* `SAin`:  IPsec security association for inbound connections.
+* `SAout`: IPsec security association for outbound connections.
+* `SPout`: IPsec security policy for outbound connections. Used to match
+           outbound flows to SAout.
+
+## SA Key Derivation
+
+For each connection direction, a different AES-GCM key and salt is used.
+The pairs are derived with HKDF ([RFC 5869][hkdf]):
+
+```
+SAin[KeyAndSalt] = HKDF(sha256, ikm=sessionKey, salt=nonceIn, info=localPeerName)
+SAout[KeyAndSalt] = HKDF(sha256, ikm=sessionKey, salt=nonceOut, info=remotePeerName)
+```
+
+Where:
+
+* the mutual `sessionKey` is derived by the [Mesh][mesh] library during
+  the control plane connection establishment between the local peer and
+  the remote peer.
+* `nonceIn` and `nonceOut` are randomly generated 32byte nonces which
+  are exchanged over the encrypted control plane channel.
+
+## SPI
+
+A directional secure connection between two peers is identified with SPI.
+
+The kernel requires the pair of SPI and dst IP to be unique among security
+associations. Thus, to avoid collisions, we generate outbound SPI on a remote
+peer.
+
+## Connection Establishment
+
+```
+Peer A                                                        Peer B
+-----------------------------------------------------------------------------------------------------------------
+
+fastdp.fwd.Confirm():
+    install iptables blocking rules,                          fastdp.fwd.Confirm():
+    nonce_BA = rand(),                                            install iptables blocking rules,
+    {key,salt}_BA = hkdf(sessionKey, nonce_BA, A),                nonce_AB = rand(),
+    spi_BA = allocspi(),                                          {key,salt}_AB = hkdf(sessionKey, nonce_AB, B),
+    create SA_BA(B<-A, spi_BA, key_BA, salt_BA),                  spi_AB = allocspi(),
+    send InitSARemote(spi_BA, nonce_BA). -->                      create SA_AB(A<-B, spi_AB, key_AB, salt_AB),
+                                                              <-- send InitSARemote(spi_AB, nonce_AB).
+
+
+                                                          --> recv InitSARemote(spi_BA, nonce_BA):
+recv InitSARemote(spi_AB, nonce_AB): <--                          {key,salt}_BA = hkdf(sessionKey, nonce_BA, A),
+    {key,salt}_AB = hkdf(sessionKey, nonce_AB, B),                create SA_BA(B<-A, spi_BA, key_BA, salt_BA),
+    create SA_AB(A<-B, spi_AB, key_AB, salt_AB),                  create SP_BA(B<-A, spi_BA),
+    create SP_AB(A<-B, spi_AB),                                   install marking rule.
+    install marking rule.
+```
+
+# Implementation Details
+
+## XFRM
+
+The implementation is based on the kernel IP packet transformation framework
+called XFRM. Unfortunately, docs are barely existing and the complexity of
+the framework is high. The best resource I found is Chapter 10 in
+"Linux Kernel Networking: Implementation and Theory" by Rami Rosen.
+
+## iptables Rules
+
+The kernel VXLAN driver does not set a dst port of a tunnel in the ip flow
+descriptor, thus XFRM policy lookup cannot match a policy (SPout) which includes
+the port. This makes impossible to encrypt only tunneled traffic between
+peers. To work around, we mark such outgoing packets with iptables and set
+the same mark in the policy selector (funnily enough, `iptables_mangle` module
+eventually sets the missing dst port in the flow descriptor). The challenge
+here is to pick a mark that it would not interfere with other networking
+applications before OUTPUT'ing a packet. For example, Kubernetes by default
+uses 1<<14 and 1<<15 marks and we choose 1<<17 (0x20000). Additionally,
+such workaround brings the requirement for at least the 4.2 kernel.
+
+The marking rules are the following:
+
+```
+iptables -t mangle -A OUTPUT -j WEAVE-IPSEC-OUT
+iptables -t mangle -A WEAVE-IPSEC-OUT -s ${LOCAL_PEER_IP} -d ${REMOTE_PEER_IP} \
+         -p udp --dport ${TUNNEL_PORT} -j WEAVE-IPSEC-OUT-MARK
+iptables -t mangle -A WEAVE-IPSEC-OUT-MARK --set-xmark ${MARK} -j MARK
+```
+
+As Linux does not implement [IP Security Levels][ipseclevel], we install
+additional iptables rules to prevent from accidentally sending unencrypted
+traffic between peers which have previously established the secure connection.
+
+For inbound traffic, we mark each ESP packet with the mark and drop non-marked
+tunnel traffic:
+
+```
+iptables -t mangle -A INPUT -j WEAVE-IPSEC-IN
+iptables -t mangle -A WEAVE-IPSEC-IN -s ${REMOTE_PEER_IP} -d ${LOCAL_PEER_IP} \
+         -m esp --espspi ${SPIin} -j WEAVE-IPSEC-IN-MARK
+iptables -t mangle -A WEAVE-IPSEC-IN-MARK --set-xmark ${MARK} -j MARK
+
+iptables -t filter -A INPUT -j WEAVE-IPSEC-IN
+iptables -t filter -A WEAVE-IPSEC-IN -s ${REMOTE_PEER_IP} -d ${LOCAL_PEER_IP} \
+         -p udp --dport ${TUNNEL_PORT} -m mark ! --mark ${MARK} -j DROP
+```
+
+For outbound traffic, we drop marked traffic which does not match any SPout:
+
+```
+iptables -t filter -A OUTPUT ! -p esp -m policy --dir out --pol none \
+         -m mark --mark ${MARK} -j DROP
+```
+
+## ESN
+
+To prevent from cycling SeqNo which makes replay attacks possible, we use
+64-bit extended sequence numbers known as [ESN](esn).
+
+## MTU
+
+In addition to the VXLAN overhead, the MTU calculation should take into account
+the ESP overhead which is 34-37 bytes (encrypted Payload is 4 bytes aligned) and
+consists of:
+
+* 4 bytes (SPI).
+* 4 bytes (Sequence Number).
+* 8 bytes (ESP IV).
+* 1 byte (Pad Length).
+* 1 byte (NextHeader).
+* 16 bytes (ICV).
+* 0-3 bytes (Padding).
+
+
+[esp]:              https://tools.ietf.org/html/rfc2406
+[aesgcm]:           https://tools.ietf.org/html/rfc4106
+[hkdf]:             https://tools.ietf.org/html/rfc5869
+[esn]:              https://tools.ietf.org/html/rfc4304
+[ipseclevel]:       https://tools.ietf.org/html/draft-mcdonald-simple-ipsec-api-01)
+[mesh]:             https://github.com/weaveworks/mesh
@@ -0,0 +1,170 @@
+# Overlay Method Selection
+
+Weave automatically selects the best working overlay implementation
+when connecting to a new peer. It does this by initiating both
+forwarders in parallel, eventually settling on the most preferred
+established forwarder at which point the other forwarder is shut down,
+assuming it had not failed already. In the event that the remaining
+forwarder fails at a later time (e.g. because of a heartbeat timeout)
+the control plane TCP connection will recycle allowing the selection
+algorithm to run again. By default fast datapath is preferred over
+Sleeve.
+
+# Local Bridging
+
+Unlike the weave bridge netdev used by Sleeve, the OVS datapath has no
+inherent bridging capability. Consequently fastdp implements an
+ethernet bridge in addition to the vxlan overlay, maintaining its own
+forwarding database by learning MAC addresses and dispatching
+broadcast, unicast & multicast traffic accordingly.
+
+# Short Peer IDs
+
+The router relies on having access to the source and destination peer
+when deciding how to forward packets. When the Sleeve overlay is in
+use this information is conveyed directly within the encapsulation by
+including the names of the peers in question, a solution not
+accommodated directly by the vxlan wire format. Fortunately vxlan does
+have a twenty four bit segment ID field in the header we can use to
+encode this data - the challenge is to identify peers uniquely with a
+twelve bit identifier instead of the seventeen bytes used by Sleeve.
+In practice this is achieved by peers adopting a random 'short ID' and
+resolving ownership collisions of same via the existing gossip
+mechanism.
+
+# Heartbeats via vxlan
+
+Heartbeats are implemented by sending an encapsulated ethernet frame
+with source and destination MAC addresses of `00:00:00:00:00:00` and a
+payload consisting of the connection UID and total ethernet frame
+length. The vxlan vport miss handler on the receiving side detects the
+all-zero addresses and acknowledges the heartbeat via the TCP control
+channel after validating the connection UID and frame length against
+their expected values.
+
+# PMTU Discovery
+
+The length of the heartbeat ethernet frame is set to the MTU of the
+datapath interface. In the event the vxlan packet is dropped or
+truncated, the heartbeat will not be acknowledged; this lack of
+acknowledgement will cause the peers to fall back to the Sleeve
+overlay, which has a more sophisticated dynamic mechanism for coping
+with low path MTUs.
+
+To avoid triggering this fallback in typical deployments, the datapath
+interface is statically configured with an MTU of 1376 bytes allowing
+it to work with most underlay network provider MTUs, including GCE at
+1460 bytes (the eighty four byte difference accommodates the encrypted vxlan overhead).
+This value can be overridden by setting `WEAVE_MTU` at launch if
+necessary.
+
+# Virtual Ports
+
+There are three kinds of virtual port associated with the weave
+datapath:
+
+* `internal` - one of these is created automatically, named after the
+  datapath. It corresponds to the network interface of the same name
+  that appears on the host when the datapath is created; it is the
+  ingress/egress port for `weave expose`
+* `netdev` - one for each application container, corresponding to the
+  host end of the veth pair
+* `vxlan` - one for each UDP port on which we are listening for vxlan
+  encapsulated packets. Typically there is only one of these, but
+  there may be more in a network in which peers do not all use the same
+  port configuration
+
+# Misses and Flow Creation
+
+As mentioned above, the datapath has no inherent behaviour - any
+ingressing packet which does not match a flow is passed to a userspace
+miss handler. The miss handler is then responsible for a) instructing
+the datapath to take actions for the packet that caused the miss (for
+example by copying it to one or more ports) and optionally b)
+installing flow rules which will allow the datapath to act on similar
+packets in future without invoking the miss handler.
+
+At a high level the fast datapath implementation can be viewed as a
+set of miss handlers that determine what actions the OVS datapath
+should take based on router topology information, together with some
+additional machinery that manages the expiry of resulting flows when
+that state changes.
+
+When a miss handler is invoked, it has two pieces of context: a byte
+array containing the packet that triggered the miss, and a set of
+'flow keys' that have been extracted from the packet by the kernel
+module. The following flow keys are of interest to weave:
+
+* `InPort` - the identifier of the ingress virtual port
+* `Ethernet` - the source and destination MAC addresses of the
+  ethernet frame
+* `Tunnel` - tunnel identifier (see section on short peer IDs above),
+  and source/dest IPv4 addresses. Only present for ingress via a vxlan
+  vport
+
+Crucially, the router must use this information alone to determine
+which actions to take - this allows the specification of a flow which
+matches these keys and instructs the kernel to take action
+automatically in future without further userspace involvement.
+
+Two actions are of interest:
+
+* `Output` - output the packet to the specified vport
+* `SetTunnel` - update the effective vxlan tunnel parameters. Only
+  required prior to output from a vxlan vport
+
+It is possible to have multiple actions in a flow, so the router can
+for example create a single rule that copies broadcast traffic to all
+'local' vports (save the ingress vport obviously) as well as to vxlan
+vports for onward peers as dictated by the routing topology.
+
+Every flow specified by the router has the following characteristics:
+
+* An `InPort` key matching the ingress vport
+* A `Tunnel` key if the ingress vport is of type vxlan
+* An `Ethernet` key matching the source and destination MAC
+* A list of `Output` (for internal and netdev vports) and
+  `SetTunnel`+`Output` (for vxlan vports) actions
+
+Under certain circumstances the miss handler will instruct the
+datapath to execute a set of actions for a packet without creating a
+corresponding flow:
+
+* When broadcasting packets to discover the path to unknown
+  destination MACs. This condition is transient in the overwhelming
+  majority of cases and so the benefit of creating a flow is
+  outweighed by the need to invalidate it once the MAC is learned
+* When the packet also needs to be forwarded to one or more peers via
+  Sleeve. In this case the router needs to handle all subsequent
+  matching packets as there is no OVS flow action to do it without
+  userspace involvement
+
+# Flow Invalidation
+
+Once a flow has been created for a particular combination of keys, the
+miss handler will never again be invoked for matching packets. It is
+therefore extremely important that we detect events which invalidate
+existing flow actions:
+
+* Addition of netdev vports to the datapath
+* Route invalidation (topology change)
+* Short peer ID collision
+
+The response in each case is the same - delete all flows from the
+datapath, allowing them to be recreated taking into account the
+updated state.
+
+In addition to these event based invalidations there is an expiry
+process that executes every five minutes. This process enumerates all
+flows in the datapath, removing any which have not been used since the
+last check; this cleans up:
+
+* Flows referring to netdev vports which have been removed
+* Flows created by forwarders which have been stopped
+* Flows related to MAC addresses which have not communicated recently
+
+Flows keyed on tunnel IPv4 address do not need to be cleared when a
+peer appears to change IP address due to NAT; this will cause a miss
+resulting in a new flow, and the old flow will expire naturally via
+the timer mechanism.
+
@@ -65,14 +65,30 @@ peer for space:
  - it will continue to ask for space until it receives some, or its
    copy of the ring tells it all peers are full in that subnet.

+### Data persistence
+
+Key IPAM data is saved to disk, in a [BoltDB](https://github.com/boltdb/bolt)
+file, stored within a [data volume container](https://docs.docker.com/engine/userguide/containers/dockervolumes/#creating-and-mounting-a-data-volume-container)
+named `weavedb`.
+
+This file is used to persist data for various Weave Net components; for IPAM it contains:
+
+* the division of the IP allocation range amongst peers, and
+* allocation of addresses to containers on the local peer,
+
+so that it is immediately available when the peer restarts.
+
 ### Claiming an address

-If a Weave process is restarted, in most cases it will hear from
-another peer which ranges it used to own, but it needs to know which
-individual IP addresses are assigned to containers in order to avoid
-giving the same address out in subsequent allocation requests. The
-weave script invokes the `claim` command in `populate_ipam` to do
-this.
+If a Weave process is restarted,
+
+- if `weavedb` is present, then it loads persisted IPAM data from there, as
+described, in the previous section;
+- else, it learns from other peers which ranges it used to own.
+
+Weave Net then consults the allocator to know which individual IP addresses
+are assigned to containers, and therefore avoid giving the same address out in
+subsequent allocation requests.

 When the Allocator is told to claim an address, there are four
 possibilities:
@@ -210,11 +226,3 @@ reached the peer that does the `rmpeer`, then we get an inconsistent
 ring.  We could make it safe, up to partitions, by inquiring of every
 live peer what its view of the dead peer is, and making sure we use
 the latest info.)
-
-## Limitations
-
-Nothing is persisted. If one peer restarts it should receive gossip
-from other peers showing what it used to own, and in that case it can
-recover quite well. If, however, there are no peers left alive, or
-this peer cannot establish network communication with those that are,
-then it cannot recover.
@@ -4,9 +4,7 @@
 * Up-to-date versions of `git`, `go` etc
 * Install GitHub release tool `go get github.com/weaveworks/github-release`
 * Create a [github token for
-  github-release](https://help.github.com/articles/creating-an-access-token-for-command-line-use/);
-set and export `$GITHUB_TOKEN` with this value
-* Update all dependencies with `make update`
+  github-release](https://help.github.com/articles/creating-an-access-token-for-command-line-use/); select the `repo` OAuth scope; set and export `$GITHUB_TOKEN` with this value

 ## Release Types

@@ -63,9 +61,10 @@ This has the following effects:
 ## Draft Phase
 ### Push Version Tag Upstream

-First you must push your version tag upstream, so that an associated
-GitHub release may be created:
+First you must push your branch and version tag upstream, so that an
+associated GitHub release may be created:

+    git push git@github.com:weaveworks/weave
    git push git@github.com:weaveworks/weave $TAG

 N.B. if you're testing the release process, push to your fork
@@ -128,6 +127,18 @@ Finally, for **Mainline** releases only:

 * Images tagged `latest` are updated on DockerHub

+### Finish up
+
+* If not on master, merge branch into master and push to GitHub.
+* Close the [milestone](https://github.com/weaveworks/weave/milestones) in GitHub and create the next milestone
+* Update the `#weavenetwork` topic heading on freenode (requires 'chanops' permission)
+* For a mainline release vX.Y.0, create a release branch X.Y from the
+  tag and push to GitHub - this will result in X.Y.0 site docs being
+  published to https://www.weave.works
+* Add the new version of `weave-net` to the checkpoint system at
+  https://checkpoint-api.weave.works/admin
+* File a PR to update the version of the daemonset at
+  https://github.com/kubernetes/kops/tree/master/upup/models/cloudup/resources/addons/networking.weave

 ## Troubleshooting

@@ -9,7 +9,7 @@ The model is that each host has a service that is notified of
 hostnames and weave addresses for containers on the host.  Like IPAM,
 this service is embedded within the router.  It binds to
 the host bridge to answer DNS queries from local containers; for
-anything it can't answer, it uses the infomation in the host's
+anything it can't answer, it uses the information in the host's
 /etc/resolv.conf to query an 'fallback' server.

 The service is comprised of a DNS server, which answers all DNS queries
@@ -33,7 +33,7 @@ are re-broadcast around the cluster.
 The DNS server also listens to the Docker event stream, and removes
 entries for containers when they die.  Entries removed in this way are
 tombstoned, and the tombstone lazily broadcast around the cluster.
-After a short timeout the tombstones are independantly removed from
+After a short timeout the tombstones are independently removed from
 each host.


@@ -0,0 +1,108 @@
+# Overview
+
+# ipsets
+
+The policy controller maintains a number of ipsets which are
+subsequently referred to by the iptables rules used to effect network
+policy specifications. These ipsets are created, modified and
+destroyed automatically in response to Pod, Namespace and
+NetworkPolicy object updates from the k8s API server:
+
+* A `hash:ip` set per namespace, containing the IP addresses of all
+  pods in that namespace
+* A `list:set` per distinct (across all network policies in all
+  namespaces) namespace selector mentioned in a network policy,
+  containing the names of any of the above hash:ip sets whose
+  corresponding namespace labels match the selector
+* A `hash:ip` set for each distinct (within the scope of the
+  containing network policy's namespace) pod selector mentioned in a
+  network policy, containing the IP addresses of all pods in the
+  namespace whose labels match that selector
+
+IPsets are implemented by the kernel module `xt_set`, without which
+weave-npc will not work.
+
+ipset names are generated deterministically from a string
+representation of the corresponding label selector. Because ipset
+names are limited to 31 characters in length, this is done by taking a
+SHA hash of the selector string and then printing that out as a base
+85 string with a "weave-" prefix e.g.:
+
+    weave-k?Z;25^M}|1s7P3|H9i;*;MhG
+
+Because pod selectors are scoped to a namespace, we need to make sure
+that if the same selector definition is used in different namespaces
+that we maintain distinct ipsets. Consequently, for such selectors the
+namespace name is prepended to the label selector string before
+hashing to avoid clashes.
+
+# iptables chains
+
+The policy controller maintains two iptables chains in response to
+changes to pods, namespaces and network policies. One chain contains
+the ingress rules that implement the network policy specifications,
+and the other is used to bypass the ingress rules for namespaces which
+have an ingress isolation policy of `DefaultAllow`.
+
+## Dynamically maintained `WEAVE-NPC-DEFAULT` chain
+
+The policy controller maintains a rule in this chain for every
+namespace whose ingress isolation policy is `DefaultAllow`. The
+purpose of this rule is simply to ACCEPT any traffic destined for such
+namespaces before it reaches the ingress chain.
+
+```
+iptables -A WEAVE-NPC-DEFAULT -m set --match-set $NSIPSET dst -j ACCEPT
+```
+
+## Dynamically maintained `WEAVE-NPC-INGRESS` chain
+
+For each namespace network policy ingress rule peer/port combination:
+
+```
+iptables -A WEAVE-NPC-INGRESS -p $PROTO [-m set --match-set $SRCSET] -m set --match-set $DSTSET --dport $DPORT -j ACCEPT
+```
+
+## Static `WEAVE-NPC` chain
+
+Static configuration:
+
+```
+iptables -A WEAVE-NPC -m state --state RELATED,ESTABLISHED -j ACCEPT
+iptables -A WEAVE-NPC -m state --state NEW -j WEAVE-NPC-DEFAULT
+iptables -A WEAVE-NPC -m state --state NEW -j WEAVE-NPC-INGRESS
+```
+
+# Steering traffic into the policy engine
+
+To direct traffic into the policy engine:
+
+```
+iptables -A FORWARD -o weave -m physdev ! --physdev-out vethwe-bridge -j WEAVE-NPC
+iptables -A FORWARD -o weave -m physdev ! --physdev-out vethwe-bridge -j DROP
+```
+
+Note this only affects traffic which egresses the bridge on a physical
+port which is not the Weave Net router - in other words, it is
+destined for an application container veth. The following traffic is
+affected:
+
+* Traffic bridged between local application containers
+* Traffic bridged from the router to a local application container
+* Traffic originating from the internet destined for nodeports - this
+  is routed via the FORWARD chain to a container pod IP after DNAT
+
+The following traffic is NOT affected:
+
+* Traffic bridged from a local application container to the router
+* Traffic originating from processes in the host network namespace
+  (e.g. kubelet health checks)
+* Traffic routed from an application container to the internet
+
+The above mechanism relies on the kernel module `br_netfilter` being
+loaded and enabled via `/proc/sys/net/bridge/bridge-nf-call-iptables`.
+
+See these resources for helpful context:
+
+* http://ebtables.netfilter.org/br_fw_ia/br_fw_ia.html
+* https://commons.wikimedia.org/wiki/File:Netfilter-packet-flow.svg
@@ -3,6 +3,7 @@ package ipam
 import (
 	"fmt"

+	"github.com/weaveworks/weave/api"
 	"github.com/weaveworks/weave/net/address"
 )

@@ -13,8 +14,9 @@ type allocateResult struct {

 type allocate struct {
 	resultChan       chan<- allocateResult
-	ident            string
-	r                address.Range // Range we are trying to allocate within
+	ident            string       // a container ID, something like "weave:expose", or api.NoContainerID
+	r                address.CIDR // Subnet we are trying to allocate within
+	isContainer      bool         // true if ident is a container ID
 	hasBeenCancelled func() bool
 }

@@ -25,29 +27,36 @@ func (g *allocate) Try(alloc *Allocator) bool {
 		return true
 	}

-	if addr, found := alloc.lookupOwned(g.ident, g.r); found {
-		g.resultChan <- allocateResult{addr, nil}
+	if addrs := alloc.ownedInRange(g.ident, g.r.Range()); len(addrs) > 0 {
+		// If we had heard that this container died, resurrect it
+		delete(alloc.dead, g.ident) // delete is no-op if key not in map
+		g.resultChan <- allocateResult{addrs[0].Addr, nil}
 		return true
 	}

-	if !alloc.universe.Overlaps(g.r) {
-		g.resultChan <- allocateResult{0, fmt.Errorf("range %s out of bounds: %s", g.r, alloc.universe)}
+	if !alloc.universe.Range().Overlaps(g.r.Range()) {
+		g.resultChan <- allocateResult{err: fmt.Errorf("range %s out of bounds: %s", g.r, alloc.universe)}
 		return true
 	}

 	alloc.establishRing()

-	if ok, addr := alloc.space.Allocate(g.r); ok {
+	if ok, addr := alloc.space.Allocate(g.r.HostRange()); ok {
+		// If caller hasn't supplied a unique ID, file it under the IP address
+		// which lets the caller then release the address using DELETE /ip/address
+		if g.ident == api.NoContainerID {
+			g.ident = addr.String()
+		}
 		alloc.debugln("Allocated", addr, "for", g.ident, "in", g.r)
-		alloc.addOwned(g.ident, addr)
+		alloc.addOwned(g.ident, address.MakeCIDR(g.r, addr), g.isContainer)
 		g.resultChan <- allocateResult{addr, nil}
 		return true
 	}

 	// out of space
-	donors := alloc.ring.ChoosePeersToAskForSpace(g.r.Start, g.r.End)
+	donors := alloc.ring.ChoosePeersToAskForSpace(g.r.Addr, g.r.Range().End)
 	for _, donor := range donors {
-		if err := alloc.sendSpaceRequest(donor, g.r); err != nil {
+		if err := alloc.sendSpaceRequest(donor, g.r.Range()); err != nil {
 			alloc.debugln("Problem asking peer", donor, "for space:", err)
 		} else {
 			alloc.debugln("Decided to ask peer", donor, "for space in range", g.r)
@@ -59,7 +68,7 @@ func (g *allocate) Try(alloc *Allocator) bool {
 }

 func (g *allocate) Cancel() {
-	g.resultChan <- allocateResult{0, &errorCancelled{"Allocate", g.ident}}
+	g.resultChan <- allocateResult{err: &errorCancelled{"Allocate", g.ident}}
 }

 func (g *allocate) ForContainer(ident string) bool {
@@ -9,6 +9,7 @@ import (

 	"github.com/stretchr/testify/require"

+	"github.com/weaveworks/weave/api"
 	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
 	"github.com/weaveworks/weave/testing/gossip"
@@ -22,6 +23,18 @@ const (

 func returnFalse() bool { return false }

+func (alloc *Allocator) SimplyAllocate(ident string, cidr address.CIDR) (address.Address, error) {
+	return alloc.Allocate(ident, cidr, true, returnFalse)
+}
+
+func (alloc *Allocator) SimplyClaim(ident string, cidr address.CIDR) error {
+	return alloc.Claim(ident, cidr, true, true, returnFalse)
+}
+
+func (alloc *Allocator) AnonymousClaim(cidr address.CIDR) error {
+	return alloc.Claim(api.NoContainerID, cidr, false, true, returnFalse)
+}
+
 func TestAllocFree(t *testing.T) {
 	const (
 		container1 = "abcdef"
@@ -37,40 +50,54 @@ func TestAllocFree(t *testing.T) {

 	alloc, subnet := makeAllocatorWithMockGossip(t, "01:00:00:01:00:00", universe, 1)
 	defer alloc.Stop()
-	_, cidr1, _ := address.ParseCIDR(subnet1)
-	_, cidr2, _ := address.ParseCIDR(subnet2)
+	cidr1, _ := address.ParseCIDR(subnet1)
+	cidr2, _ := address.ParseCIDR(subnet2)

 	alloc.claimRingForTesting()
-	addr1, err := alloc.Allocate(container1, cidr1.HostRange(), returnFalse)
+	addr1, err := alloc.SimplyAllocate(container1, cidr1)
 	require.NoError(t, err)
 	require.Equal(t, testAddr1, addr1.String(), "address")

-	addr2, err := alloc.Allocate(container1, cidr2.HostRange(), returnFalse)
+	addr2, err := alloc.SimplyAllocate(container1, cidr2)
 	require.NoError(t, err)
 	require.Equal(t, testAddr2, addr2.String(), "address")

+	addrs, err := alloc.Lookup(container1, subnet.Range())
+	require.NoError(t, err)
+	require.Equal(t, []address.CIDR{address.MakeCIDR(cidr1, addr1), address.MakeCIDR(cidr2, addr2)}, addrs)
+
 	// Ask for another address for a different container and check it's different
-	addr1b, _ := alloc.Allocate(container2, cidr1.HostRange(), returnFalse)
+	addr1b, _ := alloc.SimplyAllocate(container2, cidr1)
 	if addr1b.String() == testAddr1 {
 		t.Fatalf("Expected different address but got %s", addr1b.String())
 	}

 	// Ask for the first container again and we should get the same addresses again
-	addr1a, _ := alloc.Allocate(container1, cidr1.HostRange(), returnFalse)
+	addr1a, _ := alloc.SimplyAllocate(container1, cidr1)
 	require.Equal(t, testAddr1, addr1a.String(), "address")
-	addr2a, _ := alloc.Allocate(container1, cidr2.HostRange(), returnFalse)
+	addr2a, _ := alloc.SimplyAllocate(container1, cidr2)
 	require.Equal(t, testAddr2, addr2a.String(), "address")

 	// Now delete the first container, and we should get its addresses back
 	require.NoError(t, alloc.Delete(container1))
-	addr3, _ := alloc.Allocate(container3, cidr1.HostRange(), returnFalse)
+	addr3, _ := alloc.SimplyAllocate(container3, cidr1)
 	require.Equal(t, testAddr1, addr3.String(), "address")
-	addr4, _ := alloc.Allocate(container3, cidr2.HostRange(), returnFalse)
+	addr4, _ := alloc.SimplyAllocate(container3, cidr2)
 	require.Equal(t, testAddr2, addr4.String(), "address")

 	alloc.ContainerDied(container2)
+
+	// Resurrect
+	addr1c, err := alloc.SimplyAllocate(container2, cidr1)
+	require.NoError(t, err)
+	require.Equal(t, addr1b, addr1c, "address")
+
 	alloc.ContainerDied(container3)
-	require.Equal(t, address.Offset(spaceSize), alloc.NumFreeAddresses(subnet))
+	alloc.Encode() // sync up
+	// Move the clock forward and clear out the dead container
+	alloc.actionChan <- func() { alloc.now = func() time.Time { return time.Now().Add(containerDiedTimeout * 2) } }
+	alloc.actionChan <- func() { alloc.removeDeadContainers() }
+	require.Equal(t, address.Count(spaceSize+1), alloc.NumFreeAddresses(subnet.Range()))
 }

 func TestBootstrap(t *testing.T) {
@@ -95,7 +122,7 @@ func TestBootstrap(t *testing.T) {
 	ExpectBroadcastMessage(alloc1, nil) // alloc1 will try to form consensus
 	done := make(chan bool)
 	go func() {
-		alloc1.Allocate("somecontainer", subnet, returnFalse)
+		alloc1.Allocate("somecontainer", subnet, true, returnFalse)
 		done <- true
 	}()
 	time.Sleep(100 * time.Millisecond)
@@ -103,6 +130,7 @@ func TestBootstrap(t *testing.T) {

 	CheckAllExpectedMessagesSent(alloc1, alloc2)

+	// Check that tryPendingOps doesn't cause anything to happen here
 	alloc1.actionChan <- func() { alloc1.tryPendingOps() }
 	AssertNothingSent(t, done)

@@ -115,14 +143,24 @@ func TestBootstrap(t *testing.T) {
 	ExpectBroadcastMessage(alloc1, nil)
 	alloc1.OnGossipBroadcast(alloc2.ourName, alloc2.Encode())

-	// both nodes will get consensus now so initialize the ring
-	ExpectBroadcastMessage(alloc2, nil)
+	CheckAllExpectedMessagesSent(alloc1, alloc2)
+
+	// at this point, both nodes have seen each other, but don't have a valid proposal
+	// so there is another exchange of messages
 	ExpectBroadcastMessage(alloc2, nil)
 	alloc2.OnGossipBroadcast(alloc1.ourName, alloc1.Encode())

+	// we have consensus now so alloc1 will initialize the ring
+	ExpectBroadcastMessage(alloc1, nil)
+	alloc1.OnGossipBroadcast(alloc2.ourName, alloc2.Encode())
+
+	CheckAllExpectedMessagesSent(alloc1, alloc2)
+
+	// alloc2 receives the ring and does not reply
+	alloc2.OnGossipBroadcast(alloc1.ourName, alloc1.Encode())
+
 	CheckAllExpectedMessagesSent(alloc1, alloc2)

-	alloc1.OnGossipBroadcast(alloc2.ourName, alloc2.Encode())
 	// now alloc1 should have space

 	AssertSent(t, done)
@@ -135,43 +173,67 @@ func TestAllocatorClaim(t *testing.T) {
 		container1 = "abcdef"
 		container3 = "b01df00d"
 		universe   = "10.0.3.0/24"
-		testAddr1  = "10.0.3.2"
-		testAddr2  = "10.0.4.2"
+		testAddr1  = "10.0.3.5/24"
+		testAddr2  = "10.0.4.2/24"
+		testAddr4  = "10.0.2.3/23"
+		subnet4    = "10.0.2.0/23"
+		testPre    = "10.0.3.1/24"
 	)

-	allocs, _, subnet := makeNetworkOfAllocators(2, universe)
+	preAddr, _ := address.ParseCIDR(testPre)
+	allocs, router, subnet := makeNetworkOfAllocators(2, universe, []PreClaim{{container1, true, preAddr}})
+	defer stopNetworkOfAllocators(allocs, router)
 	alloc := allocs[1]
-	defer alloc.Stop()
-	addr1, _ := address.ParseIP(testAddr1)
+	addr1, _ := address.ParseCIDR(testAddr1)

 	// First claim should trigger "dunno, I'm going to wait"
-	err := alloc.Claim(container3, addr1, true)
+	err := alloc.SimplyClaim(container3, addr1)
 	require.NoError(t, err)

-	// Do one allocate to ensure paxos is all done
-	alloc.Allocate("unused", subnet, returnFalse)
+	alloc.Prime()
 	// Do an allocate on the other peer, which we will try to claim later
-	addrx, err := allocs[0].Allocate(container1, subnet, returnFalse)
+	addrx, err := allocs[0].Allocate(container1, subnet, true, returnFalse)
+	require.NoError(t, err)
+	// Should not get the address we pre-claimed
+	require.NotEqual(t, addrx, preAddr)
+	router.Flush()

 	// Now try the claim again
-	err = alloc.Claim(container3, addr1, true)
+	err = alloc.SimplyClaim(container3, addr1)
 	require.NoError(t, err)
 	// Check we get this address back if we try an allocate
-	addr3, _ := alloc.Allocate(container3, subnet, returnFalse)
-	require.Equal(t, testAddr1, addr3.String(), "address")
+	addr3, _ := alloc.SimplyAllocate(container3, subnet)
+	require.Equal(t, testAddr1, address.MakeCIDR(subnet, addr3).String(), "address")
 	// one more claim should still work
-	err = alloc.Claim(container3, addr1, true)
+	err = alloc.SimplyClaim(container3, addr1)
 	require.NoError(t, err)
 	// claim for a different container should fail
-	err = alloc.Claim(container1, addr1, true)
+	err = alloc.SimplyClaim(container1, addr1)
 	require.Error(t, err)
+	// claim for an anonymous container, will assume it's the same one
+	err = alloc.AnonymousClaim(addr1)
+	require.NoError(t, err)
 	// claiming the address allocated on the other peer should fail
-	err = alloc.Claim(container1, addrx, true)
+	err = alloc.SimplyClaim(container1, address.MakeCIDR(subnet, addrx))
+	require.Error(t, err, "claiming address allocated on other peer should fail")
+	// claiming the pre-claimed address should fail on both peers
+	err = alloc.SimplyClaim(container3, preAddr)
+	require.Error(t, err, "claiming address allocated on other peer should fail")
+	err = allocs[0].SimplyClaim(container3, preAddr)
 	require.Error(t, err, "claiming address allocated on other peer should fail")
 	// Check an address outside of our universe
-	addr2, _ := address.ParseIP(testAddr2)
-	err = alloc.Claim(container1, addr2, true)
+	addr2, _ := address.ParseCIDR(testAddr2)
+	err = alloc.SimplyClaim(container1, addr2)
 	require.NoError(t, err)
+	err = alloc.SimplyClaim(container3, addr2)
+	require.Error(t, err)
+	// Anonymous claim outside of our universe
+	addr4, _ := address.ParseCIDR(testAddr4)
+	subnet4cidr, _ := address.ParseCIDR(subnet4)
+	require.NoError(t, alloc.AnonymousClaim(addr4))
+	addr5, err := alloc.Allocate(api.NoContainerID, subnet4cidr, false, returnFalse)
+	require.NoError(t, err)
+	require.NotEqual(t, addr4.Start(), addr5)
 }

 func (alloc *Allocator) pause() func() {
@@ -187,16 +249,13 @@ func (alloc *Allocator) pause() func() {
 }

 func TestCancel(t *testing.T) {
-	const (
-		CIDR = "10.0.1.7/26"
-	)
-
+	const cidr = "10.0.4.0/26"
 	router := gossip.NewTestRouter(0.0)

-	alloc1, subnet := makeAllocator("01:00:00:02:00:00", CIDR, 2)
+	alloc1, subnet := makeAllocator("01:00:00:02:00:00", cidr, 2)
 	alloc1.SetInterfaces(router.Connect(alloc1.ourName, alloc1))

-	alloc2, _ := makeAllocator("02:00:00:02:00:00", CIDR, 2)
+	alloc2, _ := makeAllocator("02:00:00:02:00:00", cidr, 2)
 	alloc2.SetInterfaces(router.Connect(alloc2.ourName, alloc2))
 	alloc1.claimRingForTesting(alloc1, alloc2)
 	alloc2.claimRingForTesting(alloc1, alloc2)
@@ -208,9 +267,9 @@ func TestCancel(t *testing.T) {
 	alloc1.OnGossipBroadcast(alloc2.ourName, alloc2.Encode())

 	// Get some IPs, so each allocator has some space
-	res1, _ := alloc1.Allocate("foo", subnet, returnFalse)
+	res1, _ := alloc1.Allocate("foo", subnet, true, returnFalse)
 	common.Log.Debugf("res1 = %s", res1.String())
-	res2, _ := alloc2.Allocate("bar", subnet, returnFalse)
+	res2, _ := alloc2.Allocate("bar", subnet, true, returnFalse)
 	common.Log.Debugf("res2 = %s", res2.String())
 	if res1 == res2 {
 		require.FailNow(t, "Error: got same ips!")
@@ -221,13 +280,13 @@ func TestCancel(t *testing.T) {
 	unpause := alloc2.pause()

 	// Use up all the IPs that alloc1 owns, so the allocation after this will prompt a request to alloc2
-	for i := 0; alloc1.NumFreeAddresses(subnet) > 0; i++ {
-		alloc1.Allocate(fmt.Sprintf("tmp%d", i), subnet, returnFalse)
+	for i := 0; alloc1.NumFreeAddresses(subnet.HostRange()) > 0; i++ {
+		alloc1.Allocate(fmt.Sprintf("tmp%d", i), subnet, true, returnFalse)
 	}
 	cancelChan := make(chan bool, 1)
 	doneChan := make(chan bool)
 	go func() {
-		_, ok := alloc1.Allocate("baz", subnet,
+		_, ok := alloc1.Allocate("baz", subnet, true,
 			func() bool {
 				select {
 				case <-cancelChan:
@@ -251,18 +310,18 @@ func TestCancel(t *testing.T) {

 func TestCancelOnDied(t *testing.T) {
 	const (
-		CIDR       = "10.0.1.7/26"
+		cidr       = "10.0.4.0/26"
 		container1 = "abcdef"
 	)

 	router := gossip.NewTestRouter(0.0)
-	alloc1, subnet := makeAllocator("01:00:00:02:00:00", CIDR, 2)
+	alloc1, subnet := makeAllocator("01:00:00:02:00:00", cidr, 2)
 	alloc1.SetInterfaces(router.Connect(alloc1.ourName, alloc1))
 	alloc1.Start()

 	doneChan := make(chan bool)
 	f := func() {
-		_, ok := alloc1.Allocate(container1, subnet, returnFalse)
+		_, ok := alloc1.Allocate(container1, subnet, true, returnFalse)
 		doneChan <- ok == nil
 	}

@@ -293,62 +352,80 @@ func TestGossipShutdown(t *testing.T) {
 	defer alloc.Stop()

 	alloc.claimRingForTesting()
-	alloc.Allocate(container1, subnet, returnFalse)
+	alloc.SimplyAllocate(container1, subnet)

 	alloc.Shutdown()

-	_, err := alloc.Allocate(container2, subnet, returnFalse) // trying to allocate after shutdown should fail
+	_, err := alloc.SimplyAllocate(container2, subnet) // trying to allocate after shutdown should fail
 	require.False(t, err == nil, "no address")

 	CheckAllExpectedMessagesSent(alloc)
 }

+func TestNoFrag(t *testing.T) {
+	const cidr = "10.0.4.0/22"
+	resultChan := make(chan int)
+	for i := 0; i < 100; i++ {
+		allocs, router, subnet := makeNetworkOfAllocators(3, cidr)
+		allocs[1].Allocate("foo", subnet, true, returnFalse)
+		allocs[2].Allocate("bar", subnet, true, returnFalse)
+		allocs[2].actionChan <- func() {
+			resultChan <- len(allocs[2].ring.Entries)
+		}
+		require.True(t, <-resultChan < 6, "excessive ring fragmentation")
+		stopNetworkOfAllocators(allocs, router)
+	}
+}
+
 func TestTransfer(t *testing.T) {
-	const (
-		cidr = "10.0.1.7/22"
-	)
+	const cidr = "10.0.4.0/22"
 	allocs, router, subnet := makeNetworkOfAllocators(3, cidr)
-	alloc1 := allocs[0]
-	alloc2 := allocs[1]
-	alloc3 := allocs[2] // This will be 'master' and get the first range
+	defer stopNetworkOfAllocators(allocs, router)
+	alloc0 := allocs[0]
+	alloc1 := allocs[1]
+	alloc2 := allocs[2]

-	_, err := alloc2.Allocate("foo", subnet, returnFalse)
+	_, err := alloc1.Allocate("foo", subnet, true, returnFalse)
 	require.True(t, err == nil, "Failed to get address")

-	_, err = alloc3.Allocate("bar", subnet, returnFalse)
+	_, err = alloc2.Allocate("bar", subnet, true, returnFalse)
 	require.True(t, err == nil, "Failed to get address")

+	// simulation of periodic gossip
 	alloc2.gossip.GossipBroadcast(alloc2.Gossip())
 	router.Flush()
-	alloc2.gossip.GossipBroadcast(alloc3.Gossip())
+	alloc1.gossip.GossipBroadcast(alloc1.Gossip())
 	router.Flush()
+
+	free1 := alloc1.NumFreeAddresses(subnet.Range())
+	free2 := alloc2.NumFreeAddresses(subnet.Range())
+
+	router.RemovePeer(alloc1.ourName)
 	router.RemovePeer(alloc2.ourName)
-	router.RemovePeer(alloc3.ourName)
-	alloc2.Stop()
-	alloc3.Stop()
-	router.Flush()
-	require.NoError(t, alloc1.AdminTakeoverRanges(alloc2.ourName.String()))
-	require.NoError(t, alloc1.AdminTakeoverRanges(alloc3.ourName.String()))
-	router.Flush()
-
-	require.Equal(t, address.Offset(1022), alloc1.NumFreeAddresses(subnet))
-
-	_, err = alloc1.Allocate("foo", subnet, returnFalse)
-	require.True(t, err == nil, "Failed to get address")
 	alloc1.Stop()
+	alloc2.Stop()
+	router.Flush()
+
+	require.Equal(t, free1+1, alloc0.AdminTakeoverRanges(alloc1.ourName.String()))
+	require.Equal(t, free2+1, alloc0.AdminTakeoverRanges(alloc2.ourName.String()))
+	router.Flush()
+
+	require.Equal(t, address.Count(1024), alloc0.NumFreeAddresses(subnet.Range()))
+
+	_, err = alloc0.Allocate("foo", subnet, true, returnFalse)
+	require.True(t, err == nil, "Failed to get address")
+	alloc0.Stop()
 }

 func TestFakeRouterSimple(t *testing.T) {
-	const (
-		cidr = "10.0.1.7/22"
-	)
-	allocs, _, subnet := makeNetworkOfAllocators(2, cidr)
-	defer stopNetworkOfAllocators(allocs)
+	const cidr = "10.0.4.0/22"
+	allocs, router, subnet := makeNetworkOfAllocators(2, cidr)
+	defer stopNetworkOfAllocators(allocs, router)

 	alloc1 := allocs[0]
 	//alloc2 := allocs[1]

-	_, err := alloc1.Allocate("foo", subnet, returnFalse)
+	_, err := alloc1.Allocate("foo", subnet, true, returnFalse)
 	require.NoError(t, err, "Failed to get address")
 }

@@ -359,10 +436,10 @@ func TestAllocatorFuzz(t *testing.T) {
 		nodes        = 5
 		maxAddresses = 1000
 		concurrency  = 5
-		cidr         = "10.0.1.7/22"
+		cidr         = "10.0.4.0/22"
 	)
-	allocs, _, subnet := makeNetworkOfAllocators(nodes, cidr)
-	defer stopNetworkOfAllocators(allocs)
+	allocs, router, subnet := makeNetworkOfAllocators(nodes, cidr)
+	defer stopNetworkOfAllocators(allocs, router)

 	// Test state
 	// For each IP issued we store the allocator
@@ -429,7 +506,7 @@ func TestAllocatorFuzz(t *testing.T) {
 		allocIndex := rand.Int31n(nodes)
 		alloc := allocs[allocIndex]
 		//common.Log.Infof("Allocate: asking allocator %d", allocIndex)
-		addr, err := alloc.Allocate(name, subnet, returnFalse)
+		addr, err := alloc.SimplyAllocate(name, subnet)

 		if err != nil {
 			panic(fmt.Sprintf("Could not allocate addr"))
@@ -486,7 +563,7 @@ func TestAllocatorFuzz(t *testing.T) {

 		//common.Log.Infof("Asking for %s (%s) on allocator %d again", res.name, addr, res.alloc)

-		newAddr, _ := alloc.Allocate(res.name, subnet, returnFalse)
+		newAddr, _ := alloc.SimplyAllocate(res.name, subnet)
 		oldAddr, _ := address.ParseIP(addr)
 		if newAddr != oldAddr {
 			panic(fmt.Sprintf("Got different address for repeat request for %s: %s != %s", res.name, newAddr, oldAddr))
@@ -506,8 +583,8 @@ func TestAllocatorFuzz(t *testing.T) {
 		allocIndex := rand.Int31n(nodes)
 		addressIndex := rand.Int31n(int32(subnet.Size()))
 		alloc := allocs[allocIndex]
-		addr := address.Add(subnet.Start, address.Offset(addressIndex))
-		err := alloc.Claim(name, addr, true)
+		addr := address.Add(subnet.Addr, address.Offset(addressIndex))
+		err := alloc.SimplyClaim(name, address.MakeCIDR(subnet, addr))
 		if err == nil {
 			noteAllocation(allocIndex, name, addr)
 		}
@@ -563,14 +640,36 @@ func TestAllocatorFuzz(t *testing.T) {
 	})
 }

-func TestGossipSkew(t *testing.T) {
-	alloc1, _ := makeAllocatorWithMockGossip(t, "01:00:00:01:00:00", "10.0.1.0/22", 2)
-	defer alloc1.Stop()
-	alloc2, _ := makeAllocatorWithMockGossip(t, "02:00:00:02:00:00", "10.0.1.0/22", 2)
-	alloc2.now = func() time.Time { return time.Now().Add(time.Hour * 2) }
-	defer alloc2.Stop()
+func TestSpaceRequest(t *testing.T) {
+	const (
+		container1 = "cont-1"
+		container2 = "cont-2"
+		universe   = "10.32.0.0/12"
+	)
+	allocs, router, subnet := makeNetworkOfAllocators(1, universe)
+	defer stopNetworkOfAllocators(allocs, router)
+	alloc1 := allocs[0]

-	if _, err := alloc1.OnGossipBroadcast(alloc2.ourName, alloc2.Encode()); err == nil {
-		t.Fail()
-	}
+	addr, err := alloc1.Allocate(container1, subnet, true, returnFalse)
+	require.Nil(t, err, "")
+	// free it again so the donation splits the range neatly
+	err = alloc1.Free(container1, addr)
+	require.Nil(t, err, "")
+	require.Equal(t, cidrRanges(universe), alloc1.OwnedRanges(), "")
+
+	// Start a new peer
+	alloc2, _ := makeAllocator("02:00:00:02:00:00", universe, 2)
+	alloc2.SetInterfaces(router.Connect(alloc2.ourName, alloc2))
+	alloc2.Start()
+	defer alloc2.Stop()
+	alloc2.Allocate(container2, subnet, true, returnFalse)
+
+	// Test whether the universe has been split into two equal halves (GH #2009)
+	require.Equal(t, cidrRanges("10.32.0.0/13"), alloc1.OwnedRanges(), "")
+	require.Equal(t, cidrRanges("10.40.0.0/13"), alloc2.OwnedRanges(), "")
+}
+
+func cidrRanges(s string) []address.Range {
+	c, _ := address.ParseCIDR(s)
+	return []address.Range{c.Range()}
 }
@@ -3,16 +3,20 @@ package ipam
 import (
 	"fmt"

+	"github.com/weaveworks/mesh"
+
+	"github.com/weaveworks/weave/api"
 	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

 type claim struct {
 	resultChan       chan<- error
-	ident            string
-	addr             address.Address
-	noErrorOnUnknown bool
+	ident            string       // a container ID, something like "weave:expose", or api.NoContainerID
+	cidr             address.CIDR // single address being claimed
+	isContainer      bool         // true if ident is a container ID
+	noErrorOnUnknown bool         // if false, error or block if we don't know; if true return ok but keep trying
+	hasBeenCancelled func() bool
 }

 // Send an error (or nil for success) back to caller listening on resultChan
@@ -31,33 +35,60 @@ func (c *claim) sendResult(result error) {

 // Try returns true for success (or failure), false if we need to try again later
 func (c *claim) Try(alloc *Allocator) bool {
-	if !alloc.ring.Contains(c.addr) {
-		// Address not within our universe; assume user knows what they are doing
-		alloc.infof("Ignored address %s claimed by %s - not in our universe", c.addr, c.ident)
+	if c.hasBeenCancelled != nil && c.hasBeenCancelled() {
+		c.Cancel()
+		return true
+	}
+
+	addOwned := func() {
+		if c.ident == api.NoContainerID {
+			alloc.addOwned(c.cidr.Addr.String(), c.cidr, c.isContainer)
+		} else {
+			alloc.addOwned(c.ident, c.cidr, c.isContainer)
+		}
+	}
+
+	if !alloc.ring.Contains(c.cidr.Addr) {
+		alloc.infof("Address %s claimed by %s - not in our range", c.cidr, c.ident)
+		previousOwner := alloc.findOwner(c.cidr.Addr)
+		switch {
+		case previousOwner == "":
+			addOwned()
+		case previousOwner == c.ident: // already owned by this ID
+		case c.ident == api.NoContainerID: // already owned by anonymous container
+			// do nothing (no automatic fall-through in Go)
+		case !alloc.dead[previousOwner].IsZero(): // already owned by dead container
+			alloc.removeOwned(previousOwner, c.cidr.Addr)
+			addOwned()
+		default:
+			c.sendResult(fmt.Errorf("address %s already in use by %s", c.cidr, previousOwner))
+		}
 		c.sendResult(nil)
 		return true
 	}

 	alloc.establishRing()

-	switch owner := alloc.ring.Owner(c.addr); owner {
+	// If we had heard that this container died, resurrect it
+	delete(alloc.dead, c.ident) // (delete is no-op if key not in map)
+
+	switch owner := alloc.ring.Owner(c.cidr.Addr); owner {
 	case alloc.ourName:
 		// success
-	case router.UnknownPeerName:
+	case mesh.UnknownPeerName:
 		// If our ring doesn't know, it must be empty.
+		alloc.infof("Claim %s for %s: is in the range %s, but the allocator is not initialized yet; will try later.",
+			c.cidr, c.ident, alloc.universe)
 		if c.noErrorOnUnknown {
-			alloc.infof("Claim %s for %s: address allocator still initializing; will try later.", c.addr, c.ident)
 			c.sendResult(nil)
-		} else {
-			c.sendResult(fmt.Errorf("%s is in the range %s, but the allocator is not initialized yet", c.addr, alloc.universe.AsCIDRString()))
 		}
 		return false
 	default:
-		alloc.debugf("requesting address %s from other peer %s", c.addr, owner)
-		err := alloc.sendSpaceRequest(owner, address.NewRange(c.addr, 1))
+		alloc.debugf("requesting address %s from other peer %s", c.cidr, owner)
+		err := alloc.sendSpaceRequest(owner, address.NewRange(c.cidr.Addr, 1))
 		if err != nil { // can't speak to owner right now
 			if c.noErrorOnUnknown {
-				alloc.infof("Claim %s for %s: %s; will try later.", c.addr, c.ident, err)
+				alloc.infof("Claim %s for %s: %s; will try later.", c.cidr, c.ident, err)
 				c.sendResult(nil)
 			} else { // just tell the user they can't do this.
 				c.deniedBy(alloc, owner)
@@ -67,31 +98,42 @@ func (c *claim) Try(alloc *Allocator) bool {
 	}

 	// We are the owner, check we haven't given it to another container
-	switch existingIdent := alloc.findOwner(c.addr); existingIdent {
-	case "":
-		if err := alloc.space.Claim(c.addr); err == nil {
-			alloc.debugln("Claimed", c.addr, "for", c.ident)
-			alloc.addOwned(c.ident, c.addr)
+	existingIdent := alloc.findOwner(c.cidr.Addr)
+	switch {
+	case existingIdent == "":
+		// Unused address, we try to claim it:
+		if err := alloc.space.Claim(c.cidr.Addr); err == nil {
+			alloc.debugln("Claimed", c.cidr, "for", c.ident)
+			addOwned()
 			c.sendResult(nil)
 		} else {
 			c.sendResult(err)
 		}
-	case c.ident:
+	case (existingIdent == c.ident) || (c.ident == api.NoContainerID && existingIdent == c.cidr.Addr.String()):
 		// same identifier is claiming same address; that's OK
+		alloc.debugln("Re-Claimed", c.cidr, "for", c.ident)
+		c.sendResult(nil)
+	case existingIdent == c.cidr.Addr.String():
+		// Address already allocated via api.NoContainerID name and current ID is a real container ID:
+		c.sendResult(fmt.Errorf("address %s already in use", c.cidr))
+	case c.ident == api.NoContainerID:
+		// We do not know whether this is the same container or another one,
+		// but we also cannot prove otherwise, so we let it reclaim the address:
+		alloc.debugln("Re-Claimed", c.cidr, "for ID", c.ident, "having existing ID as", existingIdent)
 		c.sendResult(nil)
 	default:
 		// Addr already owned by container on this machine
-		c.sendResult(fmt.Errorf("address %s is already owned by %s", c.addr.String(), existingIdent))
+		c.sendResult(fmt.Errorf("address %s is already owned by %s", c.cidr.String(), existingIdent))
 	}
 	return true
 }

-func (c *claim) deniedBy(alloc *Allocator, owner router.PeerName) {
+func (c *claim) deniedBy(alloc *Allocator, owner mesh.PeerName) {
 	name, found := alloc.nicknames[owner]
 	if found {
 		name = " (" + name + ")"
 	}
-	c.sendResult(fmt.Errorf("address %s is owned by other peer %s%s", c.addr.String(), owner, name))
+	c.sendResult(fmt.Errorf("address %s is owned by other peer %s%s", c.cidr.String(), owner, name))
 }

 func (c *claim) Cancel() {
@@ -1,6 +1,7 @@
 package ipam

 import (
+	"encoding/json"
 	"fmt"
 	"net/http"

@@ -16,87 +17,153 @@ func badRequest(w http.ResponseWriter, err error) {
 	common.Log.Warningln("[allocator]:", err.Error())
 }

-func parseCIDR(w http.ResponseWriter, cidrStr string) (address.CIDR, bool) {
-	subnetAddr, cidr, err := address.ParseCIDR(cidrStr)
+func parseCIDR(w http.ResponseWriter, cidrStr string, net bool) (address.CIDR, bool) {
+	var cidr address.CIDR
+	var err error
+	if net {
+		cidr, err = ParseCIDRSubnet(cidrStr)
+	} else {
+		cidr, err = address.ParseCIDR(cidrStr)
+	}
 	if err != nil {
 		badRequest(w, err)
 		return address.CIDR{}, false
 	}
-	if cidr.Start != subnetAddr {
-		badRequest(w, fmt.Errorf("Invalid subnet %s - bits after network prefix are not all zero", cidrStr))
-		return address.CIDR{}, false
-	}
 	return cidr, true
 }

+func writeAddresses(w http.ResponseWriter, cidrs []address.CIDR) {
+	for i, cidr := range cidrs {
+		fmt.Fprint(w, cidr)
+		if i < len(cidrs)-1 {
+			w.Write([]byte{' '})
+		}
+	}
+}
+
+func hasBeenCancelled(dockerCli *docker.Client, closedChan <-chan bool, ident string, checkAlive bool) func() bool {
+	return func() bool {
+		select {
+		case <-closedChan:
+			return true
+		default:
+			res := checkAlive && dockerCli != nil && dockerCli.IsContainerNotRunning(ident)
+			checkAlive = false // we check only once; if the container dies later we learn about that through events
+			return res
+		}
+	}
+}
+
+func cancellationErr(w http.ResponseWriter, err error) bool {
+	if _, ok := err.(*errorCancelled); ok {
+		common.Log.Infoln("[allocator]:", err.Error())
+		fmt.Fprint(w, "cancelled")
+		return true
+	}
+	return false
+}
+
 func (alloc *Allocator) handleHTTPAllocate(dockerCli *docker.Client, w http.ResponseWriter, ident string, checkAlive bool, subnet address.CIDR) {
-	closedChan := w.(http.CloseNotifier).CloseNotify()
-	addr, err := alloc.Allocate(ident, subnet.HostRange(),
-		func() bool {
-			select {
-			case <-closedChan:
-				return true
-			default:
-				res := checkAlive && dockerCli != nil && dockerCli.IsContainerNotRunning(ident)
-				checkAlive = false // we check only once; if the container dies later we learn about that through events
-				return res
-			}
-		})
+	addr, err := alloc.Allocate(ident, subnet, checkAlive,
+		hasBeenCancelled(dockerCli, w.(http.CloseNotifier).CloseNotify(), ident, checkAlive))
 	if err != nil {
-		if _, ok := err.(*errorCancelled); ok { // cancellation is not really an error
-			common.Log.Infoln("[allocator]:", err.Error())
-			fmt.Fprint(w, "cancelled")
-			return
+		if !cancellationErr(w, err) {
+			badRequest(w, err)
 		}
-		badRequest(w, err)
 		return
 	}
-
 	fmt.Fprintf(w, "%s/%d", addr, subnet.PrefixLen)
 }

-// HandleHTTP wires up ipams HTTP endpoints to the provided mux.
-func (alloc *Allocator) HandleHTTP(router *mux.Router, defaultSubnet address.CIDR, dockerCli *docker.Client) {
-	router.Methods("PUT").Path("/ip/{id}/{ip}").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		vars := mux.Vars(r)
-		ident := vars["id"]
-		ipStr := vars["ip"]
-		noErrorOnUnknown := r.FormValue("noErrorOnUnknown") == "true"
-		if ip, err := address.ParseIP(ipStr); err != nil {
-			badRequest(w, err)
-			return
-		} else if err := alloc.Claim(ident, ip, noErrorOnUnknown); err != nil {
+func (alloc *Allocator) handleHTTPClaim(dockerCli *docker.Client, w http.ResponseWriter, ident string, cidr address.CIDR, checkAlive, noErrorOnUnknown bool) {
+	err := alloc.Claim(ident, cidr, checkAlive, noErrorOnUnknown,
+		hasBeenCancelled(dockerCli, w.(http.CloseNotifier).CloseNotify(), ident, checkAlive))
+	if err != nil {
+		if !cancellationErr(w, err) {
 			badRequest(w, fmt.Errorf("Unable to claim: %s", err))
-			return
 		}
+		return
+	}
+	w.WriteHeader(204)
+}

-		w.WriteHeader(204)
+// HandleHTTP wires up ipams HTTP endpoints to the provided mux.
+func (alloc *Allocator) HandleHTTP(router *mux.Router, defaultSubnet address.CIDR, tracker string, dockerCli *docker.Client) {
+	router.Methods("GET").Path("/ipinfo/defaultsubnet").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		fmt.Fprintf(w, "%s", defaultSubnet)
+	})
+
+	router.Methods("PUT").Path("/ip/{id}/{ip}/{prefixlen}").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		vars := mux.Vars(r)
+		if cidr, ok := parseCIDR(w, vars["ip"]+"/"+vars["prefixlen"], false); ok {
+			ident := vars["id"]
+			checkAlive := r.FormValue("check-alive") == "true"
+			noErrorOnUnknown := r.FormValue("noErrorOnUnknown") == "true"
+			alloc.handleHTTPClaim(dockerCli, w, ident, cidr, checkAlive, noErrorOnUnknown)
+		}
+	})
+
+	router.Methods("GET").Path("/ring").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		alloc.Prime()
 	})

 	router.Methods("GET").Path("/ip/{id}/{ip}/{prefixlen}").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		vars := mux.Vars(r)
-		if subnet, ok := parseCIDR(w, vars["ip"]+"/"+vars["prefixlen"]); ok {
-			addr, err := alloc.Lookup(vars["id"], subnet.HostRange())
+		if subnet, ok := parseCIDR(w, vars["ip"]+"/"+vars["prefixlen"], true); ok {
+			cidrs, err := alloc.Lookup(vars["id"], subnet.HostRange())
 			if err != nil {
 				http.NotFound(w, r)
 				return
 			}
-			fmt.Fprintf(w, "%s/%d", addr, subnet.PrefixLen)
+			writeAddresses(w, cidrs)
 		}
 	})

 	router.Methods("GET").Path("/ip/{id}").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		addr, err := alloc.Lookup(mux.Vars(r)["id"], defaultSubnet.HostRange())
+		addrs, err := alloc.Lookup(mux.Vars(r)["id"], defaultSubnet.HostRange())
 		if err != nil {
 			http.NotFound(w, r)
 			return
 		}
-		fmt.Fprintf(w, "%s/%d", addr, defaultSubnet.PrefixLen)
+		writeAddresses(w, addrs)
+	})
+
+	router.Methods("GET").Path("/ip").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		type mapping struct {
+			ContainerID string   `json:"containerid"`
+			Addrs       []string `json:"addrs"`
+		}
+
+		type mappings struct {
+			Owned []mapping `json:"owned"`
+		}
+
+		resultChan := make(chan mappings)
+		alloc.actionChan <- func() {
+			ms := mappings{}
+			for containerid, d := range alloc.owned {
+				m := mapping{
+					ContainerID: containerid,
+					Addrs:       []string{},
+				}
+				for _, addr := range d.Cidrs {
+					m.Addrs = append(m.Addrs, addr.String())
+				}
+				ms.Owned = append(ms.Owned, m)
+			}
+			resultChan <- ms
+		}
+		ms := <-resultChan
+
+		w.Header().Set("Content-Type", "application/json")
+		if err := json.NewEncoder(w).Encode(ms); err != nil {
+			common.Log.Warningln("[allocator]:", err.Error())
+		}
 	})

 	router.Methods("POST").Path("/ip/{id}/{ip}/{prefixlen}").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
 		vars := mux.Vars(r)
-		if subnet, ok := parseCIDR(w, vars["ip"]+"/"+vars["prefixlen"]); ok {
+		if subnet, ok := parseCIDR(w, vars["ip"]+"/"+vars["prefixlen"], true); ok {
 			alloc.handleHTTPAllocate(dockerCli, w, vars["id"], r.FormValue("check-alive") == "true", subnet)
 		}
 	})
@@ -131,18 +198,7 @@ func (alloc *Allocator) HandleHTTP(router *mux.Router, defaultSubnet address.CID
 		w.WriteHeader(204)
 	})

-	router.Methods("DELETE").Path("/peer").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		alloc.Shutdown()
-		w.WriteHeader(204)
-	})
-
-	router.Methods("DELETE").Path("/peer/{id}").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		ident := mux.Vars(r)["id"]
-		if err := alloc.AdminTakeoverRanges(ident); err != nil {
-			badRequest(w, err)
-			return
-		}
-
-		w.WriteHeader(204)
+	router.Methods("GET").Path("/ipinfo/tracker").HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		fmt.Fprintf(w, tracker)
 	})
 }
@@ -14,7 +14,6 @@ import (
 	"github.com/stretchr/testify/require"
 	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
-	wt "github.com/weaveworks/weave/testing"
 )

 func HTTPPost(t *testing.T, url string) string {
@@ -44,7 +43,7 @@ func listenHTTP(alloc *Allocator, subnet address.CIDR) int {
 	router.HandleFunc("/status", func(w http.ResponseWriter, r *http.Request) {
 		io.WriteString(w, fmt.Sprintln(alloc))
 	})
-	alloc.HandleHTTP(router, subnet, nil)
+	alloc.HandleHTTP(router, subnet, "", nil)

 	httpListener, err := net.Listen("tcp", ":0")
 	if err != nil {
@@ -81,7 +80,7 @@ func TestHttp(t *testing.T) {
 	)

 	alloc, _ := makeAllocatorWithMockGossip(t, "08:00:27:01:c3:9a", universe, 1)
-	_, cidr, _ := address.ParseCIDR(universe)
+	cidr, _ := address.ParseCIDR(universe)
 	port := listenHTTP(alloc, cidr)
 	alloc.claimRingForTesting()

@@ -123,7 +122,7 @@ func TestBadHttp(t *testing.T) {

 	alloc, _ := makeAllocatorWithMockGossip(t, "08:00:27:01:c3:9a", testCIDR1, 1)
 	defer alloc.Stop()
-	_, cidr, _ := address.ParseCIDR(testCIDR1)
+	cidr, _ := address.ParseCIDR(testCIDR1)
 	port := listenHTTP(alloc, cidr)

 	alloc.claimRingForTesting()
@@ -145,40 +144,30 @@ func TestBadHttp(t *testing.T) {
 }

 func TestHTTPCancel(t *testing.T) {
-	wt.RunWithTimeout(t, 2*time.Second, func() {
-		impTestHTTPCancel(t)
-	})
-}
-
-func impTestHTTPCancel(t *testing.T) {
 	var (
 		containerID = "deadbeef"
 		testCIDR1   = "10.0.3.0/29"
 	)

-	alloc, _ := makeAllocatorWithMockGossip(t, "08:00:27:01:c3:9a", testCIDR1, 1)
+	// Say quorum=2, so the allocate won't go ahead
+	alloc, _ := makeAllocatorWithMockGossip(t, "08:00:27:01:c3:9a", testCIDR1, 2)
 	defer alloc.Stop()
-	alloc.claimRingForTesting()
-	_, cidr, _ := address.ParseCIDR(testCIDR1)
+	ExpectBroadcastMessage(alloc, nil) // trying to form consensus
+	cidr, _ := address.ParseCIDR(testCIDR1)
 	port := listenHTTP(alloc, cidr)

-	// Stop the alloc so nothing actually works
-	unpause := alloc.pause()
-
 	// Ask the http server for a new address
-	done := make(chan *http.Response)
 	req, _ := http.NewRequest("POST", allocURL(port, testCIDR1, containerID), nil)
+	// On another goroutine, wait for a bit then cancel the request
 	go func() {
-		res, _ := http.DefaultClient.Do(req)
-		done <- res
+		time.Sleep(100 * time.Millisecond)
+		common.Log.Debug("Cancelling allocate")
+		http.DefaultTransport.(*http.Transport).CancelRequest(req)
 	}()

-	time.Sleep(100 * time.Millisecond)
-	fmt.Println("Cancelling allocate")
-	http.DefaultTransport.(*http.Transport).CancelRequest(req)
-	unpause()
-	res := <-done
+	res, _ := http.DefaultClient.Do(req)
 	if res != nil {
-		require.FailNow(t, "Error: Allocate returned non-nil")
+		body, _ := ioutil.ReadAll(res.Body)
+		require.FailNow(t, "Error: Allocate returned non-nil", string(body))
 	}
 }
@@ -0,0 +1,44 @@
+package paxos
+
+type Participant interface {
+	GossipState() GossipState
+	Update(from GossipState) bool
+	SetQuorum(uint)
+	Propose()
+	Think() bool
+	Consensus() (bool, AcceptedValue)
+	IsElector() bool
+}
+
+type Observer struct {
+}
+
+func NewObserver() Participant {
+	return &Observer{}
+}
+
+func (observer *Observer) GossipState() GossipState {
+	return nil
+}
+
+func (observer *Observer) Update(from GossipState) bool {
+	return false
+}
+
+func (observer *Observer) Propose() {
+}
+
+func (observer *Observer) SetQuorum(uint) {
+}
+
+func (observer *Observer) Think() bool {
+	return false
+}
+
+func (observer *Observer) Consensus() (bool, AcceptedValue) {
+	return false, AcceptedValue{}
+}
+
+func (observer *Observer) IsElector() bool {
+	return false
+}
@@ -1,7 +1,7 @@
 package paxos

 import (
-	"github.com/weaveworks/weave/router"
+	"github.com/weaveworks/mesh"
 )

 // The node identifier.  The use of the UID here is important: Paxos
@@ -9,8 +9,8 @@ import (
 // node does not restart and lose its Paxos state but claim to have
 // the same ID.
 type NodeID struct {
-	Name router.PeerName
-	UID  router.PeerUID
+	Name mesh.PeerName
+	UID  mesh.PeerUID
 }

 // note all fields exported in structs so we can Gob them
@@ -39,7 +39,7 @@ func (a ProposalID) valid() bool {
 }

 // For seeding IPAM, the value we want consensus on is a set of peer names
-type Value []router.PeerName
+type Value []mesh.PeerName

 // An AcceptedValue is a Value plus the proposal which originated that
 // Value.  The origin is not essential, but makes comparing
@@ -72,7 +72,7 @@ type Node struct {
 	knows  GossipState
 }

-func NewNode(name router.PeerName, uid router.PeerUID, quorum uint) *Node {
+func NewNode(name mesh.PeerName, uid mesh.PeerUID, quorum uint) *Node {
 	return &Node{
 		id:     NodeID{name, uid},
 		quorum: quorum,
@@ -80,6 +80,10 @@ func NewNode(name router.PeerName, uid router.PeerUID, quorum uint) *Node {
 	}
 }

+func (node *Node) SetQuorum(quorum uint) {
+	node.quorum = quorum
+}
+
 func (node *Node) GossipState() GossipState {
 	return node.knows
 }
@@ -124,6 +128,9 @@ func max(a uint, b uint) uint {
 // simply a matter of gossipping a new proposal that supersedes all
 // others.
 func (node *Node) Propose() {
+	if node.quorum == 0 {
+		panic("Paxos node.Propose() called with no quorum set")
+	}
 	// Find the highest round number around
 	round := uint(0)

@@ -223,7 +230,7 @@ func (node *Node) Think() bool {
 // about.  This is not necessarily all peer names, but it is at least
 // a quorum, and so good enough for seeding the ring.
 func (node *Node) pickValue() Value {
-	val := make([]router.PeerName, len(node.knows))
+	val := make([]mesh.PeerName, len(node.knows))
 	i := 0
 	for id := range node.knows {
 		val[i] = id.Name
@@ -234,6 +241,9 @@ func (node *Node) pickValue() Value {

 // Has a consensus been reached, based on the known claims of other nodes?
 func (node *Node) Consensus() (bool, AcceptedValue) {
+	if node.quorum == 0 {
+		return false, AcceptedValue{}
+	}
 	counts := map[ProposalID]uint{}

 	for _, claims := range node.knows {
@@ -250,14 +260,16 @@ func (node *Node) Consensus() (bool, AcceptedValue) {
 	return false, AcceptedValue{}
 }

+func (node *Node) IsElector() bool {
+	return true
+}
+
 type Status struct {
+	Elector    bool
 	KnownNodes int
 	Quorum     uint
 }

 func NewStatus(node *Node) *Status {
-	if node == nil {
-		return nil
-	}
-	return &Status{len(node.knows), node.quorum}
+	return &Status{true, len(node.knows), node.quorum}
 }
@@ -6,7 +6,7 @@ import (
 	"testing"
 	"time"

-	"github.com/weaveworks/weave/router"
+	"github.com/weaveworks/mesh"
 )

 type TestNode struct {
@@ -97,8 +97,8 @@ func makeRandomModel(params *TestParams, r *rand.Rand, t *testing.T) *Model {
 	}

 	for i := range m.nodes {
-		m.nodes[i].Node = NewNode(router.PeerName(i/2+1),
-			router.PeerUID(r.Int63()), m.quorum)
+		m.nodes[i].Node = NewNode(mesh.PeerName(i/2+1),
+			mesh.PeerUID(r.Int63()), m.quorum)
 		m.nodes[i].Propose()
 	}

@@ -172,8 +172,8 @@ func (m *Model) isolateNode(node *TestNode) {

 // Restart a node
 func (m *Model) restart(node *TestNode) {
-	node.Node = NewNode(router.PeerName(m.nextID),
-		router.PeerUID(m.r.Int63()), m.quorum)
+	node.Node = NewNode(mesh.PeerName(m.nextID),
+		mesh.PeerUID(m.r.Int63()), m.quorum)
 	m.nextID++
 	node.Propose()

@@ -300,14 +300,14 @@ func (m *Model) validate() {
 	for i := range m.nodes {
 		ok, val := m.nodes[i].Consensus()
 		if !ok {
-			//m.dump()
+			m.dump()
 			m.t.Fatal("Node doesn't know about consensus")
 		}

 		firstConsensus := m.nodes[i].firstConsensus
 		if firstConsensus.Origin.valid() &&
 			firstConsensus.Origin != val.Origin {
-			//m.dump()
+			m.dump()
 			m.t.Fatal("Consensus mismatch")
 		}

@@ -0,0 +1,24 @@
+package ipam
+
+type prime struct {
+	resultChan chan<- struct{}
+}
+
+func (c *prime) Try(alloc *Allocator) bool {
+	if !alloc.ring.Empty() {
+		close(c.resultChan)
+		return true
+	}
+
+	alloc.establishRing()
+
+	return false
+}
+
+func (c *prime) Cancel() {
+	close(c.resultChan)
+}
+
+func (c *prime) ForContainer(ident string) bool {
+	return false
+}
@@ -3,17 +3,18 @@ package ring
 import (
 	"sort"

+	"github.com/weaveworks/mesh"
+
 	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

 // Entry represents entries around the ring
 type entry struct {
 	Token   address.Address // The start of this range
-	Peer    router.PeerName // Who owns this range
+	Peer    mesh.PeerName   // Who owns this range
 	Version uint32          // Version of this range
-	Free    address.Offset  // Number of free IPs in this range
+	Free    address.Count   // Number of free IPs in this range
 }

 func (e *entry) Equal(e2 *entry) bool {
@@ -21,7 +22,7 @@ func (e *entry) Equal(e2 *entry) bool {
 		e.Version == e2.Version
 }

-func (e *entry) update(peername router.PeerName, free address.Offset) {
+func (e *entry) update(peername mesh.PeerName, free address.Count) {
 	e.Peer = peername
 	e.Version++
 	e.Free = free
@@ -1,31 +0,0 @@
-package ring
-
-/* Exported variables for monitoring and management */
-
-import (
-	"expvar"
-	"strconv"
-)
-
-var (
-	expRingSize    = expvar.NewMap("ipam.ringSize")
-	expRingEntries = expvar.NewMap("ipam.ringEntries")
-)
-
-type _int int
-
-func (i _int) String() string {
-	return strconv.Itoa(int(i))
-}
-
-type _uint32 uint32
-
-func (i _uint32) String() string {
-	return strconv.FormatUint(uint64(i), 10)
-}
-
-func (r *Ring) updateExportedVariables() {
-	ringName := r.Start.String()
-	expRingSize.Set(ringName, _uint32(r.End-r.Start))
-	expRingEntries.Set(ringName, _int(len(r.Entries)))
-}
@@ -12,17 +12,21 @@ import (
 	"sort"
 	"time"

+	"github.com/weaveworks/mesh"
+
 	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

+type OnUpdate func(prev, curr []address.Range, local bool)
+
 // Ring represents the ring itself
 type Ring struct {
-	Start, End address.Address   // [min, max) tokens in this ring.  Due to wrapping, min == max (effectively)
-	Peer       router.PeerName   // name of peer owning this ring instance
-	Entries    entries           // list of entries sorted by token
-	Seeds      []router.PeerName // peers with which the ring was seeded
+	Start, End address.Address // [min, max) tokens in this ring.  Due to wrapping, min == max (effectively)
+	Peer       mesh.PeerName   // name of peer owning this ring instance
+	Entries    entries         // list of entries sorted by token
+	Seeds      []mesh.PeerName // peers with which the ring was seeded
+	onUpdate   OnUpdate
 }

 func (r *Ring) assertInvariants() {
@@ -37,42 +41,52 @@ var (
 	ErrNotSorted       = errors.New("Ring not sorted")
 	ErrTokenRepeated   = errors.New("Token appears twice in ring")
 	ErrTokenOutOfRange = errors.New("Token is out of range")
-	ErrDifferentSeeds  = errors.New("Received ring was seeded differently from ours!")
-	ErrDifferentRange  = errors.New("Received range differs from ours!")
-	ErrNewerVersion    = errors.New("Received new version for entry I own!")
-	ErrInvalidEntry    = errors.New("Received invalid state update!")
-	ErrEntryInMyRange  = errors.New("Received new entry in my range!")
-	ErrNotFound        = errors.New("No entries for peer found")
+	ErrDifferentSeeds  = errors.New("Received ring was seeded differently from ours")
+	ErrDifferentRange  = errors.New("Received range differs from ours")
 )

+func errInconsistentEntry(mine, theirs *entry) error {
+	return fmt.Errorf("Inconsistent entries for %s: owned by %s but incoming message says %s", mine.Token, mine.Peer, theirs.Peer)
+}
+func errEntryInMyRange(theirs *entry) error {
+	return fmt.Errorf("Peer %s says it owns the IP range from %s, which I think I own", theirs.Peer, theirs.Token)
+}
+func errNewerVersion(mine, theirs *entry) error {
+	return fmt.Errorf("Received update for IP range I own at %s v%d: incoming message says owner %s v%d", mine.Token, mine.Version, theirs.Peer, theirs.Version)
+}
+
 func (r *Ring) checkInvariants() error {
-	if !sort.IsSorted(r.Entries) {
+	return r.checkEntries(r.Entries)
+}
+
+func (r *Ring) checkEntries(entries entries) error {
+	if !sort.IsSorted(entries) {
 		return ErrNotSorted
 	}

 	// Check no token appears twice
 	// We know it's sorted...
-	for i := 1; i < len(r.Entries); i++ {
-		if r.Entries[i-1].Token == r.Entries[i].Token {
+	for i := 1; i < len(entries); i++ {
+		if entries[i-1].Token == entries[i].Token {
 			return ErrTokenRepeated
 		}
 	}

-	if len(r.Entries) == 0 {
+	if len(entries) == 0 {
 		return nil
 	}

 	// Check tokens are in range
-	if r.Entries.entry(0).Token < r.Start {
+	if entries.entry(0).Token < r.Start {
 		return ErrTokenOutOfRange
 	}
-	if r.Entries.entry(-1).Token >= r.End {
+	if entries.entry(-1).Token >= r.End {
 		return ErrTokenOutOfRange
 	}

 	// Check all the freespaces are in range
-	for i, entry := range r.Entries {
-		next := r.Entries.entry(i + 1)
+	for i, entry := range entries {
+		next := entries.entry(i + 1)
 		distance := r.distance(entry.Token, next.Token)

 		if entry.Free > distance {
@@ -83,27 +97,46 @@ func (r *Ring) checkInvariants() error {
 	return nil
 }

+func (r *Ring) trackUpdates() func() {
+	return r.trackUpdatesOfPeer(r.Peer)
+}
+
+func (r *Ring) trackUpdatesOfPeer(peer mesh.PeerName) func() {
+	if r.onUpdate == nil {
+		return func() {}
+	}
+	ranges := r.OwnedRangesOfPeer(peer)
+	return func() { r.onUpdate(ranges, r.OwnedRangesOfPeer(peer), peer == r.Peer) }
+}
+
 // New creates an empty ring belonging to peer.
-func New(start, end address.Address, peer router.PeerName) *Ring {
+func New(start, end address.Address, peer mesh.PeerName, f OnUpdate) *Ring {
 	common.Assert(start < end)

-	ring := &Ring{Start: start, End: end, Peer: peer, Entries: make([]*entry, 0)}
-	ring.updateExportedVariables()
+	ring := &Ring{Start: start, End: end, Peer: peer, Entries: make([]*entry, 0), onUpdate: f}
 	return ring
 }

+func (r *Ring) Restore(other *Ring) {
+	defer r.trackUpdates()()
+
+	onUpdate := r.onUpdate
+	*r = *other
+	r.onUpdate = onUpdate
+}
+
 func (r *Ring) Range() address.Range {
 	return address.Range{Start: r.Start, End: r.End}
 }

 // Returns the distance between two tokens on this ring, dealing
 // with ranges which cross the origin
-func (r *Ring) distance(start, end address.Address) address.Offset {
+func (r *Ring) distance(start, end address.Address) address.Count {
 	if end > start {
-		return address.Offset(end - start)
+		return address.Count(end - start)
 	}

-	return address.Offset((r.End - start) + (end - r.Start))
+	return address.Count((r.End - start) + (end - r.Start))
 }

 // GrantRangeToHost modifies the ring such that range [start, end)
@@ -111,12 +144,12 @@ func (r *Ring) distance(start, end address.Address) address.Offset {
 // Preconditions:
 // - start < end
 // - [start, end) must be owned by the calling peer
-func (r *Ring) GrantRangeToHost(start, end address.Address, peer router.PeerName) {
+func (r *Ring) GrantRangeToHost(start, end address.Address, peer mesh.PeerName) {
 	//fmt.Printf("%s GrantRangeToHost [%v,%v) -> %s\n", r.Peer, start, end, peer)

 	r.assertInvariants()
+	defer r.trackUpdates()()
 	defer r.assertInvariants()
-	defer r.updateExportedVariables()

 	// ----------------- Start of Checks -----------------

@@ -178,43 +211,69 @@ func (r *Ring) GrantRangeToHost(start, end address.Address, peer router.PeerName
 	r.Entries.insert(entry{Token: end, Peer: r.Peer, Free: endFree})
 }

-// Merge the given ring into this ring and return any new ranges added
-func (r *Ring) Merge(gossip Ring) error {
+// Merge the given ring into this ring and indicate whether this ring
+// got updated as a result.
+func (r *Ring) Merge(gossip Ring) (bool, error) {
 	r.assertInvariants()
-	defer r.assertInvariants()
-	defer r.updateExportedVariables()
+	defer r.trackUpdates()()

 	// Don't panic when checking the gossiped in ring.
 	// In this case just return any error found.
 	if err := gossip.checkInvariants(); err != nil {
-		return err
+		return false, err
 	}

 	if len(gossip.Seeds) > 0 && len(r.Seeds) > 0 {
 		if len(gossip.Seeds) != len(r.Seeds) {
-			return ErrDifferentSeeds
+			return false, ErrDifferentSeeds
 		}
 		for i, seed := range gossip.Seeds {
 			if seed != r.Seeds[i] {
-				return ErrDifferentSeeds
+				return false, ErrDifferentSeeds
 			}
 		}
 	}

 	if r.Start != gossip.Start || r.End != gossip.End {
-		return ErrDifferentRange
+		return false, ErrDifferentRange
 	}

-	// Now merge their ring with yours, in a temporary ring.
-	var result entries
-	addToResult := func(e entry) { result = append(result, &e) }
+	result, updated, err := r.Entries.merge(gossip.Entries, r.Peer)

+	if err != nil {
+		return false, err
+	}
+
+	if err := r.checkEntries(result); err != nil {
+		return false, fmt.Errorf("Merge of incoming data causes: %s", err)
+	}
+
+	if len(r.Seeds) == 0 {
+		r.Seeds = gossip.Seeds
+	}
+	r.Entries = result
+
+	return updated, nil
+}
+
+// Merge other entries into ours, and complain when that stomps on
+// entries belonging to ourPeer. Returns the merged entries and an
+// indication whether the merge resulted in any changes, i.e. the
+// result differs from the original.
+func (es entries) merge(other entries, ourPeer mesh.PeerName) (result entries, updated bool, err error) {
 	var mine, theirs *entry
-	var previousOwner *router.PeerName
-	// i is index into r.Entries; j is index into gossip.Entries
+	var previousOwner *mesh.PeerName
+	addToResult := func(e entry) { result = append(result, &e) }
+	addTheirs := func(e entry) {
+		addToResult(e)
+		updated = true
+		previousOwner = nil
+	}
+
+	// i is index into es; j is index into other
 	var i, j int
-	for i < len(r.Entries) && j < len(gossip.Entries) {
-		mine, theirs = r.Entries[i], gossip.Entries[j]
+	for i < len(es) && j < len(other) {
+		mine, theirs = es[i], other[j]
 		switch {
 		case mine.Token < theirs.Token:
 			addToResult(*mine)
@@ -222,56 +281,52 @@ func (r *Ring) Merge(gossip Ring) error {
 			i++
 		case mine.Token > theirs.Token:
 			// insert, checking that a range owned by us hasn't been split
-			if previousOwner != nil && *previousOwner == r.Peer && theirs.Peer != r.Peer {
-				return ErrEntryInMyRange
+			if previousOwner != nil && *previousOwner == ourPeer && theirs.Peer != ourPeer {
+				err = errEntryInMyRange(theirs)
+				return
 			}
-			addToResult(*theirs)
-			previousOwner = nil
+			addTheirs(*theirs)
 			j++
 		case mine.Token == theirs.Token:
 			// merge
 			switch {
 			case mine.Version >= theirs.Version:
 				if mine.Version == theirs.Version && !mine.Equal(theirs) {
-					common.Log.Debugf("Error merging entries at %s - %v != %v", mine.Token, mine, theirs)
-					return ErrInvalidEntry
+					err = errInconsistentEntry(mine, theirs)
+					return
 				}
 				addToResult(*mine)
 				previousOwner = &mine.Peer
 			case mine.Version < theirs.Version:
-				if mine.Peer == r.Peer { // We shouldn't receive updates to our own tokens
-					return ErrNewerVersion
+				if mine.Peer == ourPeer { // We shouldn't receive updates to our own tokens
+					err = errNewerVersion(mine, theirs)
+					return
 				}
-				addToResult(*theirs)
-				previousOwner = nil
+				addTheirs(*theirs)
 			}
 			i++
 			j++
 		}
 	}

-	// At this point, either i is at the end of r or j is at the end
-	// of gossip, so copy over the remaining entries.
+	// At this point, either i is at the end of es or j is at the end
+	// of other, so copy over the remaining entries.

-	for ; i < len(r.Entries); i++ {
-		mine = r.Entries[i]
+	for ; i < len(es); i++ {
+		mine = es[i]
 		addToResult(*mine)
 	}

-	for ; j < len(gossip.Entries); j++ {
-		theirs = gossip.Entries[j]
-		if previousOwner != nil && *previousOwner == r.Peer && theirs.Peer != r.Peer {
-			return ErrEntryInMyRange
+	for ; j < len(other); j++ {
+		theirs = other[j]
+		if previousOwner != nil && *previousOwner == ourPeer && theirs.Peer != ourPeer {
+			err = errEntryInMyRange(theirs)
+			return
 		}
-		addToResult(*theirs)
-		previousOwner = nil
+		addTheirs(*theirs)
 	}

-	if len(r.Seeds) == 0 {
-		r.Seeds = gossip.Seeds
-	}
-	r.Entries = result
-	return nil
+	return
 }

 // Empty returns true if the ring has no entries
@@ -304,10 +359,14 @@ func (r *Ring) splitRangesOverZero(ranges []address.Range) []address.Range {
 // ranges are owned by this peer.  Will split ranges which
 // span 0 in the ring.
 func (r *Ring) OwnedRanges() (result []address.Range) {
+	return r.OwnedRangesOfPeer(r.Peer)
+}
+
+func (r *Ring) OwnedRangesOfPeer(peer mesh.PeerName) (result []address.Range) {
 	r.assertInvariants()

 	for i, entry := range r.Entries {
-		if entry.Peer == r.Peer {
+		if entry.Peer == peer {
 			nextEntry := r.Entries.entry(i + 1)
 			result = append(result, address.Range{Start: entry.Token, End: nextEntry.Token})
 		}
@@ -316,41 +375,58 @@ func (r *Ring) OwnedRanges() (result []address.Range) {
 	return r.splitRangesOverZero(result)
 }

-// ClaimForPeers claims the entire ring for the array of peers passed
-// in.  Only works for empty rings.
-func (r *Ring) ClaimForPeers(peers []router.PeerName) {
-	common.Assert(r.Empty())
-	defer r.assertInvariants()
-	defer r.updateExportedVariables()
+// For printing status
+type RangeInfo struct {
+	Peer mesh.PeerName
+	address.Range
+	Version uint32
+}

-	totalSize := r.distance(r.Start, r.End)
-	share := totalSize/address.Offset(len(peers)) + 1
-	remainder := totalSize % address.Offset(len(peers))
-	pos := r.Start
-
-	for i, peer := range peers {
-		if address.Offset(i) == remainder {
-			share--
-			if share == 0 {
-				break
-			}
+func (r *Ring) AllRangeInfo() (result []RangeInfo) {
+	for i, entry := range r.Entries {
+		nextEntry := r.Entries.entry(i + 1)
+		ranges := []address.Range{{Start: entry.Token, End: nextEntry.Token}}
+		ranges = r.splitRangesOverZero(ranges)
+		for _, r := range ranges {
+			result = append(result, RangeInfo{entry.Peer, r, entry.Version})
 		}
-
-		if e, found := r.Entries.get(pos); found {
-			e.update(peer, share)
-		} else {
-			r.Entries.insert(entry{Token: pos, Peer: peer, Free: share})
-		}
-
-		pos += address.Address(share)
 	}
+	return
+}

-	common.Assert(pos == r.End)
+// ClaimForPeers claims the entire ring for the array of peers passed
+// in.  Only works for empty rings. Each claimed range is CIDR-aligned.
+func (r *Ring) ClaimForPeers(peers []mesh.PeerName) {
+	common.Assert(r.Empty())

+	defer r.trackUpdates()()
+	defer r.assertInvariants()
+	defer func() {
+		e := r.Entries[len(r.Entries)-1]
+		common.Assert(address.Add(e.Token, address.Offset(e.Free)) == r.End)
+	}()
+
+	r.subdivide(r.Start, r.End, peers)
 	r.Seeds = peers
 }

-func (r *Ring) FprintWithNicknames(w io.Writer, m map[router.PeerName]string) {
+// subdivide subdivides the [from,to) CIDR for the given peers into
+// CIDR-aligned subranges.
+func (r *Ring) subdivide(from, to address.Address, peers []mesh.PeerName) {
+	share := address.Length(to, from)
+	if share == 0 {
+		return
+	}
+	if share == 1 || len(peers) == 1 {
+		r.Entries.insert(entry{Token: from, Peer: peers[0], Free: share})
+		return
+	}
+	mid := address.Add(from, address.Offset(share/2))
+	r.subdivide(from, mid, peers[:len(peers)/2])
+	r.subdivide(address.Add(mid, address.Offset(share%2)), to, peers[len(peers)/2:])
+}
+
+func (r *Ring) FprintWithNicknames(w io.Writer, m map[mesh.PeerName]string) {
 	for _, entry := range r.Entries {
 		nickname, found := m[entry.Peer]
 		if found {
@@ -365,17 +441,16 @@ func (r *Ring) FprintWithNicknames(w io.Writer, m map[router.PeerName]string) {
 func (r *Ring) String() string {
 	var buffer bytes.Buffer
 	fmt.Fprintf(&buffer, "Ring [%s, %s)", r.Start, r.End)
-	r.FprintWithNicknames(&buffer, make(map[router.PeerName]string))
+	r.FprintWithNicknames(&buffer, make(map[mesh.PeerName]string))
 	return buffer.String()
 }

 // ReportFree is used by the allocator to tell the ring how many free
 // ips are in a given range, so that ChoosePeersToAskForSpace can make
-// more intelligent decisions.
-func (r *Ring) ReportFree(freespace map[address.Address]address.Offset) {
+// more intelligent decisions.  Returns true if any changes made.
+func (r *Ring) ReportFree(freespace map[address.Address]address.Count) (updated bool) {
 	r.assertInvariants()
 	defer r.assertInvariants()
-	defer r.updateExportedVariables()

 	common.Assert(!r.Empty())
 	entries := r.Entries
@@ -402,20 +477,22 @@ func (r *Ring) ReportFree(freespace map[address.Address]address.Offset) {
 		// Check we're not reporting more space than the range
 		entry, next := entries.entry(i), entries.entry(i+1)
 		maxSize := r.distance(entry.Token, next.Token)
-		common.Assert(free <= maxSize)
+		common.Assert(free <= address.Count(maxSize))

 		if entries[i].Free == free {
-			return
+			continue
 		}

 		entries[i].Free = free
 		entries[i].Version++
+		updated = true
 	}
+	return
 }

 type weightedPeer struct {
 	weight   float64
-	peername router.PeerName
+	peername mesh.PeerName
 }
 type weightedPeers []weightedPeer

@@ -426,11 +503,8 @@ func (ws weightedPeers) Swap(i, j int)      { ws[i], ws[j] = ws[j], ws[i] }

 // ChoosePeersToAskForSpace returns all peers we can ask for space in
 // the range [start, end), in weighted-random order.  Assumes start<end.
-func (r *Ring) ChoosePeersToAskForSpace(start, end address.Address) []router.PeerName {
-	var (
-		sum               address.Offset
-		totalSpacePerPeer = make(map[router.PeerName]address.Offset) // Compute total free space per peer
-	)
+func (r *Ring) ChoosePeersToAskForSpace(start, end address.Address) []mesh.PeerName {
+	totalSpacePerPeer := make(map[mesh.PeerName]address.Count)

 	// iterate through tokens
 	for i, entry := range r.Entries {
@@ -452,7 +526,6 @@ func (r *Ring) ChoosePeersToAskForSpace(start, end address.Address) []router.Pee
 		}

 		totalSpacePerPeer[entry.Peer] += entry.Free
-		sum += entry.Free
 	}

 	// Compute weighted random numbers, then sort.
@@ -463,46 +536,41 @@ func (r *Ring) ChoosePeersToAskForSpace(start, end address.Address) []router.Pee
 		ws = append(ws, weightedPeer{weight: float64(space) * rand.Float64(), peername: peername})
 	}
 	sort.Sort(ws)
-	result := make([]router.PeerName, len(ws))
+	result := make([]mesh.PeerName, len(ws))
 	for i, wp := range ws {
 		result[i] = wp.peername
 	}
 	return result
 }

-func (r *Ring) PickPeerForTransfer(isValid func(router.PeerName) bool) router.PeerName {
+func (r *Ring) PickPeerForTransfer(isValid func(mesh.PeerName) bool) mesh.PeerName {
 	for _, entry := range r.Entries {
 		if entry.Peer != r.Peer && isValid(entry.Peer) {
 			return entry.Peer
 		}
 	}
-	return router.UnknownPeerName
+	return mesh.UnknownPeerName
 }

 // Transfer will mark all entries associated with 'from' peer as owned by 'to' peer
 // and return ranges indicating the new space we picked up
-func (r *Ring) Transfer(from, to router.PeerName) ([]address.Range, error) {
+func (r *Ring) Transfer(from, to mesh.PeerName) []address.Range {
 	r.assertInvariants()
+	defer r.trackUpdates()()
+	defer r.trackUpdatesOfPeer(from)()
 	defer r.assertInvariants()
-	defer r.updateExportedVariables()

 	var newRanges []address.Range
-	found := false

 	for i, entry := range r.Entries {
 		if entry.Peer == from {
-			found = true
 			entry.Peer = to
 			entry.Version++
 			newRanges = append(newRanges, address.Range{Start: entry.Token, End: r.Entries.entry(i + 1).Token})
 		}
 	}

-	if !found {
-		return nil, ErrNotFound
-	}
-
-	return r.splitRangesOverZero(newRanges), nil
+	return r.splitRangesOverZero(newRanges)
 }

 // Contains returns true if addr is in this ring
@@ -511,13 +579,13 @@ func (r *Ring) Contains(addr address.Address) bool {
 }

 // Owner returns the peername which owns the range containing addr
-func (r *Ring) Owner(token address.Address) router.PeerName {
+func (r *Ring) Owner(token address.Address) mesh.PeerName {
 	common.Assert(r.Start <= token && token < r.End)

 	r.assertInvariants()
 	// There can be no owners on an empty ring
 	if r.Empty() {
-		return router.UnknownPeerName
+		return mesh.UnknownPeerName
 	}

 	// Look for the right-most entry, less than or equal to token
@@ -530,8 +598,8 @@ func (r *Ring) Owner(token address.Address) router.PeerName {
 }

 // Get the set of PeerNames mentioned in the ring
-func (r *Ring) PeerNames() map[router.PeerName]struct{} {
-	res := make(map[router.PeerName]struct{})
+func (r *Ring) PeerNames() map[mesh.PeerName]struct{} {
+	res := make(map[mesh.PeerName]struct{})

 	for _, entry := range r.Entries {
 		res[entry.Peer] = struct{}{}
@@ -8,17 +8,19 @@ import (
 	"testing"

 	"github.com/stretchr/testify/require"
+	"github.com/weaveworks/mesh"
+
 	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

 var (
-	peer1name, _ = router.PeerNameFromString("01:00:00:00:02:00")
-	peer2name, _ = router.PeerNameFromString("02:00:00:00:02:00")
-	peer3name, _ = router.PeerNameFromString("03:00:00:00:02:00")
+	peer1name, _ = mesh.PeerNameFromString("01:00:00:00:02:00")
+	peer2name, _ = mesh.PeerNameFromString("02:00:00:00:02:00")
+	peer3name, _ = mesh.PeerNameFromString("03:00:00:00:02:00")

-	start, end    = ParseIP("10.0.0.0"), ParseIP("10.0.0.255")
+	start, end    = ParseIP("10.0.0.0"), ParseIP("10.0.1.0")
+	dot8          = ParseIP("10.0.0.8")
 	dot10, dot245 = ParseIP("10.0.0.10"), ParseIP("10.0.0.245")
 	dot250        = ParseIP("10.0.0.250")
 	middle        = ParseIP("10.0.0.128")
@@ -29,8 +31,17 @@ func ParseIP(s string) address.Address {
 	return addr
 }

+func merge(r1, r2 *Ring) error {
+	_, err := r1.Merge(*r2)
+	return err
+}
+
+func NewRing(start, end address.Address, peer mesh.PeerName) *Ring {
+	return New(start, end, peer, nil)
+}
+
 func TestInvariants(t *testing.T) {
-	ring := New(start, end, peer1name)
+	ring := NewRing(start, end, peer1name)

 	// Check ring is sorted
 	ring.Entries = []*entry{{Token: dot245, Peer: peer1name}, {Token: dot10, Peer: peer2name}}
@@ -41,7 +52,7 @@ func TestInvariants(t *testing.T) {
 	require.True(t, ring.checkInvariants() == ErrTokenRepeated, "Expected error")

 	// Check tokens are in bounds
-	ring = New(dot10, dot245, peer1name)
+	ring = NewRing(dot10, dot245, peer1name)
 	ring.Entries = []*entry{{Token: start, Peer: peer1name}}
 	require.True(t, ring.checkInvariants() == ErrTokenOutOfRange, "Expected error")

@@ -50,7 +61,7 @@ func TestInvariants(t *testing.T) {
 }

 func TestInsert(t *testing.T) {
-	ring := New(start, end, peer1name)
+	ring := NewRing(start, end, peer1name)
 	ring.Entries = []*entry{{Token: start, Peer: peer1name, Free: 255}}

 	require.Panics(t, func() {
@@ -59,17 +70,23 @@ func TestInsert(t *testing.T) {

 	ring.Entries.entry(0).Free = 0
 	ring.Entries.insert(entry{Token: dot245, Peer: peer1name})
-	ring2 := New(start, end, peer1name)
-	ring2.Entries = []*entry{{Token: start, Peer: peer1name, Free: 0}, {Token: dot245, Peer: peer1name}}
-	require.Equal(t, ring2, ring)
+	check := []RangeInfo{
+		{Peer: peer1name, Range: address.Range{Start: start, End: dot245}},
+		{Peer: peer1name, Range: address.Range{Start: dot245, End: end}},
+	}
+	require.Equal(t, check, ring.AllRangeInfo())

 	ring.Entries.insert(entry{Token: dot10, Peer: peer1name})
-	ring2.Entries = []*entry{{Token: start, Peer: peer1name, Free: 0}, {Token: dot10, Peer: peer1name}, {Token: dot245, Peer: peer1name}}
-	require.Equal(t, ring2, ring)
+	check2 := []RangeInfo{
+		{Peer: peer1name, Range: address.Range{Start: start, End: dot10}},
+		{Peer: peer1name, Range: address.Range{Start: dot10, End: dot245}},
+		{Peer: peer1name, Range: address.Range{Start: dot245, End: end}},
+	}
+	require.Equal(t, check2, ring.AllRangeInfo())
 }

 func TestBetween(t *testing.T) {
-	ring1 := New(start, end, peer1name)
+	ring1 := NewRing(start, end, peer1name)
 	ring1.Entries = []*entry{{Token: start, Peer: peer1name, Free: 255}}

 	// First off, in a ring where everything is owned by the peer
@@ -111,29 +128,29 @@ func TestBetween(t *testing.T) {
 }

 func TestGrantSimple(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	// Claim everything for peer1
 	ring1.ClaimItAll()
-	require.Equal(t, entries{{Token: start, Peer: peer1name, Free: 255}}, ring1.Entries)
+	require.Equal(t, entries{{Token: start, Peer: peer1name, Free: 256}}, ring1.Entries)

 	// Now grant everything to peer2
 	ring1.GrantRangeToHost(start, end, peer2name)
-	ring2.Entries = []*entry{{Token: start, Peer: peer2name, Free: 255, Version: 1}}
+	ring2.Entries = []*entry{{Token: start, Peer: peer2name, Free: 256, Version: 1}}
 	require.Equal(t, ring2.Entries, ring1.Entries)

 	// Now spint back to peer 1
 	ring2.GrantRangeToHost(dot10, end, peer1name)
 	ring1.Entries = []*entry{{Token: start, Peer: peer2name, Free: 10, Version: 2},
-		{Token: dot10, Peer: peer1name, Free: 245}}
+		{Token: dot10, Peer: peer1name, Free: 246}}
 	require.Equal(t, ring2.Entries, ring1.Entries)

 	// And spint back to peer 2 again
 	ring1.GrantRangeToHost(dot245, end, peer2name)
 	require.Equal(t, entries{{Token: start, Peer: peer2name, Free: 10, Version: 2},
 		{Token: dot10, Peer: peer1name, Free: 235, Version: 1},
-		{Token: dot245, Peer: peer2name, Free: 10}}, ring1.Entries)
+		{Token: dot245, Peer: peer2name, Free: 11}}, ring1.Entries)

 	// Grant range spanning a live token
 	ring1.Entries = []*entry{{Token: start, Peer: peer1name, Free: 10, Version: 2},
@@ -146,19 +163,19 @@ func TestGrantSimple(t *testing.T) {
 }

 func TestGrantSplit(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	// Claim everything for peer1
-	ring1.Entries = []*entry{{Token: start, Peer: peer1name, Free: 255}}
-	ring2.Merge(*ring1)
+	ring1.Entries = []*entry{{Token: start, Peer: peer1name, Free: 256}}
+	merge(ring2, ring1)
 	require.Equal(t, ring2.Entries, ring1.Entries)

 	// Now grant a split range to peer2
 	ring1.GrantRangeToHost(dot10, dot245, peer2name)
 	require.Equal(t, entries{{Token: start, Peer: peer1name, Free: 10, Version: 1},
 		{Token: dot10, Peer: peer2name, Free: 235},
-		{Token: dot245, Peer: peer1name, Free: 10}}, ring1.Entries)
+		{Token: dot245, Peer: peer1name, Free: 11}}, ring1.Entries)

 	ring1.assertInvariants()

@@ -169,22 +186,22 @@ func TestGrantSplit(t *testing.T) {
 	require.Equal(t, entries{{Token: start, Peer: peer1name, Free: 10, Version: 2},
 		{Token: dot10, Peer: peer2name, Free: 118, Version: 1},
 		{Token: middle, Peer: peer2name, Free: 117, Version: 1},
-		{Token: dot245, Peer: peer1name, Free: 10, Version: 0}}, ring1.Entries)
+		{Token: dot245, Peer: peer1name, Free: 11, Version: 0}}, ring1.Entries)

 	ring1.assertInvariants()
 }

 func TestMergeSimple(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	// Claim everything for peer1
 	ring1.ClaimItAll()
 	ring1.GrantRangeToHost(middle, end, peer2name)
-	require.NoError(t, ring2.Merge(*ring1))
+	require.NoError(t, merge(ring2, ring1))

 	require.Equal(t, entries{{Token: start, Peer: peer1name, Free: 128, Version: 1},
-		{Token: middle, Peer: peer2name, Free: 127}}, ring1.Entries)
+		{Token: middle, Peer: peer2name, Free: 128}}, ring1.Entries)

 	require.Equal(t, ring2.Entries, ring1.Entries)

@@ -193,47 +210,48 @@ func TestMergeSimple(t *testing.T) {
 	ring1.GrantRangeToHost(start, middle, peer2name)
 	ring2.GrantRangeToHost(middle, end, peer1name)

-	require.NoError(t, ring2.Merge(*ring1))
-	require.NoError(t, ring1.Merge(*ring2))
+	require.NoError(t, merge(ring2, ring1))
+	require.NoError(t, merge(ring1, ring2))

 	require.Equal(t, entries{{Token: start, Peer: peer2name, Free: 128, Version: 2},
-		{Token: middle, Peer: peer1name, Free: 127, Version: 1}}, ring1.Entries)
+		{Token: middle, Peer: peer1name, Free: 128, Version: 1}}, ring1.Entries)

 	require.Equal(t, ring2.Entries, ring1.Entries)
 }

 func TestMergeErrors(t *testing.T) {
 	// Cannot Merge in an invalid ring
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)
 	ring2.Entries = []*entry{{Token: middle, Peer: peer2name}, {Token: start, Peer: peer2name}}
-	require.True(t, ring1.Merge(*ring2) == ErrNotSorted, "Expected ErrNotSorted")
+	require.True(t, merge(ring1, ring2) == ErrNotSorted, "Expected ErrNotSorted")

 	// Should Merge two rings for different ranges
-	ring2 = New(start, middle, peer2name)
+	ring2 = NewRing(start, middle, peer2name)
 	ring2.Entries = []*entry{}
-	require.True(t, ring1.Merge(*ring2) == ErrDifferentRange, "Expected ErrDifferentRange")
+	require.True(t, merge(ring1, ring2) == ErrDifferentRange, "Expected ErrDifferentRange")

 	// Cannot Merge newer version of entry I own
-	ring2 = New(start, end, peer2name)
+	ring2 = NewRing(start, end, peer2name)
 	ring1.Entries = []*entry{{Token: start, Peer: peer1name}}
 	ring2.Entries = []*entry{{Token: start, Peer: peer1name, Version: 1}}
-	require.True(t, ring1.Merge(*ring2) == ErrNewerVersion, "Expected ErrNewerVersion")
+	fmt.Println(merge(ring1, ring2))
+	require.Error(t, merge(ring1, ring2), "Expected error")

 	// Cannot Merge two entries with same version but different hosts
 	ring1.Entries = []*entry{{Token: start, Peer: peer1name}}
 	ring2.Entries = []*entry{{Token: start, Peer: peer2name}}
-	require.True(t, ring1.Merge(*ring2) == ErrInvalidEntry, "Expected ErrInvalidEntry")
+	require.Error(t, merge(ring1, ring2), "Expected error")

 	// Cannot Merge an entry into a range I own
 	ring1.Entries = []*entry{{Token: start, Peer: peer1name}}
 	ring2.Entries = []*entry{{Token: middle, Peer: peer2name}}
-	require.True(t, ring1.Merge(*ring2) == ErrEntryInMyRange, "Expected ErrEntryInMyRange")
+	require.Error(t, merge(ring1, ring2), "Expected error")
 }

 func TestMergeMore(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	assertRing := func(ring *Ring, entries entries) {
 		require.Equal(t, entries, ring.Entries)
@@ -244,74 +262,74 @@ func TestMergeMore(t *testing.T) {

 	// Claim everything for peer1
 	ring1.ClaimItAll()
-	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 255}})
+	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 256}})
 	assertRing(ring2, []*entry{})

 	// Check the Merge sends it to the other ring
-	require.NoError(t, ring2.Merge(*ring1))
-	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 255}})
-	assertRing(ring2, []*entry{{Token: start, Peer: peer1name, Free: 255}})
+	require.NoError(t, merge(ring2, ring1))
+	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 256}})
+	assertRing(ring2, []*entry{{Token: start, Peer: peer1name, Free: 256}})

 	// Give everything to peer2
 	ring1.GrantRangeToHost(start, end, peer2name)
-	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 255, Version: 1}})
-	assertRing(ring2, []*entry{{Token: start, Peer: peer1name, Free: 255}})
+	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 256, Version: 1}})
+	assertRing(ring2, []*entry{{Token: start, Peer: peer1name, Free: 256}})

-	require.NoError(t, ring2.Merge(*ring1))
-	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 255, Version: 1}})
-	assertRing(ring2, []*entry{{Token: start, Peer: peer2name, Free: 255, Version: 1}})
+	require.NoError(t, merge(ring2, ring1))
+	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 256, Version: 1}})
+	assertRing(ring2, []*entry{{Token: start, Peer: peer2name, Free: 256, Version: 1}})

 	// And carve off some space
 	ring2.GrantRangeToHost(middle, end, peer1name)
 	assertRing(ring2, []*entry{{Token: start, Peer: peer2name, Free: 128, Version: 2},
-		{Token: middle, Peer: peer1name, Free: 127}})
-	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 255, Version: 1}})
+		{Token: middle, Peer: peer1name, Free: 128}})
+	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 256, Version: 1}})

 	// And Merge back
-	require.NoError(t, ring1.Merge(*ring2))
+	require.NoError(t, merge(ring1, ring2))
 	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 128, Version: 2},
-		{Token: middle, Peer: peer1name, Free: 127}})
+		{Token: middle, Peer: peer1name, Free: 128}})
 	assertRing(ring2, []*entry{{Token: start, Peer: peer2name, Free: 128, Version: 2},
-		{Token: middle, Peer: peer1name, Free: 127}})
+		{Token: middle, Peer: peer1name, Free: 128}})

 	// This should be a no-op
-	require.NoError(t, ring2.Merge(*ring1))
+	require.NoError(t, merge(ring2, ring1))
 	assertRing(ring1, []*entry{{Token: start, Peer: peer2name, Free: 128, Version: 2},
-		{Token: middle, Peer: peer1name, Free: 127}})
+		{Token: middle, Peer: peer1name, Free: 128}})
 	assertRing(ring2, []*entry{{Token: start, Peer: peer2name, Free: 128, Version: 2},
-		{Token: middle, Peer: peer1name, Free: 127}})
+		{Token: middle, Peer: peer1name, Free: 128}})
 }

 func TestMergeSplit(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	// Claim everything for peer2
-	ring1.Entries = []*entry{{Token: start, Peer: peer2name, Free: 255}}
-	require.NoError(t, ring2.Merge(*ring1))
+	ring1.Entries = []*entry{{Token: start, Peer: peer2name, Free: 256}}
+	require.NoError(t, merge(ring2, ring1))
 	require.Equal(t, ring2.Entries, ring1.Entries)

 	// Now grant a split range to peer1
 	ring2.GrantRangeToHost(dot10, dot245, peer1name)
 	require.Equal(t, entries{{Token: start, Peer: peer2name, Free: 10, Version: 1},
 		{Token: dot10, Peer: peer1name, Free: 235},
-		{Token: dot245, Peer: peer2name, Free: 10}}, ring2.Entries)
+		{Token: dot245, Peer: peer2name, Free: 11}}, ring2.Entries)

-	require.NoError(t, ring1.Merge(*ring2))
+	require.NoError(t, merge(ring1, ring2))
 	require.Equal(t, entries{{Token: start, Peer: peer2name, Free: 10, Version: 1},
 		{Token: dot10, Peer: peer1name, Free: 235},
-		{Token: dot245, Peer: peer2name, Free: 10}}, ring1.Entries)
+		{Token: dot245, Peer: peer2name, Free: 11}}, ring1.Entries)

 	require.Equal(t, ring2.Entries, ring1.Entries)
 }

 func TestMergeSplit2(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	// Claim everything for peer2
 	ring1.Entries = []*entry{{Token: start, Peer: peer2name, Free: 250}, {Token: dot250, Peer: peer2name, Free: 5}}
-	require.NoError(t, ring2.Merge(*ring1))
+	require.NoError(t, merge(ring2, ring1))
 	require.Equal(t, ring2.Entries, ring1.Entries)

 	// Now grant a split range to peer1
@@ -320,7 +338,7 @@ func TestMergeSplit2(t *testing.T) {
 		{Token: dot10, Peer: peer1name, Free: 235},
 		{Token: dot245, Peer: peer2name, Free: 5}, {Token: dot250, Peer: peer2name, Free: 5}}, ring2.Entries)

-	require.NoError(t, ring1.Merge(*ring2))
+	require.NoError(t, merge(ring1, ring2))
 	require.Equal(t, entries{{Token: start, Peer: peer2name, Free: 10, Version: 1},
 		{Token: dot10, Peer: peer1name, Free: 235},
 		{Token: dot245, Peer: peer2name, Free: 5}, {Token: dot250, Peer: peer2name, Free: 5}}, ring1.Entries)
@@ -330,8 +348,8 @@ func TestMergeSplit2(t *testing.T) {

 // A simple test, very similar to above, but using the marshalling to byte[]s
 func TestGossip(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	assertRing := func(ring *Ring, entries entries) {
 		require.Equal(t, entries, ring.Entries)
@@ -342,23 +360,23 @@ func TestGossip(t *testing.T) {

 	// Claim everything for peer1
 	ring1.ClaimItAll()
-	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 255}})
+	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 256}})
 	assertRing(ring2, []*entry{})

 	// Check the Merge sends it to the other ring
-	require.NoError(t, ring2.Merge(*ring1))
-	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 255}})
-	assertRing(ring2, []*entry{{Token: start, Peer: peer1name, Free: 255}})
+	require.NoError(t, merge(ring2, ring1))
+	assertRing(ring1, []*entry{{Token: start, Peer: peer1name, Free: 256}})
+	assertRing(ring2, []*entry{{Token: start, Peer: peer1name, Free: 256}})
 }

-func assertPeersWithSpace(t *testing.T, ring *Ring, start, end address.Address, expected int) []router.PeerName {
+func assertPeersWithSpace(t *testing.T, ring *Ring, start, end address.Address, expected int) []mesh.PeerName {
 	peers := ring.ChoosePeersToAskForSpace(start, end)
 	require.Equal(t, expected, len(peers))
 	return peers
 }

 func TestFindFree(t *testing.T) {
-	ring1 := New(start, end, peer1name)
+	ring1 := NewRing(start, end, peer1name)

 	assertPeersWithSpace(t, ring1, start, end, 0)

@@ -366,7 +384,7 @@ func TestFindFree(t *testing.T) {
 	assertPeersWithSpace(t, ring1, start, end, 0)

 	// We shouldn't return outselves
-	ring1.ReportFree(map[address.Address]address.Offset{start: 10})
+	ring1.ReportFree(map[address.Address]address.Count{start: 10})
 	assertPeersWithSpace(t, ring1, start, end, 0)

 	ring1.Entries = []*entry{{Token: start, Peer: peer1name, Free: 1},
@@ -375,7 +393,7 @@ func TestFindFree(t *testing.T) {
 	ring1.assertInvariants()

 	// We should return others
-	var peers []router.PeerName
+	var peers []mesh.PeerName

 	ring1.Entries = []*entry{{Token: start, Peer: peer2name, Free: 1}}
 	peers = assertPeersWithSpace(t, ring1, start, end, 1)
@@ -394,14 +412,14 @@ func TestFindFree(t *testing.T) {
 }

 func TestReportFree(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	ring1.ClaimItAll()
 	ring1.GrantRangeToHost(middle, end, peer2name)
-	require.NoError(t, ring2.Merge(*ring1))
+	require.NoError(t, merge(ring2, ring1))

-	freespace := make(map[address.Address]address.Offset)
+	freespace := make(map[address.Address]address.Count)
 	for _, r := range ring2.OwnedRanges() {
 		freespace[r.Start] = 0
 	}
@@ -409,7 +427,7 @@ func TestReportFree(t *testing.T) {
 }

 func TestMisc(t *testing.T) {
-	ring := New(start, end, peer1name)
+	ring := NewRing(start, end, peer1name)

 	require.True(t, ring.Empty(), "empty")

@@ -418,38 +436,38 @@ func TestMisc(t *testing.T) {
 }

 func TestEmptyGossip(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	ring1.ClaimItAll()
 	// This used to panic, and it shouldn't
-	require.NoError(t, ring1.Merge(*ring2))
+	require.NoError(t, merge(ring1, ring2))
 }

 func TestMergeOldMessage(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	ring1.ClaimItAll()
-	require.NoError(t, ring2.Merge(*ring1))
+	require.NoError(t, merge(ring2, ring1))

 	ring1.GrantRangeToHost(middle, end, peer1name)
-	require.NoError(t, ring1.Merge(*ring2))
+	require.NoError(t, merge(ring1, ring2))
 }

 func TestSplitRangeAtBeginning(t *testing.T) {
-	ring1 := New(start, end, peer1name)
-	ring2 := New(start, end, peer2name)
+	ring1 := NewRing(start, end, peer1name)
+	ring2 := NewRing(start, end, peer2name)

 	ring1.ClaimItAll()
-	require.NoError(t, ring2.Merge(*ring1))
+	require.NoError(t, merge(ring2, ring1))

 	ring1.GrantRangeToHost(start, middle, peer2name)
-	require.NoError(t, ring2.Merge(*ring1))
+	require.NoError(t, merge(ring2, ring1))
 }

 func TestOwnedRange(t *testing.T) {
-	ring1 := New(start, end, peer1name)
+	ring1 := NewRing(start, end, peer1name)
 	ring1.ClaimItAll()

 	require.Equal(t, []address.Range{{Start: start, End: end}}, ring1.OwnedRanges())
@@ -457,8 +475,8 @@ func TestOwnedRange(t *testing.T) {
 	ring1.GrantRangeToHost(middle, end, peer2name)
 	require.Equal(t, []address.Range{{Start: start, End: middle}}, ring1.OwnedRanges())

-	ring2 := New(start, end, peer2name)
-	ring2.Merge(*ring1)
+	ring2 := NewRing(start, end, peer2name)
+	merge(ring2, ring1)
 	require.Equal(t, []address.Range{{Start: middle, End: end}}, ring2.OwnedRanges())

 	ring2.Entries = []*entry{{Token: middle, Peer: peer2name}}
@@ -472,28 +490,28 @@ func TestOwnedRange(t *testing.T) {

 func TestTransfer(t *testing.T) {
 	// First test just checks if we can grant some range to a host, when we transfer it, we get it back
-	ring1 := New(start, end, peer1name)
+	ring1 := NewRing(start, end, peer1name)
 	ring1.ClaimItAll()
 	ring1.GrantRangeToHost(middle, end, peer2name)
 	ring1.Transfer(peer2name, peer1name)
-	require.Equal(t, []address.Range{{start, middle}, {middle, end}}, ring1.OwnedRanges())
+	require.Equal(t, []address.Range{{Start: start, End: middle}, {Start: middle, End: end}}, ring1.OwnedRanges())

 	// Second test is what happens when a token exists at the end of a range but is transferred
 	// - does it get resurrected correctly?
-	ring1 = New(start, end, peer1name)
+	ring1 = NewRing(start, end, peer1name)
 	ring1.ClaimItAll()
 	ring1.GrantRangeToHost(middle, end, peer2name)
 	ring1.Transfer(peer2name, peer1name)
 	ring1.GrantRangeToHost(dot10, middle, peer2name)
-	require.Equal(t, []address.Range{{start, dot10}, {middle, end}}, ring1.OwnedRanges())
+	require.Equal(t, []address.Range{{Start: start, End: dot10}, {Start: middle, End: end}}, ring1.OwnedRanges())
 }

 func TestOwner(t *testing.T) {
-	ring1 := New(start, end, peer1name)
+	ring1 := NewRing(start, end, peer1name)
 	require.True(t, ring1.Contains(start), "start should be in ring")
 	require.False(t, ring1.Contains(end), "end should not be in ring")

-	require.Equal(t, router.UnknownPeerName, ring1.Owner(start))
+	require.Equal(t, mesh.UnknownPeerName, ring1.Owner(start))

 	ring1.ClaimItAll()
 	ring1.GrantRangeToHost(middle, end, peer2name)
@@ -505,6 +523,36 @@ func TestOwner(t *testing.T) {

 }

+func makePeerName(i int) mesh.PeerName {
+	if i >= 10000 {
+		panic("makePeerName: invalid value")
+	}
+	peer, _ := mesh.PeerNameFromString(fmt.Sprintf("%02d:%02d:00:00:00:ff", i/100, i%100))
+	return peer
+}
+
+func makePeers(numPeers int) []mesh.PeerName {
+	peers := make([]mesh.PeerName, numPeers)
+	for i := 0; i < numPeers; i++ {
+		peers[i] = makePeerName(i)
+	}
+	return peers
+}
+
+func TestClaimForPeers(t *testing.T) {
+	const numPeers = 12
+	// Different end to usual so we get a number of addresses that a)
+	// is smaller than the max number of peers, and b) is divisible by
+	// some number of peers. This maximises coverage of edge cases.
+	end := dot8
+	peers := makePeers(numPeers)
+	// Test for a range of peer counts
+	for i := 0; i < numPeers; i++ {
+		ring := NewRing(start, end, peers[0])
+		ring.ClaimForPeers(peers[:i+1])
+	}
+}
+
 type addressSlice []address.Address

 func (s addressSlice) Len() int           { return len(s) }
@@ -517,10 +565,9 @@ func TestFuzzRing(t *testing.T) {
 		iterations = 1000
 	)

-	peers := make([]router.PeerName, numPeers)
+	peers := make([]mesh.PeerName, numPeers)
 	for i := 0; i < numPeers; i++ {
-		peer, _ := router.PeerNameFromString(fmt.Sprintf("%02d:00:00:00:02:00", i))
-		peers[i] = peer
+		peers[i] = makePeerName(i)
 	}

 	// Make a valid, random ring
@@ -539,7 +586,7 @@ func TestFuzzRing(t *testing.T) {
 		sort.Sort(addressSlice(tokens))

 		peer := peers[rand.Intn(len(peers))]
-		ring := New(start, end, peer)
+		ring := NewRing(start, end, peer)
 		for _, token := range tokens {
 			peer = peers[rand.Intn(len(peers))]
 			ring.Entries = append(ring.Entries, &entry{Token: start + token, Peer: peer})
@@ -556,7 +603,7 @@ func TestFuzzRing(t *testing.T) {

 		// Merge them - this might fail, we don't care
 		// We just want to make sure it doesn't panic
-		ring1.Merge(*ring2)
+		merge(ring1, ring2)

 		// Check whats left still passes assertions
 		ring1.assertInvariants()
@@ -573,7 +620,7 @@ func TestFuzzRing(t *testing.T) {
 		}

 		peer := peers[rand.Intn(len(peers))]
-		ring := New(start, end, peer)
+		ring := NewRing(start, end, peer)
 		for _, token := range tokens {
 			peer = peers[rand.Intn(len(peers))]
 			ring.Entries = append(ring.Entries, &entry{Token: start + token, Peer: peer})
@@ -589,7 +636,7 @@ func TestFuzzRing(t *testing.T) {

 		// Merge them - this might fail, we don't care
 		// We just want to make sure it doesn't panic
-		ring1.Merge(*ring2)
+		merge(ring1, ring2)

 		// Check whats left still passes assertions
 		ring1.assertInvariants()
@@ -601,17 +648,17 @@ func TestFuzzRingHard(t *testing.T) {
 	var (
 		numPeers   = 100
 		iterations = 3000
-		peers      []router.PeerName
+		peers      []mesh.PeerName
 		rings      []*Ring
 		nextPeerID = 0
 	)

 	addPeer := func() {
-		peer, _ := router.PeerNameFromString(fmt.Sprintf("%02d:%02d:00:00:00:00", nextPeerID/10, nextPeerID%10))
+		peer := makePeerName(nextPeerID)
 		common.Log.Debugf("%s: Adding peer", peer)
 		nextPeerID++
 		peers = append(peers, peer)
-		rings = append(rings, New(start, end, peer))
+		rings = append(rings, NewRing(start, end, peer))
 	}

 	for i := 0; i < numPeers; i++ {
@@ -620,7 +667,7 @@ func TestFuzzRingHard(t *testing.T) {

 	rings[0].ClaimItAll()

-	randomPeer := func(exclude int) (int, router.PeerName, *Ring) {
+	randomPeer := func(exclude int) (int, mesh.PeerName, *Ring) {
 		var peerIndex int
 		if exclude >= 0 {
 			peerIndex = rand.Intn(len(peers) - 1)
@@ -650,7 +697,7 @@ func TestFuzzRingHard(t *testing.T) {

 		// We need to be in a ~converged ring to rmpeer
 		for _, ring := range rings {
-			require.NoError(t, otherRing.Merge(*ring))
+			require.NoError(t, merge(otherRing, ring))
 		}

 		common.Log.Debugf("%s: transferring from peer %s", otherPeername, peername)
@@ -664,7 +711,7 @@ func TestFuzzRingHard(t *testing.T) {
 		// And now tell everyone about the transfer - rmpeer is
 		// not partition safe
 		for i, ring := range rings {
-			require.NoError(t, ring.Merge(*otherRing))
+			require.NoError(t, merge(ring, otherRing))
 			theRanges[i] = ring.OwnedRanges()
 		}
 	}
@@ -693,7 +740,7 @@ func TestFuzzRingHard(t *testing.T) {
 			// Now 'gossip' this to a random host (note, note could be same host as above)
 			otherIndex, _, otherRing := randomPeer(-1)
 			common.Log.Debugf("%s: 'Gossiping' to %s", ring.Peer, otherRing.Peer)
-			require.NoError(t, otherRing.Merge(*ring))
+			require.NoError(t, merge(otherRing, ring))

 			theRanges[indexWithRanges] = ring.OwnedRanges()
 			theRanges[otherIndex] = otherRing.OwnedRanges()
@@ -712,7 +759,7 @@ func TestFuzzRingHard(t *testing.T) {
 		ring1 := ringsWithEntries[rand.Intn(len(ringsWithEntries))]
 		ring2index, _, ring2 := randomPeer(-1)
 		common.Log.Debugf("%s: 'Gossiping' to %s", ring1.Peer, ring2.Peer)
-		require.NoError(t, ring2.Merge(*ring1))
+		require.NoError(t, merge(ring2, ring1))
 		theRanges[ring2index] = ring2.OwnedRanges()
 	}

@@ -729,7 +776,7 @@ func TestFuzzRingHard(t *testing.T) {
 }

 func (r *Ring) ClaimItAll() {
-	r.ClaimForPeers([]router.PeerName{r.Peer})
+	r.ClaimForPeers([]mesh.PeerName{r.Peer})
 }

 func (es entries) String() string {
@@ -31,8 +31,8 @@ func (s *Space) Add(start address.Address, size address.Offset) {

 // Clear removes all spaces from this space set.  Used during node shutdown.
 func (s *Space) Clear() {
-	s.free = s.free[:0]
 	s.ours = s.ours[:0]
+	s.free = s.free[:0]
 }

 // Walk down the free list calling f() on the in-range portions, until
@@ -88,8 +88,24 @@ func (s *Space) Claim(addr address.Address) error {
 	return nil
 }

-func (s *Space) NumFreeAddressesInRange(r address.Range) address.Offset {
-	res := address.Offset(0)
+func (s *Space) NumOwnedAddresses() address.Count {
+	res := address.Count(0)
+	for i := 0; i < len(s.ours); i += 2 {
+		res += address.Length(s.ours[i+1], s.ours[i])
+	}
+	return res
+}
+
+func (s *Space) NumFreeAddresses() address.Count {
+	res := address.Count(0)
+	for i := 0; i < len(s.free); i += 2 {
+		res += address.Length(s.free[i+1], s.free[i])
+	}
+	return res
+}
+
+func (s *Space) NumFreeAddressesInRange(r address.Range) address.Count {
+	res := address.Count(0)
 	s.walkFree(r, func(chunk address.Range) bool {
 		res += chunk.Size()
 		return false
@@ -111,11 +127,14 @@ func (s *Space) Free(addr address.Address) error {
 }

 func (s *Space) biggestFreeRange(r address.Range) (biggest address.Range) {
-	biggestSize := address.Offset(0)
+	biggestSize := address.Count(0)
 	s.walkFree(r, func(chunk address.Range) bool {
 		if size := chunk.Size(); size >= biggestSize {
-			biggest = chunk
-			biggestSize = size
+			chunk = chunk.BiggestCIDRRange()
+			if size = chunk.Size(); size >= biggestSize {
+				biggest = chunk
+				biggestSize = size
+			}
 		}
 		return false
 	})
@@ -129,9 +148,12 @@ func (s *Space) Donate(r address.Range) (address.Range, bool) {
 		return address.Range{}, false
 	}

-	// Donate half of that biggest free range. Note size/2 rounds down, so
-	// the resulting donation size rounds up, and in particular can't be empty.
-	biggest.Start = address.Add(biggest.Start, biggest.Size()/2)
+	// Donate no more than half what we have available
+	if biggest.Size() > s.NumFreeAddressesInRange(r)/2 {
+		// Note size/2 rounds down, so the resulting donation size
+		// rounds up, and in particular can't be empty.
+		biggest.Start = address.Add(biggest.Start, address.Offset(biggest.Size()/2))
+	}

 	s.ours = subtract(s.ours, biggest.Start, biggest.End)
 	s.free = subtract(s.free, biggest.Start, biggest.End)
@@ -19,16 +19,6 @@ func ip(s string) address.Address {
 	return addr
 }

-// Helper function to avoid 'NumFreeAddressesInRange(start, end)'
-// dozens of times in tests
-func (s *Space) NumFreeAddresses() address.Offset {
-	res := address.Offset(0)
-	for i := 0; i < len(s.free); i += 2 {
-		res += address.Subtract(s.free[i+1], s.free[i])
-	}
-	return res
-}
-
 func TestLowlevel(t *testing.T) {
 	a := []address.Address{}
 	a = add(a, 100, 200)
@@ -61,24 +51,25 @@ func TestLowlevel(t *testing.T) {
 	require.Equal(t, []address.Address{}, a)

 	s := New()
-	require.Equal(t, address.Offset(0), s.NumFreeAddresses())
-	ok, got := s.Allocate(address.NewRange(0, 1000))
+	require.Equal(t, address.Count(0), s.NumFreeAddresses())
+	ok, _ := s.Allocate(address.NewRange(0, 1000))
 	require.False(t, ok, "allocate in empty space should fail")

 	s.Add(100, 100)
-	require.Equal(t, address.Offset(100), s.NumFreeAddresses())
-	ok, got = s.Allocate(address.NewRange(0, 1000))
+	require.Equal(t, address.Count(100), s.NumFreeAddresses())
+	ok, got := s.Allocate(address.NewRange(0, 1000))
 	require.True(t, ok && got == 100, "allocate")
-	require.Equal(t, address.Offset(99), s.NumFreeAddresses())
+	require.Equal(t, address.Count(99), s.NumFreeAddresses())
 	require.NoError(t, s.Claim(150))
-	require.Equal(t, address.Offset(98), s.NumFreeAddresses())
+	require.Equal(t, address.Count(98), s.NumFreeAddresses())
 	require.NoError(t, s.Free(100))
-	require.Equal(t, address.Offset(99), s.NumFreeAddresses())
+	require.Equal(t, address.Count(99), s.NumFreeAddresses())
 	wt.AssertErrorInterface(t, (*error)(nil), s.Free(0), "free not allocated")
 	wt.AssertErrorInterface(t, (*error)(nil), s.Free(100), "double free")

 	r, ok := s.Donate(address.NewRange(0, 1000))
-	require.True(t, ok && r.Start == 125 && r.Size() == 25, "donate")
+	require.True(t, ok, "donate")
+	require.Equal(t, address.NewRange(0xa0, 0x20), r, "donate")

 	// test Donate when addresses are scarce
 	s = New()
@@ -88,7 +79,8 @@ func TestLowlevel(t *testing.T) {
 	require.NoError(t, s.Claim(0))
 	require.NoError(t, s.Claim(2))
 	r, ok = s.Donate(address.NewRange(0, 1000))
-	require.True(t, ok && r.Start == 1 && r.End == 2, "donate")
+	require.True(t, ok, "donate")
+	require.Equal(t, address.NewRange(1, 1), r, "donate")
 	r, ok = s.Donate(address.NewRange(0, 1000))
 	require.True(t, !ok, "donate should fail")
 }
@@ -107,19 +99,22 @@ func TestSpaceAllocate(t *testing.T) {
 	)

 	space1 := makeSpace(start, size)
-	require.Equal(t, address.Offset(20), space1.NumFreeAddresses())
+	require.Equal(t, address.Count(20), space1.NumFreeAddresses())
+	require.Equal(t, address.Count(0), space1.NumOwnedAddresses())
 	space1.assertInvariants()

 	_, addr1 := space1.Allocate(address.NewRange(start, size))
 	require.Equal(t, testAddr1, addr1.String(), "address")
-	require.Equal(t, address.Offset(19), space1.NumFreeAddresses())
+	require.Equal(t, address.Count(19), space1.NumFreeAddresses())
+	require.Equal(t, address.Count(1), space1.NumOwnedAddresses())
 	space1.assertInvariants()

 	_, addr2 := space1.Allocate(address.NewRange(start, size))
 	require.False(t, addr2.String() == testAddr1, "address")
-	require.Equal(t, address.Offset(18), space1.NumFreeAddresses())
-	require.Equal(t, address.Offset(13), space1.NumFreeAddressesInRange(address.Range{Start: ip(testAddr1), End: ip(testAddrx)}))
-	require.Equal(t, address.Offset(18), space1.NumFreeAddressesInRange(address.Range{Start: ip(testAddr1), End: ip(testAddry)}))
+	require.Equal(t, address.Count(18), space1.NumFreeAddresses())
+	require.Equal(t, address.Count(2), space1.NumOwnedAddresses())
+	require.Equal(t, address.Count(13), space1.NumFreeAddressesInRange(address.Range{Start: ip(testAddr1), End: ip(testAddrx)}))
+	require.Equal(t, address.Count(18), space1.NumFreeAddressesInRange(address.Range{Start: ip(testAddr1), End: ip(testAddry)}))
 	space1.assertInvariants()

 	space1.Free(addr2)
@@ -134,20 +129,21 @@ func TestSpaceAllocate(t *testing.T) {

 func TestSpaceFree(t *testing.T) {
 	const (
-		testAddr1   = "10.0.3.4"
+		testAddr1   = "10.0.3.16"
 		testAddrx   = "10.0.3.19"
 		testAddry   = "10.0.9.19"
 		containerID = "deadbeef"
+		size        = 16
 	)

-	entireRange := address.NewRange(ip(testAddr1), 20)
-	space := makeSpace(ip(testAddr1), 20)
+	entireRange := address.NewRange(ip(testAddr1), size)
+	space := makeSpace(ip(testAddr1), size)

 	// Check we are prepared to give up the entire space
 	r := space.biggestFreeRange(entireRange)
-	require.True(t, r.Start == ip(testAddr1) && r.Size() == 20, "Wrong space")
+	require.Equal(t, address.NewRange(ip(testAddr1), size), r, "Wrong space")

-	for i := 0; i < 20; i++ {
+	for i := 0; i < size; i++ {
 		ok, _ := space.Allocate(entireRange)
 		require.True(t, ok, "Failed to get address")
 	}
@@ -155,48 +151,48 @@ func TestSpaceFree(t *testing.T) {
 	// Check we are full
 	ok, _ := space.Allocate(entireRange)
 	require.True(t, !ok, "Should have failed to get address")
-	r, ok = space.Donate(entireRange)
+	r, _ = space.Donate(entireRange)
 	require.True(t, r.Size() == 0, "Wrong space")

 	// Free in the middle
-	require.NoError(t, space.Free(ip("10.0.3.13")))
-	r = space.biggestFreeRange(entireRange)
-	require.True(t, r.Start == ip("10.0.3.13") && r.Size() == 1, "Wrong space")
-
-	// Free one at the end
 	require.NoError(t, space.Free(ip("10.0.3.23")))
 	r = space.biggestFreeRange(entireRange)
 	require.True(t, r.Start == ip("10.0.3.23") && r.Size() == 1, "Wrong space")

+	// Free one at the end
+	require.NoError(t, space.Free(ip("10.0.3.31")))
+	r = space.biggestFreeRange(entireRange)
+	require.True(t, r.Start == ip("10.0.3.31") && r.Size() == 1, "Wrong space")
+
 	// Now free a few at the end
+	require.NoError(t, space.Free(ip("10.0.3.30")))
+	require.NoError(t, space.Free(ip("10.0.3.29")))
+
+	require.Equal(t, address.Count(4), space.NumFreeAddresses())
+
+	// Now get the biggest free space; should be 3.30
+	r = space.biggestFreeRange(entireRange)
+	require.Equal(t, address.NewRange(ip("10.0.3.30"), 2), r, "Wrong space")
+
+	// Now free a few in the middle
+	require.NoError(t, space.Free(ip("10.0.3.24")))
 	require.NoError(t, space.Free(ip("10.0.3.22")))
 	require.NoError(t, space.Free(ip("10.0.3.21")))

-	require.Equal(t, address.Offset(4), space.NumFreeAddresses())
+	require.Equal(t, address.Count(7), space.NumFreeAddresses())

-	// Now get the biggest free space; should be 3.21
+	// Now get the biggest free space; should be 3.30
 	r = space.biggestFreeRange(entireRange)
-	require.True(t, r.Start == ip("10.0.3.21") && r.Size() == 3, "Wrong space")
+	require.Equal(t, address.NewRange(ip("10.0.3.30"), 2), r, "Wrong space")

-	// Now free a few in the middle
-	require.NoError(t, space.Free(ip("10.0.3.12")))
-	require.NoError(t, space.Free(ip("10.0.3.11")))
-	require.NoError(t, space.Free(ip("10.0.3.10")))
-
-	require.Equal(t, address.Offset(7), space.NumFreeAddresses())
-
-	// Now get the biggest free space; should be 3.21
-	r = space.biggestFreeRange(entireRange)
-	require.True(t, r.Start == ip("10.0.3.10") && r.Size() == 4, "Wrong space")
-
-	require.Equal(t, []address.Range{{Start: ip("10.0.3.4"), End: ip("10.0.3.24")}}, space.OwnedRanges())
+	require.Equal(t, []address.Range{{Start: ip("10.0.3.16"), End: ip("10.0.3.32")}}, space.OwnedRanges())
 }

 func TestDonateSimple(t *testing.T) {
 	const (
 		testAddr1 = "10.0.1.0"
 		testAddr2 = "10.0.1.32"
-		size      = 48
+		size      = 64
 	)

 	var (
@@ -209,9 +205,9 @@ func TestDonateSimple(t *testing.T) {
 	r, ok := ps1.Donate(address.NewRange(ip(testAddr1), size))
 	numGivenUp := r.Size()
 	require.True(t, ok, "Donate result")
-	require.Equal(t, "10.0.1.24", r.Start.String(), "Invalid start")
-	require.Equal(t, address.Offset(size/2), numGivenUp)
-	require.Equal(t, address.Offset(size/2), ps1.NumFreeAddresses())
+	require.Equal(t, "10.0.1.32", r.Start.String(), "Invalid start")
+	require.Equal(t, address.Count(size/2), numGivenUp)
+	require.Equal(t, address.Count(size/2), ps1.NumFreeAddresses())

 	// Now check we can give the rest up.
 	count := 0 // count to avoid infinite loop
@@ -222,8 +218,8 @@ func TestDonateSimple(t *testing.T) {
 		}
 		numGivenUp += r.Size()
 	}
-	require.Equal(t, address.Offset(0), ps1.NumFreeAddresses())
-	require.Equal(t, address.Offset(size), numGivenUp)
+	require.Equal(t, address.Count(0), ps1.NumFreeAddresses())
+	require.Equal(t, address.Count(size), numGivenUp)
 }

 func TestDonateHard(t *testing.T) {
@@ -240,7 +236,7 @@ func TestDonateHard(t *testing.T) {
 		require.True(t, ok, "Failed to get IP!")
 	}

-	require.Equal(t, address.Offset(0), spaceset.NumFreeAddresses())
+	require.Equal(t, address.Count(0), spaceset.NumFreeAddresses())

 	// Now free all but the last address
 	// this will force us to split the free list
@@ -251,13 +247,13 @@ func TestDonateHard(t *testing.T) {
 	// Now split
 	newRange, ok := spaceset.Donate(address.NewRange(start, size))
 	require.True(t, ok, "GiveUpSpace result")
-	require.Equal(t, ip("10.0.1.23"), newRange.Start)
-	require.Equal(t, address.Offset(24), newRange.Size())
-	require.Equal(t, address.Offset(23), spaceset.NumFreeAddresses())
+	require.Equal(t, address.NewRange(ip("10.0.1.16"), 16), newRange, "Wrong space")
+	require.Equal(t, address.Count(31), spaceset.NumFreeAddresses())

-	//Space set should now have 2 spaces
-	expected := New()
-	expected.Add(start, 23)
-	expected.ours = add(nil, ip("10.0.1.47"), ip("10.0.1.48"))
+	//Space set should now have 3 spaces
+	expected := &Space{
+		ours: add(nil, ip("10.0.1.47"), ip("10.0.1.48")),
+		free: add(add(nil, ip("10.0.1.0"), ip("10.0.1.16")), ip("10.0.1.32"), ip("10.0.1.47")),
+	}
 	require.Equal(t, expected, spaceset)
 }
@@ -1,6 +1,7 @@
 package ipam

 import (
+	"fmt"
 	"github.com/weaveworks/weave/ipam/paxos"
 	"github.com/weaveworks/weave/net/address"
 )
@@ -8,6 +9,8 @@ import (
 type Status struct {
 	Paxos            *paxos.Status
 	Range            string
+	RangeNumIPs      int
+	ActiveIPs        int
 	DefaultSubnet    string
 	Entries          []EntryStatus
 	PendingClaims    []ClaimStatus
@@ -15,14 +18,17 @@ type Status struct {
 }

 type EntryStatus struct {
-	Token   string
-	Peer    string
-	Version uint32
+	Token       string
+	Size        uint32
+	Peer        string
+	Nickname    string
+	IsKnownPeer bool
+	Version     uint32
 }

 type ClaimStatus struct {
-	Ident   string
-	Address address.Address
+	Ident string
+	CIDR  address.CIDR
 }

 func NewStatus(allocator *Allocator, defaultSubnet address.CIDR) *Status {
@@ -31,8 +37,14 @@ func NewStatus(allocator *Allocator, defaultSubnet address.CIDR) *Status {
 	}

 	var paxosStatus *paxos.Status
-	if allocator.paxosTicker != nil {
-		paxosStatus = paxos.NewStatus(allocator.paxos)
+	if allocator.awaitingConsensus && allocator.paxos != nil {
+		if allocator.paxos.IsElector() {
+			if node, ok := allocator.paxos.(*paxos.Node); ok {
+				paxosStatus = paxos.NewStatus(node)
+			}
+		} else {
+			paxosStatus = &paxos.Status{Elector: false}
+		}
 	}

 	resultChan := make(chan *Status)
@@ -40,6 +52,8 @@ func NewStatus(allocator *Allocator, defaultSubnet address.CIDR) *Status {
 		resultChan <- &Status{
 			paxosStatus,
 			allocator.universe.String(),
+			int(allocator.universe.Size()),
+			int(allocator.space.NumOwnedAddresses()),
 			defaultSubnet.String(),
 			newEntryStatusSlice(allocator),
 			newClaimStatusSlice(allocator),
@@ -56,8 +70,15 @@ func newEntryStatusSlice(allocator *Allocator) []EntryStatus {
 		return slice
 	}

-	for _, entry := range allocator.ring.Entries {
-		slice = append(slice, EntryStatus{entry.Token.String(), entry.Peer.String(), entry.Version})
+	for _, r := range allocator.ring.AllRangeInfo() {
+		slice = append(slice, EntryStatus{
+			Token:       r.Start.String(),
+			Size:        uint32(r.Size()),
+			Peer:        r.Peer.String(),
+			Nickname:    allocator.nicknames[r.Peer],
+			IsKnownPeer: allocator.isKnownPeer(r.Peer),
+			Version:     r.Version,
+		})
 	}

 	return slice
@@ -67,7 +88,7 @@ func newClaimStatusSlice(allocator *Allocator) []ClaimStatus {
 	var slice []ClaimStatus
 	for _, op := range allocator.pendingClaims {
 		claim := op.(*claim)
-		slice = append(slice, ClaimStatus{claim.ident, claim.addr})
+		slice = append(slice, ClaimStatus{claim.ident, claim.cidr})
 	}
 	return slice
 }
@@ -76,7 +97,7 @@ func newAllocateIdentSlice(allocator *Allocator) []string {
 	var slice []string
 	for _, op := range allocator.pendingAllocates {
 		allocate := op.(*allocate)
-		slice = append(slice, allocate.ident)
+		slice = append(slice, fmt.Sprintf("%s %s", allocate.ident, allocate.r.String()))
 	}
 	return slice
 }
@@ -7,14 +7,15 @@ import (
 	"testing"
 	"time"

+	"github.com/weaveworks/mesh"
+
 	"github.com/stretchr/testify/require"
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 	"github.com/weaveworks/weave/testing/gossip"
 )

 type mockMessage struct {
-	dst     router.PeerName
+	dst     mesh.PeerName
 	msgType byte
 	buf     []byte
 }
@@ -48,13 +49,13 @@ func (m *mockGossipComms) String() string {
 // that the contents of messages are never re-ordered.  Which, for instance,
 // requires they are not based off iterating through a map.

-func (m *mockGossipComms) GossipBroadcast(update router.GossipData) error {
+func (m *mockGossipComms) GossipBroadcast(update mesh.GossipData) {
 	m.Lock()
 	defer m.Unlock()
 	buf := []byte{}
 	if len(m.messages) == 0 {
 		require.FailNow(m, fmt.Sprintf("%s: Gossip broadcast message unexpected: \n%x", m.name, buf))
-	} else if msg := m.messages[0]; msg.dst != router.UnknownPeerName {
+	} else if msg := m.messages[0]; msg.dst != mesh.UnknownPeerName {
 		require.FailNow(m, fmt.Sprintf("%s: Expected Gossip message to %s but got broadcast", m.name, msg.dst))
 	} else if msg.buf != nil && !equalByteBuffer(msg.buf, buf) {
 		require.FailNow(m, fmt.Sprintf("%s: Gossip message not sent as expected: \nwant: %x\ngot : %x", m.name, msg.buf, buf))
@@ -62,7 +63,6 @@ func (m *mockGossipComms) GossipBroadcast(update router.GossipData) error {
 		// Swallow this message
 		m.messages = m.messages[1:]
 	}
-	return nil
 }

 func equalByteBuffer(a, b []byte) bool {
@@ -77,12 +77,12 @@ func equalByteBuffer(a, b []byte) bool {
 	return true
 }

-func (m *mockGossipComms) GossipUnicast(dstPeerName router.PeerName, buf []byte) error {
+func (m *mockGossipComms) GossipUnicast(dstPeerName mesh.PeerName, buf []byte) error {
 	m.Lock()
 	defer m.Unlock()
 	if len(m.messages) == 0 {
 		require.FailNow(m, fmt.Sprintf("%s: Gossip message to %s unexpected: \n%s", m.name, dstPeerName, buf))
-	} else if msg := m.messages[0]; msg.dst == router.UnknownPeerName {
+	} else if msg := m.messages[0]; msg.dst == mesh.UnknownPeerName {
 		require.FailNow(m, fmt.Sprintf("%s: Expected Gossip broadcast message but got dest %s", m.name, dstPeerName))
 	} else if msg.dst != dstPeerName {
 		require.FailNow(m, fmt.Sprintf("%s: Expected Gossip message to %s but got dest %s", m.name, msg.dst, dstPeerName))
@@ -99,7 +99,7 @@ func (m *mockGossipComms) GossipUnicast(dstPeerName router.PeerName, buf []byte)

 func ExpectMessage(alloc *Allocator, dst string, msgType byte, buf []byte) {
 	m := alloc.gossip.(*mockGossipComms)
-	dstPeerName, _ := router.PeerNameFromString(dst)
+	dstPeerName, _ := mesh.PeerNameFromString(dst)
 	m.Lock()
 	m.messages = append(m.messages, mockMessage{dstPeerName, msgType, buf})
 	m.Unlock()
@@ -108,7 +108,7 @@ func ExpectMessage(alloc *Allocator, dst string, msgType byte, buf []byte) {
 func ExpectBroadcastMessage(alloc *Allocator, buf []byte) {
 	m := alloc.gossip.(*mockGossipComms)
 	m.Lock()
-	m.messages = append(m.messages, mockMessage{router.UnknownPeerName, 0, buf})
+	m.messages = append(m.messages, mockMessage{mesh.UnknownPeerName, 0, buf})
 	m.Unlock()
 }

@@ -123,24 +123,36 @@ func CheckAllExpectedMessagesSent(allocs ...*Allocator) {
 	}
 }

-func makeAllocator(name string, cidrStr string, quorum uint) (*Allocator, address.Range) {
-	peername, err := router.PeerNameFromString(name)
+type mockDB struct{}
+
+func (d *mockDB) Load(_ string, _ interface{}) (bool, error) { return false, nil }
+func (d *mockDB) Save(_ string, _ interface{}) error         { return nil }
+
+func makeAllocator(name string, cidrStr string, quorum uint, preClaims ...PreClaim) (*Allocator, address.CIDR) {
+	peername, err := mesh.PeerNameFromString(name)
 	if err != nil {
 		panic(err)
 	}

-	_, cidr, err := address.ParseCIDR(cidrStr)
+	cidr, err := address.ParseCIDR(cidrStr)
 	if err != nil {
 		panic(err)
 	}

-	alloc := NewAllocator(peername, router.PeerUID(rand.Int63()),
-		"nick-"+name, cidr.Range(), quorum, func(router.PeerName) bool { return true })
-
-	return alloc, cidr.HostRange()
+	return NewAllocator(Config{
+		OurName:     peername,
+		OurUID:      mesh.PeerUID(rand.Int63()),
+		OurNickname: "nick-" + name,
+		Universe:    cidr,
+		IsObserver:  quorum == 0,
+		PreClaims:   preClaims,
+		Quorum:      func() uint { return quorum },
+		Db:          new(mockDB),
+		IsKnownPeer: func(mesh.PeerName) bool { return true },
+	}), cidr
 }

-func makeAllocatorWithMockGossip(t *testing.T, name string, universeCIDR string, quorum uint) (*Allocator, address.Range) {
+func makeAllocatorWithMockGossip(t *testing.T, name string, universeCIDR string, quorum uint) (*Allocator, address.CIDR) {
 	alloc, subnet := makeAllocator(name, universeCIDR, quorum)
 	gossip := &mockGossipComms{T: t, name: name}
 	alloc.SetInterfaces(gossip)
@@ -149,7 +161,7 @@ func makeAllocatorWithMockGossip(t *testing.T, name string, universeCIDR string,
 }

 func (alloc *Allocator) claimRingForTesting(allocs ...*Allocator) {
-	peers := []router.PeerName{alloc.ourName}
+	peers := []mesh.PeerName{alloc.ourName}
 	for _, alloc2 := range allocs {
 		peers = append(peers, alloc2.ourName)
 	}
@@ -157,14 +169,22 @@ func (alloc *Allocator) claimRingForTesting(allocs ...*Allocator) {
 	alloc.space.AddRanges(alloc.ring.OwnedRanges())
 }

-func (alloc *Allocator) NumFreeAddresses(r address.Range) address.Offset {
-	resultChan := make(chan address.Offset)
+func (alloc *Allocator) NumFreeAddresses(r address.Range) address.Count {
+	resultChan := make(chan address.Count)
 	alloc.actionChan <- func() {
 		resultChan <- alloc.space.NumFreeAddressesInRange(r)
 	}
 	return <-resultChan
 }

+func (alloc *Allocator) OwnedRanges() (result []address.Range) {
+	resultChan := make(chan []address.Range)
+	alloc.actionChan <- func() {
+		resultChan <- alloc.ring.OwnedRanges()
+	}
+	return <-resultChan
+}
+
 // Check whether or not something was sent on a channel
 func AssertSent(t *testing.T, ch <-chan bool) {
 	timeout := time.After(10 * time.Second)
@@ -194,15 +214,19 @@ func AssertNothingSentErr(t *testing.T, ch <-chan error) {
 	}
 }

-func makeNetworkOfAllocators(size int, cidr string) ([]*Allocator, *gossip.TestRouter, address.Range) {
+func makeNetworkOfAllocators(size int, cidr string, preClaims ...[]PreClaim) ([]*Allocator, *gossip.TestRouter, address.CIDR) {
 	gossipRouter := gossip.NewTestRouter(0.0)
 	allocs := make([]*Allocator, size)
-	var subnet address.Range
+	var subnet address.CIDR

 	for i := 0; i < size; i++ {
 		var alloc *Allocator
+		preClaim := []PreClaim{}
+		if i < len(preClaims) {
+			preClaim = preClaims[i]
+		}
 		alloc, subnet = makeAllocator(fmt.Sprintf("%02d:00:00:02:00:00", i),
-			cidr, uint(size/2+1))
+			cidr, uint(size/2+1), preClaim...)
 		alloc.SetInterfaces(gossipRouter.Connect(alloc.ourName, alloc))
 		alloc.Start()
 		allocs[i] = alloc
@@ -213,7 +237,10 @@ func makeNetworkOfAllocators(size int, cidr string) ([]*Allocator, *gossip.TestR
 	return allocs, gossipRouter, subnet
 }

-func stopNetworkOfAllocators(allocs []*Allocator) {
+func stopNetworkOfAllocators(allocs []*Allocator, gossipRouter *gossip.TestRouter) {
+	// NB: We must stop the router first since ipam gossip makes
+	// synchronous calls into the allocator.
+	gossipRouter.Stop()
 	for _, alloc := range allocs {
 		alloc.Stop()
 	}
@@ -0,0 +1,284 @@
+package tracker
+
+// The AWSVPC tracker tracks the IPAM ring changes and accordingly updates
+// the AWS VPC route table and, if the announced change is local
+// (denoted by local=true) to the host, the host route table.
+//
+// During the initialization, the tracker detects AWS VPC route table id which
+// is associated with the host's subnet on the AWS network. If such a table does
+// not exist, the default VPC route table is used.
+//
+// When a host A donates a range to a host B, the necessary route table
+// updates (removal) happen on the host A first, and afterwards on
+// the host B (installation).
+//
+// NB: there is a hard limit for 50 routes within any VPC route table
+// (practically, it is 48, because one route is used by the AWS Internet GW and
+// one by the AWS host subnet), therefore it is suggested to avoid
+// an excessive fragmentation within the IPAM ring which might happen due to
+// the claim operations or uneven distribution of containers across the hosts.
+
+import (
+	"fmt"
+	"net"
+	"syscall"
+
+	"github.com/aws/aws-sdk-go/aws"
+	"github.com/aws/aws-sdk-go/aws/ec2metadata"
+	"github.com/aws/aws-sdk-go/aws/session"
+	"github.com/aws/aws-sdk-go/service/ec2"
+	"github.com/vishvananda/netlink"
+
+	"github.com/weaveworks/weave/common"
+	"github.com/weaveworks/weave/net/address"
+)
+
+type AWSVPCTracker struct {
+	ec2          *ec2.EC2
+	instanceID   string // EC2 Instance ID
+	routeTableID string // VPC Route Table ID
+	linkIndex    int    // The weave bridge link index
+}
+
+// NewAWSVPCTracker creates and initialises AWS VPC based tracker.
+func NewAWSVPCTracker(bridgeName string) (*AWSVPCTracker, error) {
+	var (
+		err     error
+		session = session.New()
+		t       = &AWSVPCTracker{}
+	)
+
+	// Detect region and instance id
+	meta := ec2metadata.New(session)
+	t.instanceID, err = meta.GetMetadata("instance-id")
+	if err != nil {
+		return nil, fmt.Errorf("cannot detect instance-id: %s", err)
+	}
+	region, err := meta.Region()
+	if err != nil {
+		return nil, fmt.Errorf("cannot detect region: %s", err)
+	}
+
+	t.ec2 = ec2.New(session, aws.NewConfig().WithRegion(region))
+
+	routeTableID, err := t.detectRouteTableID()
+	if err != nil {
+		return nil, err
+	}
+	t.routeTableID = *routeTableID
+
+	// Detect Weave bridge link index
+	link, err := netlink.LinkByName(bridgeName)
+	if err != nil {
+		return nil, fmt.Errorf("cannot find \"%s\" interface: %s", bridgeName, err)
+	}
+	t.linkIndex = link.Attrs().Index
+
+	t.infof("AWSVPC has been initialized on %s instance for %s route table at %s region",
+		t.instanceID, t.routeTableID, region)
+
+	return t, nil
+}
+
+// HandleUpdate method updates the AWS VPC and the host route tables.
+func (t *AWSVPCTracker) HandleUpdate(prevRanges, currRanges []address.Range, local bool) error {
+	t.debugf("replacing %q by %q; local(%t)", prevRanges, currRanges, local)
+
+	prev, curr := removeCommon(address.NewCIDRs(merge(prevRanges)), address.NewCIDRs(merge(currRanges)))
+
+	// It might make sense to do the removal first and then add entries
+	// because of the 50 routes limit. However, in such case a container might
+	// not be reachable for a short period of time which is not a desired behavior.
+
+	// Add new entries
+	for _, cidr := range curr {
+		cidrStr := cidr.String()
+		t.debugf("adding route %s to %s", cidrStr, t.instanceID)
+		if _, err := t.createVPCRoute(cidrStr); err != nil {
+			return fmt.Errorf("createVPCRoutes failed: %s", err)
+		}
+		if local {
+			if err := t.createHostRoute(cidrStr); err != nil {
+				if errno, ok := err.(syscall.Errno); !(ok && errno == syscall.EEXIST) {
+					return fmt.Errorf("createHostRoute failed: %s", err)
+				}
+			}
+		}
+	}
+
+	// Remove obsolete entries
+	for _, cidr := range prev {
+		cidrStr := cidr.String()
+		t.debugf("removing %s route", cidrStr)
+		if _, err := t.deleteVPCRoute(cidrStr); err != nil {
+			return fmt.Errorf("deleteVPCRoute failed: %s", err)
+		}
+		if local {
+			if err := t.deleteHostRoute(cidrStr); err != nil {
+				return fmt.Errorf("deleteHostRoute failed: %s", err)
+			}
+		}
+	}
+
+	return nil
+}
+
+func (t *AWSVPCTracker) createVPCRoute(cidr string) (*ec2.CreateRouteOutput, error) {
+	route := &ec2.CreateRouteInput{
+		RouteTableId:         &t.routeTableID,
+		InstanceId:           &t.instanceID,
+		DestinationCidrBlock: &cidr,
+	}
+	return t.ec2.CreateRoute(route)
+}
+
+func (t *AWSVPCTracker) createHostRoute(cidr string) error {
+	dst, err := parseCIDR(cidr)
+	if err != nil {
+		return err
+	}
+	route := &netlink.Route{
+		LinkIndex: t.linkIndex,
+		Dst:       dst,
+		Scope:     netlink.SCOPE_LINK,
+	}
+	return netlink.RouteAdd(route)
+}
+
+func (t *AWSVPCTracker) deleteVPCRoute(cidr string) (*ec2.DeleteRouteOutput, error) {
+	route := &ec2.DeleteRouteInput{
+		RouteTableId:         &t.routeTableID,
+		DestinationCidrBlock: &cidr,
+	}
+	return t.ec2.DeleteRoute(route)
+}
+
+func (t *AWSVPCTracker) deleteHostRoute(cidr string) error {
+	dst, err := parseCIDR(cidr)
+	if err != nil {
+		return err
+	}
+	route := &netlink.Route{
+		LinkIndex: t.linkIndex,
+		Dst:       dst,
+		Scope:     netlink.SCOPE_LINK,
+	}
+	return netlink.RouteDel(route)
+}
+
+// detectRouteTableID detects AWS VPC Route Table ID of the given tracker instance.
+func (t *AWSVPCTracker) detectRouteTableID() (*string, error) {
+	instancesParams := &ec2.DescribeInstancesInput{
+		InstanceIds: []*string{aws.String(t.instanceID)},
+	}
+	instancesResp, err := t.ec2.DescribeInstances(instancesParams)
+	if err != nil {
+		return nil, fmt.Errorf("DescribeInstances failed: %s", err)
+	}
+	if len(instancesResp.Reservations) == 0 ||
+		len(instancesResp.Reservations[0].Instances) == 0 {
+		return nil, fmt.Errorf("cannot find %s instance within reservations", t.instanceID)
+	}
+	vpcID := instancesResp.Reservations[0].Instances[0].VpcId
+	subnetID := instancesResp.Reservations[0].Instances[0].SubnetId
+
+	// First try to find a routing table associated with the subnet of the instance
+	tablesParams := &ec2.DescribeRouteTablesInput{
+		Filters: []*ec2.Filter{
+			{
+				Name:   aws.String("association.subnet-id"),
+				Values: []*string{subnetID},
+			},
+		},
+	}
+	tablesResp, err := t.ec2.DescribeRouteTables(tablesParams)
+	if err != nil {
+		return nil, fmt.Errorf("DescribeRouteTables failed: %s", err)
+	}
+	if len(tablesResp.RouteTables) != 0 {
+		return tablesResp.RouteTables[0].RouteTableId, nil
+	}
+	// Fallback to the default routing table
+	tablesParams = &ec2.DescribeRouteTablesInput{
+		Filters: []*ec2.Filter{
+			{
+				Name:   aws.String("association.main"),
+				Values: []*string{aws.String("true")},
+			},
+			{
+				Name:   aws.String("vpc-id"),
+				Values: []*string{vpcID},
+			},
+		},
+	}
+	tablesResp, err = t.ec2.DescribeRouteTables(tablesParams)
+	if err != nil {
+		return nil, fmt.Errorf("DescribeRouteTables failed: %s", err)
+	}
+	if len(tablesResp.RouteTables) != 0 {
+		return tablesResp.RouteTables[0].RouteTableId, nil
+	}
+
+	return nil, fmt.Errorf("cannot find routetable for %s instance", t.instanceID)
+}
+
+func (t *AWSVPCTracker) debugf(fmt string, args ...interface{}) {
+	common.Log.Debugf("[tracker] "+fmt, args...)
+}
+
+func (t *AWSVPCTracker) infof(fmt string, args ...interface{}) {
+	common.Log.Infof("[tracker] "+fmt, args...)
+}
+
+// Helpers
+
+// merge merges adjacent range entries.
+// The given slice has to be sorted in increasing order.
+func merge(r []address.Range) []address.Range {
+	var merged []address.Range
+
+	for i := range r {
+		if prev := len(merged) - 1; prev >= 0 && merged[prev].End == r[i].Start {
+			merged[prev].End = r[i].End
+		} else {
+			merged = append(merged, r[i])
+		}
+	}
+
+	return merged
+}
+
+// removeCommon filters out CIDR ranges which are contained in both a and b slices.
+// Both slices have to be sorted in increasing order.
+func removeCommon(a, b []address.CIDR) (newA, newB []address.CIDR) {
+	i, j := 0, 0
+
+	for i < len(a) && j < len(b) {
+		switch {
+		case a[i].Start() < b[j].Start() || a[i].End() < b[j].End():
+			newA = append(newA, a[i])
+			i++
+		case a[i].Start() > b[j].Start() || a[i].End() > b[j].End():
+			newB = append(newB, b[j])
+			j++
+		default:
+			i++
+			j++
+		}
+
+	}
+	newA = append(newA, a[i:]...)
+	newB = append(newB, b[j:]...)
+
+	return
+}
+
+func parseCIDR(cidr string) (*net.IPNet, error) {
+	ip, ipnet, err := net.ParseCIDR(cidr)
+	if err != nil {
+		return nil, err
+	}
+	ipnet.IP = ip
+
+	return ipnet, nil
+}
@@ -0,0 +1,52 @@
+package tracker
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+
+	"github.com/weaveworks/weave/net/address"
+)
+
+var (
+	r0to127     = cidr("10.0.0.0", "10.0.0.127")
+	r128to255   = cidr("10.0.0.128", "10.0.0.255")
+	r0to255     = cidr("10.0.0.0", "10.0.0.255")
+	r1dot0to255 = cidr("10.0.1.0", "10.0.1.255")
+	r2dot0to255 = cidr("10.0.2.0", "10.0.2.255")
+)
+
+func TestRemoveCommon(t *testing.T) {
+	a := []address.CIDR{r0to127, r1dot0to255}
+	b := []address.CIDR{r1dot0to255, r2dot0to255}
+	newA, newB := removeCommon(a, b)
+	require.Equal(t, []address.CIDR{r0to127}, newA)
+	require.Equal(t, []address.CIDR{r2dot0to255}, newB)
+}
+
+func TestMerge(t *testing.T) {
+	ranges := []address.Range{
+		r0to127.Range(),
+		r128to255.Range(),
+		r2dot0to255.Range(),
+	}
+	require.Equal(t, []address.Range{r0to255.Range(), r2dot0to255.Range()}, merge(ranges))
+}
+
+// Helper
+
+// TODO(mp) DRY with helpers of other tests.
+
+func ip(s string) address.Address {
+	addr, _ := address.ParseIP(s)
+	return addr
+}
+
+// [start; end]
+func cidr(start, end string) address.CIDR {
+	c := address.Range{Start: ip(start), End: ip(end) + 1}.CIDRs()
+	if len(c) != 1 {
+		panic("invalid cidr")
+	}
+	return c[0]
+}
@@ -0,0 +1,20 @@
+package tracker
+
+import (
+	"github.com/weaveworks/weave/net/address"
+)
+
+// LocalRangeTracker is an interface for tracking changes in the IPAM ring.
+type LocalRangeTracker interface {
+	// HandleUpdate is called whenever an address ring gets updated.
+	//
+	// prevRanges corresponds to ranges which were owned by a peer before
+	// a change in the ring, while currRanges to the ones which are currently
+	// owned by the peer.
+	// Both slices have to be sorted in increasing order.
+	// Adjacent ranges within each slice might appear as separate ranges.
+	//
+	// The local parameter indicates whether the ranges belong to the peer
+	// by which the method is called.
+	HandleUpdate(prevRanges, currRanges []address.Range, local bool) error
+}
@@ -5,8 +5,10 @@ import (
 	"fmt"
 	"math/rand"
 	"net"
+	"os"
 	"sort"
 	"strings"
+	"sync"
 	"time"

 	"github.com/miekg/dns"
@@ -17,7 +19,6 @@ import (
 const (
 	topDomain        = "."
 	reverseDNSdomain = "in-addr.arpa."
-	etcResolvConf    = "/etc/resolv.conf"
 	udpBuffSize      = uint16(4096)
 	minUDPSize       = 512

@@ -26,16 +27,52 @@ const (
 	DefaultClientTimeout = 5 * time.Second
 )

-type DNSServer struct {
-	ns      *Nameserver
-	domain  string
-	ttl     uint32
-	address string
+type Upstream interface {
+	Config() (*dns.ClientConfig, error)
+}

-	servers   []*dns.Server
-	upstream  *dns.ClientConfig
-	tcpClient *dns.Client
-	udpClient *dns.Client
+func NewUpstream(resolvConf, filterAddress string) Upstream {
+	return &upstream{
+		resolvConf:    resolvConf,
+		filterAddress: filterAddress,
+		cachedConfig:  &dns.ClientConfig{}}
+}
+
+type upstream struct {
+	sync.Mutex
+
+	resolvConf    string
+	filterAddress string
+	cachedConfig  *dns.ClientConfig
+	lastStat      time.Time
+	lastModified  time.Time
+}
+
+func (u *upstream) Config() (*dns.ClientConfig, error) {
+	u.Lock()
+	defer u.Unlock()
+
+	// If we checked less than five seconds ago return what we already have
+	now := time.Now()
+	if now.Sub(u.lastStat) < 5*time.Second {
+		return u.cachedConfig, nil
+	}
+	u.lastStat = now
+
+	// Reread file if it has changed
+	if fi, err := os.Stat(u.resolvConf); err != nil {
+		return u.cachedConfig, err
+	} else if fi.ModTime() != u.lastModified {
+		if config, err := dns.ClientConfigFromFile(u.resolvConf); err == nil {
+			config.Servers = filter(config.Servers, u.filterAddress)
+			u.lastModified = fi.ModTime()
+			u.cachedConfig = config
+		} else {
+			return u.cachedConfig, err
+		}
+	}
+
+	return u.cachedConfig, nil
 }

 func filter(ss []string, s string) []string {
@@ -49,24 +86,30 @@ func filter(ss []string, s string) []string {
 	return ss
 }

-func NewDNSServer(ns *Nameserver, domain, address, effectiveAddress string, ttl uint32, clientTimeout time.Duration) (*DNSServer, error) {
+type DNSServer struct {
+	ns      *Nameserver
+	domain  string
+	ttl     uint32
+	address string
+
+	servers   []*dns.Server
+	upstream  Upstream
+	tcpClient *dns.Client
+	udpClient *dns.Client
+}
+
+func NewDNSServer(ns *Nameserver, domain, address string, upstream Upstream, ttl uint32, clientTimeout time.Duration) (*DNSServer, error) {
 	s := &DNSServer{
 		ns:        ns,
 		domain:    dns.Fqdn(domain),
 		ttl:       ttl,
 		address:   address,
+		upstream:  upstream,
 		tcpClient: &dns.Client{Net: "tcp", ReadTimeout: clientTimeout},
 		udpClient: &dns.Client{Net: "udp", ReadTimeout: clientTimeout, UDPSize: udpBuffSize},
 	}
-	var err error
-	if s.upstream, err = dns.ClientConfigFromFile(etcResolvConf); err != nil {
-		return nil, err
-	}
-	if s.upstream != nil {
-		s.upstream.Servers = filter(s.upstream.Servers, effectiveAddress)
-	}

-	err = s.listen(address)
+	err := s.listen(address)
 	return s, err
 }

@@ -134,7 +177,7 @@ func (d *DNSServer) createMux(client *dns.Client, defaultMaxResponseSize int) *d

 func (h *handler) handleLocal(w dns.ResponseWriter, req *dns.Msg) {
 	h.ns.debugf("local request: %+v", *req)
-	if len(req.Question) != 1 || req.Question[0].Qtype != dns.TypeA {
+	if len(req.Question) != 1 {
 		h.nameError(w, req)
 		return
 	}
@@ -149,6 +192,12 @@ func (h *handler) handleLocal(w dns.ResponseWriter, req *dns.Msg) {
 		h.nameError(w, req)
 		return
 	}
+	// Per RFC4074, if we have an A but another type was requested,
+	// return 'no error' with empty answer section
+	if req.Question[0].Qtype != dns.TypeA {
+		h.respond(w, h.makeResponse(req, nil))
+		return
+	}

 	header := dns.RR_Header{
 		Name:   req.Question[0].Name,
@@ -173,7 +222,7 @@ func (h *handler) handleReverse(w dns.ResponseWriter, req *dns.Msg) {
 		return
 	}

-	ipStr := strings.TrimSuffix(req.Question[0].Name, "."+reverseDNSdomain)
+	ipStr := strings.TrimSuffix(strings.ToLower(req.Question[0].Name), "."+reverseDNSdomain)
 	ip, err := address.ParseIP(ipStr)
 	if err != nil {
 		h.nameError(w, req)
@@ -204,7 +253,7 @@ func (h *handler) handleRecursive(w dns.ResponseWriter, req *dns.Msg) {
 	h.ns.debugf("recursive request: %+v", *req)

 	// Resolve unqualified names locally
-	if len(req.Question) == 1 && req.Question[0].Qtype == dns.TypeA {
+	if len(req.Question) == 1 {
 		hostname := dns.Fqdn(req.Question[0].Name)
 		if strings.Count(hostname, ".") == 1 {
 			h.handleLocal(w, req)
@@ -212,10 +261,14 @@ func (h *handler) handleRecursive(w dns.ResponseWriter, req *dns.Msg) {
 		}
 	}

-	for _, server := range h.upstream.Servers {
+	upstreamConfig, err := h.upstream.Config()
+	if err != nil {
+		h.ns.errorf("unable to read upstream config: %s", err)
+	}
+	for _, server := range upstreamConfig.Servers {
 		reqCopy := req.Copy()
 		reqCopy.Id = dns.Id()
-		response, _, err := h.client.Exchange(reqCopy, fmt.Sprintf("%s:%s", server, h.upstream.Port))
+		response, _, err := h.client.Exchange(reqCopy, fmt.Sprintf("%s:%s", server, upstreamConfig.Port))
 		if (err != nil && err != dns.ErrTruncated) || response == nil {
 			h.ns.debugf("error trying %s: %v", server, err)
 			continue
@@ -11,21 +11,26 @@ import (
 	"github.com/miekg/dns"
 	"github.com/stretchr/testify/require"

+	"github.com/weaveworks/mesh"
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

+type mockUpstream struct {
+	config *dns.ClientConfig
+}
+
+func (mu *mockUpstream) Config() (*dns.ClientConfig, error) {
+	return mu.config, nil
+}
+
 func startServer(t *testing.T, upstream *dns.ClientConfig) (*DNSServer, *Nameserver, int, int) {
-	peername, err := router.PeerNameFromString("00:00:00:02:00:00")
+	peername, err := mesh.PeerNameFromString("00:00:00:02:00:00")
 	require.Nil(t, err)
-	nameserver := New(peername, nil, "")
-	dnsserver, err := NewDNSServer(nameserver, "weave.local.", "0.0.0.0:0", "", 30, 5*time.Second)
+	nameserver := New(peername, "", func(mesh.PeerName) bool { return true })
+	dnsserver, err := NewDNSServer(nameserver, "weave.local.", "0.0.0.0:0", &mockUpstream{upstream}, 30, 5*time.Second)
 	require.Nil(t, err)
 	udpPort := dnsserver.servers[0].PacketConn.LocalAddr().(*net.UDPAddr).Port
 	tcpPort := dnsserver.servers[1].Listener.Addr().(*net.TCPAddr).Port
-	if upstream != nil {
-		dnsserver.upstream = upstream
-	}
 	go dnsserver.ActivateAndServe()
 	return dnsserver, nameserver, udpPort, tcpPort
 }
@@ -39,7 +44,7 @@ func TestTruncation(t *testing.T) {
 	addrs := []address.Address{}
 	for i := address.Address(0); i < 100; i++ {
 		addrs = append(addrs, i)
-		nameserver.AddEntry("foo.weave.local.", "", router.UnknownPeerName, i)
+		nameserver.AddEntry("foo.weave.local.", "", mesh.UnknownPeerName, i)
 	}

 	doRequest := func(client *dns.Client, request *dns.Msg, port int, expectedErr error) *dns.Msg {
@@ -8,15 +8,16 @@ import (
 	"strings"
 	"time"

+	"github.com/weaveworks/mesh"
+
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

 var now = func() int64 { return time.Now().Unix() }

 type Entry struct {
 	ContainerID string
-	Origin      router.PeerName
+	Origin      mesh.PeerName
 	Addr        address.Address
 	Hostname    string // as supplied
 	lHostname   string // lowercased (not exported, so not encoded by gob)
@@ -93,6 +94,9 @@ func (e1 *Entry) merge(e2 *Entry) bool {
 		e1.Version = e2.Version
 		e1.Tombstone = e2.Tombstone
 		return true
+	} else if e2.Version == e1.Version && e2.Tombstone > e1.Tombstone {
+		e1.Tombstone = e2.Tombstone
+		return true
 	}
 	return false
 }
@@ -105,9 +109,18 @@ func (e1 *Entry) addLowercase() {
 	e1.lHostname = strings.ToLower(e1.Hostname)
 }

+func (e1 *Entry) tombstone() bool {
+	if e1.Tombstone > 0 {
+		return false
+	}
+	e1.Tombstone = now()
+	e1.Version++
+	return true
+}
+
 func check(es SortableEntries) error {
 	if !sort.IsSorted(es) {
-		return fmt.Errorf("Not sorted!")
+		return fmt.Errorf("Not sorted")
 	}
 	for i := 1; i < es.Len(); i++ {
 		if es.Get(i).equal(es.Get(i - 1)) {
@@ -128,7 +141,7 @@ func (es *Entries) checkAndPanic() *Entries {
 	return es
 }

-func (es *Entries) add(hostname, containerid string, origin router.PeerName, addr address.Address) Entry {
+func (es *Entries) add(hostname, containerid string, origin mesh.PeerName, addr address.Address) Entry {
 	defer es.checkAndPanic().checkAndPanic()

 	entry := Entry{Hostname: hostname, lHostname: strings.ToLower(hostname),
@@ -176,14 +189,12 @@ func (es *Entries) merge(incoming Entries) Entries {
 }

 // f returning true means keep the entry.
-func (es *Entries) tombstone(ourname router.PeerName, f func(*Entry) bool) Entries {
+func (es *Entries) tombstone(ourname mesh.PeerName, f func(*Entry) bool) Entries {
 	defer es.checkAndPanic().checkAndPanic()

 	tombstoned := Entries{}
 	for i, e := range *es {
-		if f(&e) && e.Origin == ourname {
-			e.Version++
-			e.Tombstone = now()
+		if f(&e) && e.Origin == ourname && e.tombstone() {
 			(*es)[i] = e
 			tombstoned = append(tombstoned, e)
 		}
@@ -191,6 +202,7 @@ func (es *Entries) tombstone(ourname router.PeerName, f func(*Entry) bool) Entri
 	return tombstoned
 }

+// note f() may only modify entries such that they remain in order defined by less()
 func (es *Entries) filter(f func(*Entry) bool) {
 	defer es.checkAndPanic().checkAndPanic()

@@ -205,6 +217,16 @@ func (es *Entries) filter(f func(*Entry) bool) {
 	*es = (*es)[:i]
 }

+func (es Entries) findEqual(e *Entry) (*Entry, bool) {
+	i := sort.Search(len(es), func(i int) bool {
+		return !es[i].insensitiveLess(e)
+	})
+	if i < len(es) && es[i].equal(*e) {
+		return &es[i], true
+	}
+	return nil, false
+}
+
 func (es Entries) lookup(hostname string) Entries {
 	es.checkAndPanic()

@@ -245,12 +267,14 @@ type GossipData struct {
 	Entries
 }

-func (g *GossipData) Merge(o router.GossipData) {
+func (g *GossipData) Merge(o mesh.GossipData) mesh.GossipData {
 	other := o.(*GossipData)
-	g.Entries.merge(other.Entries)
-	if g.Timestamp < other.Timestamp {
-		g.Timestamp = other.Timestamp
+	gossip := g.copy()
+	gossip.Entries.merge(other.Entries)
+	if gossip.Timestamp < other.Timestamp {
+		gossip.Timestamp = other.Timestamp
 	}
+	return gossip
 }

 func (g *GossipData) Decode(msg []byte) error {
@@ -264,9 +288,7 @@ func (g *GossipData) Decode(msg []byte) error {
 }

 func (g *GossipData) Encode() [][]byte {
-	// Make a copy so we can sort: all outgoing data is sent in case-sensitive order
-	g2 := GossipData{Timestamp: g.Timestamp, Entries: make(Entries, len(g.Entries))}
-	copy(g2.Entries, g.Entries)
+	g2 := g.copy()
 	sort.Sort(CaseSensitive(g2.Entries))
 	buf := &bytes.Buffer{}
 	if err := gob.NewEncoder(buf).Encode(g2); err != nil {
@@ -274,3 +296,9 @@ func (g *GossipData) Encode() [][]byte {
 	}
 	return [][]byte{buf.Bytes()}
 }
+
+func (g *GossipData) copy() *GossipData {
+	g2 := &GossipData{Timestamp: g.Timestamp, Entries: make(Entries, len(g.Entries))}
+	copy(g2.Entries, g.Entries)
+	return g2
+}
@@ -4,9 +4,9 @@ import (
 	"testing"

 	"github.com/stretchr/testify/require"
+	"github.com/weaveworks/mesh"

 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

 func l(es Entries) Entries {
@@ -28,21 +28,21 @@ func TestAdd(t *testing.T) {
 	now = func() int64 { return 1234 }

 	entries := Entries{}
-	entries.add("A", "", router.UnknownPeerName, address.Address(0))
+	entries.add("A", "", mesh.UnknownPeerName, address.Address(0))
 	expected := l(Entries{
-		Entry{Hostname: "A", Origin: router.UnknownPeerName, Addr: address.Address(0)},
+		Entry{Hostname: "A", Origin: mesh.UnknownPeerName, Addr: address.Address(0)},
 	})
 	require.Equal(t, entries, expected)

-	entries.tombstone(router.UnknownPeerName, func(e *Entry) bool { return e.Hostname == "A" })
+	entries.tombstone(mesh.UnknownPeerName, func(e *Entry) bool { return e.Hostname == "A" })
 	expected = l(Entries{
-		Entry{Hostname: "A", Origin: router.UnknownPeerName, Addr: address.Address(0), Version: 1, Tombstone: 1234},
+		Entry{Hostname: "A", Origin: mesh.UnknownPeerName, Addr: address.Address(0), Version: 1, Tombstone: 1234},
 	})
 	require.Equal(t, entries, expected)

-	entries.add("A", "", router.UnknownPeerName, address.Address(0))
+	entries.add("A", "", mesh.UnknownPeerName, address.Address(0))
 	expected = l(Entries{
-		Entry{Hostname: "A", Origin: router.UnknownPeerName, Addr: address.Address(0), Version: 2},
+		Entry{Hostname: "A", Origin: mesh.UnknownPeerName, Addr: address.Address(0), Version: 2},
 	})
 	require.Equal(t, entries, expected)
 }
@@ -78,7 +78,7 @@ func TestTombstone(t *testing.T) {

 	es := makeEntries("AB")

-	es.tombstone(router.UnknownPeerName, func(e *Entry) bool {
+	es.tombstone(mesh.UnknownPeerName, func(e *Entry) bool {
 		return e.Hostname == "B"
 	})
 	expected := l(Entries{
@@ -86,6 +86,21 @@ func TestTombstone(t *testing.T) {
 		Entry{Hostname: "B", Version: 1, Tombstone: 1234},
 	})
 	require.Equal(t, expected, es)
+
+	// Now try a merge including two entries which differ only in tombstone
+	e2 := make(Entries, len(es))
+	copy(e2, es)
+	e2[1].Tombstone = 5555
+
+	diff := es.merge(e2)
+
+	expected2 := l(Entries{
+		Entry{Hostname: "A"},
+		Entry{Hostname: "B", Version: 1, Tombstone: 5555},
+	})
+	require.Equal(t, expected2, es)
+	expectedDiff := l(Entries{Entry{Hostname: "B", Version: 1, Tombstone: 5555}})
+	require.Equal(t, expectedDiff, diff)
 }

 func TestDelete(t *testing.T) {
@@ -117,7 +132,7 @@ func TestGossipDataMerge(t *testing.T) {
 	g1 := GossipData{Entries: makeEntries("AcDf")}
 	g2 := GossipData{Entries: makeEntries("BEf")}

-	g1.Merge(&g2)
+	g3 := g1.Merge(&g2).(*GossipData)

-	require.Equal(t, GossipData{Entries: makeEntries("ABcDEf")}, g1)
+	require.Equal(t, GossipData{Entries: makeEntries("ABcDEf")}, *g3)
 }
@@ -27,7 +27,7 @@ func (n *Nameserver) HandleHTTP(router *mux.Router, dockerCli *docker.Client) {
 			vars      = mux.Vars(r)
 			container = vars["container"]
 			ipStr     = vars["ip"]
-			hostname  = dns.Fqdn(r.FormValue("fqdn"))
+			fqdn      = r.FormValue("fqdn")
 			ip, err   = address.ParseIP(ipStr)
 		)
 		if err != nil {
@@ -35,21 +35,11 @@ func (n *Nameserver) HandleHTTP(router *mux.Router, dockerCli *docker.Client) {
 			return
 		}

-		if !dns.IsSubDomain(n.domain, hostname) {
-			n.infof("Ignoring registration %s %s %s (not a subdomain of %s)", hostname, ipStr, container, n.domain)
-			return
-		}
-
-		if err := n.AddEntry(hostname, container, n.ourName, ip); err != nil {
-			n.badRequest(w, fmt.Errorf("Unable to add entry: %v", err))
-			return
-		}
+		n.AddEntryFQDN(fqdn, container, n.ourName, ip)

 		if r.FormValue("check-alive") == "true" && dockerCli != nil && dockerCli.IsContainerNotRunning(container) {
 			n.infof("container '%s' is not running: removing", container)
-			if err := n.Delete(hostname, container, ipStr, ip); err != nil {
-				n.infof("failed to remove: %v", err)
-			}
+			n.Delete(dns.Fqdn(fqdn), container, ipStr, ip)
 		}

 		w.WriteHeader(204)
@@ -79,10 +69,7 @@ func (n *Nameserver) HandleHTTP(router *mux.Router, dockerCli *docker.Client) {
 			ipStr = "*"
 		}

-		if err := n.Delete(hostname, container, ipStr, ip); err != nil {
-			n.badRequest(w, fmt.Errorf("Unable to delete entries: %v", err))
-			return
-		}
+		n.Delete(hostname, container, ipStr, ip)
 		w.WriteHeader(204)
 	}
 	router.Methods("DELETE").Path("/name/{container}/{ip}").HandlerFunc(deleteHandler)
@@ -6,10 +6,10 @@ import (
 	"time"

 	"github.com/miekg/dns"
+	"github.com/weaveworks/mesh"

-	. "github.com/weaveworks/weave/common"
+	"github.com/weaveworks/weave/common"
 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
 )

 const (
@@ -34,28 +34,24 @@ const (
 // - Update is O(n) for now
 type Nameserver struct {
 	sync.RWMutex
-	ourName router.PeerName
-	domain  string
-	gossip  router.Gossip
-	entries Entries
-	peers   *router.Peers
-	quit    chan struct{}
+	ourName     mesh.PeerName
+	domain      string
+	gossip      mesh.Gossip
+	entries     Entries
+	isKnownPeer func(mesh.PeerName) bool
+	quit        chan struct{}
 }

-func New(ourName router.PeerName, peers *router.Peers, domain string) *Nameserver {
-	ns := &Nameserver{
-		ourName: ourName,
-		domain:  dns.Fqdn(domain),
-		peers:   peers,
-		quit:    make(chan struct{}),
+func New(ourName mesh.PeerName, domain string, isKnownPeer func(mesh.PeerName) bool) *Nameserver {
+	return &Nameserver{
+		ourName:     ourName,
+		domain:      dns.Fqdn(domain),
+		isKnownPeer: isKnownPeer,
+		quit:        make(chan struct{}),
 	}
-	if peers != nil {
-		peers.OnGC(ns.PeerGone)
-	}
-	return ns
 }

-func (n *Nameserver) SetGossip(gossip router.Gossip) {
+func (n *Nameserver) SetGossip(gossip mesh.Gossip) {
 	n.gossip = gossip
 }

@@ -77,22 +73,31 @@ func (n *Nameserver) Stop() {
 	n.quit <- struct{}{}
 }

-func (n *Nameserver) broadcastEntries(es ...Entry) error {
-	if n.gossip != nil {
-		return n.gossip.GossipBroadcast(&GossipData{
-			Entries:   Entries(es),
-			Timestamp: now(),
-		})
+func (n *Nameserver) broadcastEntries(es ...Entry) {
+	if n.gossip == nil || len(es) == 0 {
+		return
 	}
-	return nil
+	n.gossip.GossipBroadcast(&GossipData{
+		Entries:   Entries(es),
+		Timestamp: now(),
+	})
 }

-func (n *Nameserver) AddEntry(hostname, containerid string, origin router.PeerName, addr address.Address) error {
-	n.infof("adding entry %s -> %s", hostname, addr.String())
+func (n *Nameserver) AddEntry(hostname, containerid string, origin mesh.PeerName, addr address.Address) {
 	n.Lock()
+	n.infof("adding entry for %s: %s -> %s", containerid, hostname, addr.String())
 	entry := n.entries.add(hostname, containerid, origin, addr)
 	n.Unlock()
-	return n.broadcastEntries(entry)
+	n.broadcastEntries(entry)
+}
+
+func (n *Nameserver) AddEntryFQDN(fqdn, containerid string, origin mesh.PeerName, addr address.Address) {
+	hostname := dns.Fqdn(fqdn)
+	if !dns.IsSubDomain(n.domain, hostname) {
+		n.infof("Ignoring registration %s %s %s (not a subdomain of %s)", hostname, addr.String(), containerid, n.domain)
+		return
+	}
+	n.AddEntry(hostname, containerid, origin, addr)
 }

 func (n *Nameserver) Lookup(hostname string) []address.Address {
@@ -125,7 +130,8 @@ func (n *Nameserver) ReverseLookup(ip address.Address) (string, error) {
 	return match.Hostname, nil
 }

-func (n *Nameserver) ContainerStarted(ident string) {}
+func (n *Nameserver) ContainerStarted(ident string)   {}
+func (n *Nameserver) ContainerDestroyed(ident string) {}

 func (n *Nameserver) ContainerDied(ident string) {
 	n.Lock()
@@ -137,23 +143,19 @@ func (n *Nameserver) ContainerDied(ident string) {
 		return false
 	})
 	n.Unlock()
-	if len(entries) > 0 {
-		if err := n.broadcastEntries(entries...); err != nil {
-			n.errorf("failed to broadcast container %s death: %v", ident, err)
-		}
-	}
+	n.broadcastEntries(entries...)
 }

-func (n *Nameserver) PeerGone(peer *router.Peer) {
+func (n *Nameserver) PeerGone(peer mesh.PeerName) {
 	n.infof("peer %s gone", peer.String())
 	n.Lock()
 	defer n.Unlock()
 	n.entries.filter(func(e *Entry) bool {
-		return e.Origin != peer.Name
+		return e.Origin != peer
 	})
 }

-func (n *Nameserver) Delete(hostname, containerid, ipStr string, ip address.Address) error {
+func (n *Nameserver) Delete(hostname, containerid, ipStr string, ip address.Address) {
 	n.Lock()
 	n.infof("tombstoning hostname=%s, container=%s, ip=%s", hostname, containerid, ipStr)
 	entries := n.entries.tombstone(n.ourName, func(e *Entry) bool {
@@ -173,7 +175,7 @@ func (n *Nameserver) Delete(hostname, containerid, ipStr string, ip address.Addr
 		return true
 	})
 	n.Unlock()
-	return n.broadcastEntries(entries...)
+	n.broadcastEntries(entries...)
 }

 func (n *Nameserver) deleteTombstones() {
@@ -185,7 +187,7 @@ func (n *Nameserver) deleteTombstones() {
 	})
 }

-func (n *Nameserver) Gossip() router.GossipData {
+func (n *Nameserver) Gossip() mesh.GossipData {
 	n.RLock()
 	defer n.RUnlock()
 	gossip := &GossipData{
@@ -196,29 +198,56 @@ func (n *Nameserver) Gossip() router.GossipData {
 	return gossip
 }

-func (n *Nameserver) OnGossipUnicast(sender router.PeerName, msg []byte) error {
+func (n *Nameserver) OnGossipUnicast(sender mesh.PeerName, msg []byte) error {
 	return nil
 }

-func (n *Nameserver) receiveGossip(msg []byte) (router.GossipData, router.GossipData, error) {
+func (n *Nameserver) receiveGossip(msg []byte) (mesh.GossipData, mesh.GossipData, error) {
 	var gossip GossipData
 	if err := gossip.Decode(msg); err != nil {
 		return nil, nil, err
 	}
 	if delta := gossip.Timestamp - now(); delta > gossipWindow || delta < -gossipWindow {
-		return nil, nil, fmt.Errorf("clock skew of %d detected", delta)
+		return nil, nil, fmt.Errorf("host clock skew of %ds exceeds %ds limit", delta, gossipWindow)
 	}

+	// Filter to remove entries from unknown peers, done before we take
+	// the nameserver lock, so we don't have to worry what isKnownPeer locks.
+	gossip.Entries.filter(func(e *Entry) bool {
+		return n.isKnownPeer(e.Origin)
+	})
+
+	var overriddenEntries []Entry
 	n.Lock()
-	defer n.Unlock()

-	if n.peers != nil {
-		gossip.Entries.filter(func(e *Entry) bool {
-			return n.peers.Fetch(e.Origin) != nil
-		})
-	}
+	// Check entries claiming to originate from us against our current data
+	gossip.Entries.filter(func(e *Entry) bool {
+		if e.Origin == n.ourName {
+			if ourEntry, ok := n.entries.findEqual(e); ok {
+				if ourEntry.Version < e.Version ||
+					(ourEntry.Version == e.Version && ourEntry.Tombstone != e.Tombstone) {
+					// Take our version of the data, but make the version higher than the incoming
+					nextVersion := e.Version + 1
+					*e = *ourEntry
+					e.Version = nextVersion
+					overriddenEntries = append(overriddenEntries, *e)
+				}
+			} else { // We have no entry matching the one that came in with us as Origin
+				if e.tombstone() {
+					overriddenEntries = append(overriddenEntries, *e)
+				}
+			}
+		}
+		return true
+	})

 	newEntries := n.entries.merge(gossip.Entries)
+	n.Unlock() // unlock before attempting to broadcast
+
+	// Note that all overriddenEntries have been merged into our entries, either
+	// because we forced the version higher or because they were missing before.
+	n.broadcastEntries(overriddenEntries...)
+
 	if len(newEntries) > 0 {
 		return &GossipData{Entries: newEntries, Timestamp: now()}, &gossip, nil
 	}
@@ -227,24 +256,29 @@ func (n *Nameserver) receiveGossip(msg []byte) (router.GossipData, router.Gossip

 // merge received data into state and return "everything new I've
 // just learnt", or nil if nothing in the received data was new
-func (n *Nameserver) OnGossip(msg []byte) (router.GossipData, error) {
+func (n *Nameserver) OnGossip(msg []byte) (mesh.GossipData, error) {
 	newEntries, _, err := n.receiveGossip(msg)
 	return newEntries, err
 }

 // merge received data into state and return a representation of
 // the received data, for further propagation
-func (n *Nameserver) OnGossipBroadcast(_ router.PeerName, msg []byte) (router.GossipData, error) {
+func (n *Nameserver) OnGossipBroadcast(_ mesh.PeerName, msg []byte) (mesh.GossipData, error) {
 	_, entries, err := n.receiveGossip(msg)
 	return entries, err
 }

+// Logging
+
 func (n *Nameserver) infof(fmt string, args ...interface{}) {
-	Log.Infof("[nameserver %s] "+fmt, append([]interface{}{n.ourName}, args...)...)
+	n.logf(common.Log.Infof, fmt, args...)
 }
 func (n *Nameserver) debugf(fmt string, args ...interface{}) {
-	Log.Debugf("[nameserver %s] "+fmt, append([]interface{}{n.ourName}, args...)...)
+	n.logf(common.Log.Debugf, fmt, args...)
 }
 func (n *Nameserver) errorf(fmt string, args ...interface{}) {
-	Log.Errorf("[nameserver %s] "+fmt, append([]interface{}{n.ourName}, args...)...)
+	n.logf(common.Log.Errorf, fmt, args...)
+}
+func (n *Nameserver) logf(f func(string, ...interface{}), fmt string, args ...interface{}) {
+	f("[nameserver %s] "+fmt, append([]interface{}{n.ourName}, args...)...)
 }
@@ -10,20 +10,23 @@ import (
 	"time"

 	"github.com/stretchr/testify/require"
+	"github.com/weaveworks/mesh"

 	"github.com/weaveworks/weave/net/address"
-	"github.com/weaveworks/weave/router"
-	wt "github.com/weaveworks/weave/testing"
 	"github.com/weaveworks/weave/testing/gossip"
 )

+func makeNameserver(name mesh.PeerName) *Nameserver {
+	return New(name, "", func(mesh.PeerName) bool { return true })
+}
+
 func makeNetwork(size int) ([]*Nameserver, *gossip.TestRouter) {
 	gossipRouter := gossip.NewTestRouter(0.0)
 	nameservers := make([]*Nameserver, size)

 	for i := 0; i < size; i++ {
-		name, _ := router.PeerNameFromString(fmt.Sprintf("%02d:00:00:02:00:00", i))
-		nameserver := New(name, nil, "")
+		name, _ := mesh.PeerNameFromString(fmt.Sprintf("%02d:00:00:02:00:00", i))
+		nameserver := makeNameserver(name)
 		nameserver.SetGossip(gossipRouter.Connect(nameserver.ourName, nameserver))
 		nameserver.Start()
 		nameservers[i] = nameserver
@@ -40,7 +43,7 @@ func stopNetwork(nameservers []*Nameserver, grouter *gossip.TestRouter) {
 }

 type pair struct {
-	origin router.PeerName
+	origin mesh.PeerName
 	addr   address.Address
 }

@@ -72,12 +75,6 @@ func (m mapping) Addrs() []address.Address {
 }

 func TestNameservers(t *testing.T) {
-	wt.RunWithTimeout(t, 2*time.Minute, func() {
-		testNameservers(t)
-	})
-}
-
-func testNameservers(t *testing.T) {
 	//common.SetLogLevel("debug")

 	lookupTimeout := 10 // ms
@@ -87,7 +84,7 @@ func testNameservers(t *testing.T) {
 	badNameservers := nameservers[25:]
 	// This subset will remain well-connected, and we will deal mainly with them
 	nameservers = nameservers[:25]
-	nameserversByName := map[router.PeerName]*Nameserver{}
+	nameserversByName := map[mesh.PeerName]*Nameserver{}
 	for _, n := range nameservers {
 		nameserversByName[n.ourName] = n
 	}
@@ -120,7 +117,7 @@ func testNameservers(t *testing.T) {
 		mapping := mapping{hostname, []pair{{nameserver.ourName, addr}}}
 		mappings = append(mappings, mapping)

-		require.Nil(t, nameserver.AddEntry(hostname, "", nameserver.ourName, addr))
+		nameserver.AddEntry(hostname, "", nameserver.ourName, addr)
 		check(nameserver, mapping)
 	}

@@ -135,14 +132,14 @@ func testNameservers(t *testing.T) {
 		mapping.addrs = append(mapping.addrs, pair{nameserver.ourName, addr})
 		mappings[i] = mapping

-		require.Nil(t, nameserver.AddEntry(mapping.hostname, "", nameserver.ourName, addr))
+		nameserver.AddEntry(mapping.hostname, "", nameserver.ourName, addr)
 		check(nameserver, mapping)
 	}

 	loseConnection := func() {
 		nameserver1 := badNameservers[rand.Intn(len(badNameservers))]
 		nameserver2 := nameservers[rand.Intn(len(nameservers))]
-		nameserver1.PeerGone(&router.Peer{Name: nameserver2.ourName})
+		nameserver1.PeerGone(nameserver2.ourName)
 	}

 	deleteMapping := func() {
@@ -160,7 +157,7 @@ func testNameservers(t *testing.T) {
 		mappings[i] = mapping
 		nameserver := nameserversByName[pair.origin]

-		require.Nil(t, nameserver.Delete(mapping.hostname, "*", pair.addr.String(), pair.addr))
+		nameserver.Delete(mapping.hostname, "*", pair.addr.String(), pair.addr)
 		check(nameserver, mapping)
 	}

@@ -222,22 +219,20 @@ func testNameservers(t *testing.T) {
 }

 func TestContainerAndPeerDeath(t *testing.T) {
-	peername, err := router.PeerNameFromString("00:00:00:02:00:00")
+	peername, err := mesh.PeerNameFromString("00:00:00:02:00:00")
 	require.Nil(t, err)
-	nameserver := New(peername, nil, "")
+	nameserver := makeNameserver(peername)

-	err = nameserver.AddEntry("hostname", "containerid", peername, address.Address(0))
-	require.Nil(t, err)
+	nameserver.AddEntry("hostname", "containerid", peername, address.Address(0))
 	require.Equal(t, []address.Address{0}, nameserver.Lookup("hostname"))

 	nameserver.ContainerDied("containerid")
 	require.Equal(t, []address.Address{}, nameserver.Lookup("hostname"))

-	err = nameserver.AddEntry("hostname", "containerid", peername, address.Address(0))
-	require.Nil(t, err)
+	nameserver.AddEntry("hostname", "containerid", peername, address.Address(0))
 	require.Equal(t, []address.Address{0}, nameserver.Lookup("hostname"))

-	nameserver.PeerGone(&router.Peer{Name: peername})
+	nameserver.PeerGone(peername)
 	require.Equal(t, []address.Address{}, nameserver.Lookup("hostname"))
 }

@@ -246,19 +241,17 @@ func TestTombstoneDeletion(t *testing.T) {
 	defer func() { now = oldNow }()
 	now = func() int64 { return 1234 }

-	peername, err := router.PeerNameFromString("00:00:00:02:00:00")
+	peername, err := mesh.PeerNameFromString("00:00:00:02:00:00")
 	require.Nil(t, err)
-	nameserver := New(peername, nil, "")
+	nameserver := makeNameserver(peername)

-	err = nameserver.AddEntry("hostname", "containerid", peername, address.Address(0))
-	require.Nil(t, err)
+	nameserver.AddEntry("hostname", "containerid", peername, address.Address(0))
 	require.Equal(t, []address.Address{0}, nameserver.Lookup("hostname"))

 	nameserver.deleteTombstones()
 	require.Equal(t, []address.Address{0}, nameserver.Lookup("hostname"))

-	err = nameserver.Delete("hostname", "containerid", "", address.Address(0))
-	require.Nil(t, err)
+	nameserver.Delete("hostname", "containerid", "", address.Address(0))
 	require.Equal(t, []address.Address{}, nameserver.Lookup("hostname"))
 	require.Equal(t, l(Entries{Entry{
 		ContainerID: "containerid",
@@ -1,10 +1,11 @@
 package nameserver

 type Status struct {
-	Domain  string
-	Address string
-	TTL     uint32
-	Entries []EntryStatus
+	Domain   string
+	Upstream []string
+	Address  string
+	TTL      uint32
+	Entries  []EntryStatus
 }

 type EntryStatus struct {
@@ -35,8 +36,10 @@ func NewStatus(ns *Nameserver, dnsServer *DNSServer) *Status {
 			entry.Tombstone})
 	}

+	upstreamConfig, _ := dnsServer.upstream.Config()
 	return &Status{
 		dnsServer.domain,
+		upstreamConfig.Servers,
 		dnsServer.address,
 		dnsServer.ttl,
 		entryStatusSlice}
@@ -10,6 +10,7 @@ import (
 // Using 32-bit integer to represent IPv4 address
 type Address uint32
 type Offset uint32
+type Count uint32

 type Range struct {
 	Start, End Address // [Start, End); Start <= End
@@ -18,8 +19,8 @@ type Range struct {
 func NewRange(start Address, size Offset) Range {
 	return Range{Start: start, End: Add(start, size)}
 }
-func (r Range) Size() Offset               { return Subtract(r.End, r.Start) }
-func (r Range) String() string             { return fmt.Sprintf("[%s-%s)", r.Start, r.End) }
+func (r Range) Size() Count                { return Length(r.End, r.Start) }
+func (r Range) String() string             { return fmt.Sprintf("%s-%s", r.Start, r.End-1) }
 func (r Range) Overlaps(or Range) bool     { return !(r.Start >= or.End || r.End <= or.Start) }
 func (r Range) Contains(addr Address) bool { return addr >= r.Start && addr < r.End }

@@ -31,14 +32,81 @@ func (r Range) AsCIDRString() string {
 		}
 		prefixLen--
 	}
-	return CIDR{Start: r.Start, PrefixLen: prefixLen}.String()
+	return CIDR{Addr: r.Start, PrefixLen: prefixLen}.String()
+}
+
+// return the highest bit set in v
+// algorithm from http://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2
+func msb(v Count) Count {
+	v |= v >> 1
+	v |= v >> 2
+	v |= v >> 4
+	v |= v >> 8
+	v |= v >> 16
+	// Cast to 64-bit to avoid overflow when original v >= 0x80000000
+	return Count((uint64(v) + 1) / 2)
+}
+
+func (r Range) BiggestCIDRRange() Range {
+	sizeMsb := Offset(msb(r.Size()))
+	maskedSize := Offset(r.Size()) & (sizeMsb - 1)
+	maskedStart := Offset(r.Start) & (sizeMsb - 1)
+	if maskedStart == 0 {
+		return NewRange(r.Start, sizeMsb)
+	}
+	if sizeMsb-maskedStart <= maskedSize {
+		return NewRange(Add(r.Start, sizeMsb-maskedStart), sizeMsb)
+	}
+	size := sizeMsb / 2
+	if size >= maskedStart {
+		return NewRange(Add(r.Start, size-maskedStart), size)
+	}
+	return NewRange(Add(r.Start, sizeMsb-maskedStart), size)
+}
+
+func MakeCIDR(subnet CIDR, addr Address) CIDR {
+	return CIDR{Addr: addr, PrefixLen: subnet.PrefixLen}
 }

 type CIDR struct {
-	Start     Address
+	Addr      Address
 	PrefixLen int
 }

+// CIDRs returns a list of CIDR-aligned ranges which cover this range.
+func (r Range) CIDRs() []CIDR {
+	const (
+		fullMask         = ^Address(0)
+		cidrMaxPrefixLen = 32
+	)
+	var cidrs []CIDR
+
+	for start, end := r.Start, r.End-1; end >= start; {
+		mask, prefixLen := fullMask, cidrMaxPrefixLen
+		// Find the smallest mask which would cover some part of [start;end].
+		// Once we found such, apply it by OR'ing
+		for mask > 0 {
+			tmpMask := mask << 1
+			// Check whether mask neither too short nor too long
+			if (start&tmpMask) != start || (start|^tmpMask) > end {
+				break
+			}
+			mask = tmpMask
+			prefixLen--
+		}
+		cidrs = append(cidrs, CIDR{start, prefixLen})
+		// Apply mask
+		start |= ^mask
+		// Check for overflow
+		if start+1 < start {
+			break
+		}
+		start++
+	}
+
+	return cidrs
+}
+
 func ParseIP(s string) (Address, error) {
 	if ip := net.ParseIP(s); ip != nil {
 		return FromIP4(ip), nil
@@ -46,29 +114,55 @@ func ParseIP(s string) (Address, error) {
 	return 0, &net.ParseError{Type: "IP Address", Text: s}
 }

-func ParseCIDR(s string) (Address, CIDR, error) {
+func ParseCIDR(s string) (CIDR, error) {
 	if ip, ipnet, err := net.ParseCIDR(s); err != nil {
-		return 0, CIDR{}, err
+		return CIDR{}, err
 	} else if ipnet.IP.To4() == nil {
-		return 0, CIDR{}, &net.ParseError{Type: "Non-IPv4 address not supported", Text: s}
+		return CIDR{}, &net.ParseError{Type: "Non-IPv4 address not supported", Text: s}
 	} else {
 		prefixLen, _ := ipnet.Mask.Size()
-		return FromIP4(ip), CIDR{Start: FromIP4(ipnet.IP), PrefixLen: prefixLen}, nil
+		return CIDR{Addr: FromIP4(ip), PrefixLen: prefixLen}, nil
 	}
 }

+func NewCIDRs(ranges []Range) (cidrs []CIDR) {
+	for _, r := range ranges {
+		cidrs = append(cidrs, r.CIDRs()...)
+	}
+	return cidrs
+}
+
+func (cidr CIDR) Start() Address {
+	return cidr.Addr
+}
+
+// cidr = [Start; End)
+func (cidr CIDR) End() Address {
+	return cidr.Range().End
+}
+
+func (cidr CIDR) IsSubnet() bool {
+	mask := cidr.Size() - 1
+	return Offset(cidr.Addr)&mask == 0
+}
+
 func (cidr CIDR) Size() Offset { return 1 << uint(32-cidr.PrefixLen) }

 func (cidr CIDR) Range() Range {
-	return NewRange(cidr.Start, cidr.Size())
+	return NewRange(cidr.Addr, cidr.Size())
 }
 func (cidr CIDR) HostRange() Range {
 	// Respect RFC1122 exclusions of first and last addresses
-	return NewRange(cidr.Start+1, cidr.Size()-2)
+	return NewRange(cidr.Addr+1, cidr.Size()-2)
 }

 func (cidr CIDR) String() string {
-	return fmt.Sprintf("%s/%d", cidr.Start.String(), cidr.PrefixLen)
+	return fmt.Sprintf("%s/%d", cidr.Addr.String(), cidr.PrefixLen)
+}
+
+func (cidr CIDR) IPNet() (r *net.IPNet) {
+	mask := net.CIDRMask(cidr.PrefixLen, 8*net.IPv4len)
+	return &net.IPNet{IP: cidr.Addr.IP4(), Mask: mask}
 }

 // FromIP4 converts an ipv4 address to our integer address type
@@ -107,7 +201,12 @@ func Subtract(a, b Address) Offset {
 	return Offset(a - b)
 }

-func Min(a, b Offset) Offset {
+func Length(a, b Address) Count {
+	common.Assert(a >= b)
+	return Count(a - b)
+}
+
+func Min(a, b Count) Count {
 	if a > b {
 		return b
 	}
@@ -0,0 +1,82 @@
+package address
+
+import (
+	"net"
+	"testing"
+	"testing/quick"
+
+	"github.com/stretchr/testify/require"
+)
+
+func isPower2(x Count) bool {
+	if x == 0 {
+		return false
+	}
+	for ; x > 1; x /= 2 {
+		if x&1 != 0 {
+			return false
+		}
+	}
+	return true
+}
+
+func TestBiggestPow2AlignedRange(t *testing.T) {
+	require.Equal(t, NewRange(0, 1), NewRange(0, 1).BiggestCIDRRange())
+	require.Equal(t, NewRange(1, 1), NewRange(1, 2).BiggestCIDRRange())
+	require.Equal(t, NewRange(2, 2), NewRange(1, 3).BiggestCIDRRange())
+	require.Equal(t, NewRange(0, 0x40000000), NewRange(0, 0x7fffffff).BiggestCIDRRange())
+	require.Equal(t, NewRange(0xfffffffe, 1), NewRange(0xfffffffe, 1).BiggestCIDRRange())
+	prop := func(start Address, size Offset) bool {
+		if size > Offset(0xffffffff)-Offset(start) { // out of range
+			return true
+		}
+		r := NewRange(start, size)
+		result := r.BiggestCIDRRange()
+		return r.Contains(result.Start) &&
+			r.Contains(result.End-1) &&
+			isPower2(result.Size()) &&
+			result.Size() > r.Size()/4 &&
+			Count(result.Start)%result.Size() == 0
+	}
+	require.NoError(t, quick.Check(prop, &quick.Config{MaxCount: 1000000}))
+}
+
+func ip(s string) Address {
+	addr, _ := ParseIP(s)
+	return addr
+}
+
+func cidr(s string) CIDR {
+	c, err := ParseCIDR(s)
+	if err != nil {
+		panic(err)
+	}
+	return c
+}
+
+func TestCIDRs(t *testing.T) {
+	start := ip("10.0.0.1")
+	end := ip("10.0.0.9")
+	r := NewRange(start, Subtract(end, start))
+	require.Equal(t,
+		[]CIDR{cidr("10.0.0.1/32"), cidr("10.0.0.2/31"), cidr("10.0.0.4/30"), cidr("10.0.0.8/32")},
+		r.CIDRs())
+}
+
+func TestCIDRStartAndEnd(t *testing.T) {
+	cidr, _ := ParseCIDR("10.0.0.0/24")
+	require.Equal(t, ip("10.0.0.0"), cidr.Start(), "")
+	require.Equal(t, ip("10.0.1.0"), cidr.End(), "")
+}
+
+func TestConversions(t *testing.T) {
+	addrStr := "10.2.3.0"
+	cidrStr := addrStr + "/24"
+	addr := ip(addrStr)
+	require.Equal(t, addr.String(), addrStr)
+	require.Equal(t, addr.IP4(), net.ParseIP(addrStr).To4())
+	cidr, _ := ParseCIDR(cidrStr)
+	require.Equal(t, cidr.String(), cidrStr)
+	_, ipnet, _ := net.ParseCIDR(cidrStr)
+	require.Equal(t, cidr.IPNet(), ipnet)
+}
@@ -0,0 +1,46 @@
+package net
+
+import (
+	"fmt"
+	"io"
+	"os"
+	"path/filepath"
+)
+
+// Configure the ARP cache parameters for the given interface.  This
+// makes containers react more quickly to a change in the MAC address
+// associated with an IP address.
+func ConfigureARPCache(procPath, name string) error {
+	if err := sysctl(procPath, fmt.Sprintf("net/ipv4/neigh/%s/base_reachable_time", name), "5"); err != nil {
+		return err
+	}
+	if err := sysctl(procPath, fmt.Sprintf("net/ipv4/neigh/%s/delay_first_probe_time", name), "2"); err != nil {
+		return err
+	}
+	if err := sysctl(procPath, fmt.Sprintf("net/ipv4/neigh/%s/ucast_solicit", name), "1"); err != nil {
+		return err
+	}
+	return nil
+}
+
+func sysctl(procPath, variable, value string) error {
+	f, err := os.OpenFile(filepath.Join(procPath, "/sys/", variable), os.O_WRONLY, 0)
+	if err != nil {
+		return err
+	}
+
+	n, err := io.WriteString(f, value)
+	if err != nil {
+		return err
+	}
+	if n < len(value) {
+		return io.ErrShortWrite
+	}
+
+	err = f.Close()
+	if err != nil {
+		return err
+	}
+
+	return nil
+}
@@ -0,0 +1,521 @@
+package net
+
+import (
+	"fmt"
+	"io/ioutil"
+	"net"
+	"path/filepath"
+
+	"github.com/Sirupsen/logrus"
+	"github.com/coreos/go-iptables/iptables"
+	"github.com/j-keck/arping"
+	"github.com/pkg/errors"
+	"github.com/vishvananda/netlink"
+
+	"github.com/weaveworks/weave/common/odp"
+)
+
+/* This code implements three possible configurations to connect
+   containers to the Weave Net overlay:
+
+1. Bridge
+                 +-------+
+(container-veth)-+ weave +-(vethwe-bridge)--(vethwe-pcap)
+                 +-------+
+
+"weave" is a Linux bridge. "vethwe-pcap" (end of veth pair) is used
+to capture and inject packets, by router/pcap.go.
+
+2. BridgedFastdp
+
+                 +-------+                                    /----------\
+(container-veth)-+ weave +-(vethwe-bridge)--(vethwe-datapath)-+ datapath +
+                 +-------+                                    \----------/
+
+"weave" is a Linux bridge and "datapath" is an Open vSwitch datapath;
+they are connected via a veth pair. Packet capture and injection use
+the "datapath" device, via "router/fastdp.go:fastDatapathBridge"
+
+3. Fastdp
+
+                 /-------\
+(container-veth)-+ weave +
+                 \-------/
+
+"weave" is an Open vSwitch datapath, and capture/injection are as in
+BridgedFastdp. Not used by default due to missing conntrack support in
+datapath of old kernel versions (https://github.com/weaveworks/weave/issues/1577).
+*/
+
+const (
+	WeaveBridgeName = "weave"
+	DatapathName    = "datapath"
+	DatapathIfName  = "vethwe-datapath"
+	BridgeIfName    = "vethwe-bridge"
+	PcapIfName      = "vethwe-pcap"
+)
+
+type Bridge interface {
+	init(config *BridgeConfig) error // create and initialise bridge device(s)
+	attach(veth *netlink.Veth) error // attach veth to bridge
+	IsFastdp() bool                  // does this bridge use fastdp?
+	String() string                  // human-readable type string
+}
+
+// Used to indicate a fallback to the Bridge type
+var errBridgeNotSupported = errors.New("bridge not supported")
+
+type bridgeImpl struct{ bridge netlink.Link }
+type fastdpImpl struct{ datapathName string }
+type bridgedFastdpImpl struct {
+	bridgeImpl
+	fastdpImpl
+}
+
+// Returns a string that is consistent with the weave script
+func (bridgeImpl) String() string        { return "bridge" }
+func (fastdpImpl) String() string        { return "fastdp" }
+func (bridgedFastdpImpl) String() string { return "bridged_fastdp" }
+
+// Used to decide whether to manage ODP tunnels
+func (bridgeImpl) IsFastdp() bool        { return false }
+func (fastdpImpl) IsFastdp() bool        { return true }
+func (bridgedFastdpImpl) IsFastdp() bool { return true }
+
+func ExistingBridgeType(weaveBridgeName, datapathName string) (Bridge, error) {
+	bridge, _ := netlink.LinkByName(weaveBridgeName)
+	datapath, _ := netlink.LinkByName(datapathName)
+
+	switch {
+	case bridge == nil && datapath == nil:
+		return nil, nil
+	case isBridge(bridge) && datapath == nil:
+		return bridgeImpl{bridge: bridge}, nil
+	case isDatapath(bridge) && datapath == nil:
+		return fastdpImpl{datapathName: datapathName}, nil
+	case isDatapath(datapath) && isBridge(bridge):
+		return bridgedFastdpImpl{bridgeImpl{bridge: bridge}, fastdpImpl{datapathName: datapathName}}, nil
+	default:
+		return nil, errors.New("Inconsistent bridge state detected. Please do 'weave reset' and try again")
+	}
+}
+
+func EnforceAddrAssignType(bridgeName string) (setAddr bool, err error) {
+	sysctlFilename := filepath.Join("/sys/class/net/", bridgeName, "/addr_assign_type")
+	addrAssignType, err := ioutil.ReadFile(sysctlFilename)
+	if err != nil {
+		return false, errors.Wrapf(err, "reading %q", sysctlFilename)
+	}
+
+	// From include/uapi/linux/netdevice.h
+	// #define NET_ADDR_PERM       0   /* address is permanent (default) */
+	// #define NET_ADDR_RANDOM     1   /* address is generated randomly */
+	// #define NET_ADDR_STOLEN     2   /* address is stolen from other device */
+	// #define NET_ADDR_SET        3   /* address is set using dev_set_mac_address() */
+	// Note the file typically has a newline at the end, so we just look at the first char
+	if addrAssignType[0] != '3' {
+		link, err := netlink.LinkByName(bridgeName)
+		if err != nil {
+			return false, errors.Wrapf(err, "EnforceAddrAssignType finding bridge %s", bridgeName)
+		}
+
+		mac, err := RandomMAC()
+		if err != nil {
+			return false, errors.Wrap(err, "creating random MAC")
+		}
+
+		if err := netlink.LinkSetHardwareAddr(link, mac); err != nil {
+			return false, errors.Wrapf(err, "setting bridge %s address to %v", bridgeName, mac)
+		}
+		return true, nil
+	}
+
+	return false, nil
+}
+
+func isBridge(link netlink.Link) bool {
+	_, isBridge := link.(*netlink.Bridge)
+	return isBridge
+}
+
+func isDatapath(link netlink.Link) bool {
+	switch link.(type) {
+	case *netlink.GenericLink:
+		return link.Type() == "openvswitch"
+	case *netlink.Device:
+		// Assume it's our openvswitch device, and the kernel has not been updated to report the kind.
+		return true
+	default:
+		return false
+	}
+}
+
+func DetectHairpin(portIfName string, log *logrus.Logger) error {
+	link, err := netlink.LinkByName(portIfName)
+	if err != nil {
+		return fmt.Errorf("Unable to find link %q: %s", portIfName, err)
+	}
+
+	ch := make(chan netlink.LinkUpdate)
+	// See EnsureInterface for why done channel is not passed
+	if err := netlink.LinkSubscribe(ch, nil); err != nil {
+		return fmt.Errorf("Unable to subscribe to netlink updates: %s", err)
+	}
+
+	pi, err := netlink.LinkGetProtinfo(link)
+	if err != nil {
+		return fmt.Errorf("Unable to get link protinfo %q: %s", portIfName, err)
+	}
+	if pi.Hairpin {
+		return fmt.Errorf("Hairpin mode enabled on %q", portIfName)
+	}
+
+	go func() {
+		for up := range ch {
+			if up.Attrs().Name == portIfName && up.Attrs().Protinfo != nil &&
+				up.Attrs().Protinfo.Hairpin {
+				log.Errorf("Hairpin mode enabled on %q", portIfName)
+			}
+		}
+	}()
+
+	return nil
+}
+
+var ErrBridgeNoIP = fmt.Errorf("Bridge has no IP address")
+
+func FindBridgeIP(bridgeName string, subnet *net.IPNet) (net.IP, error) {
+	netdev, err := GetBridgeNetDev(bridgeName)
+	if err != nil {
+		return nil, fmt.Errorf("Failed to get netdev for %q bridge: %s", bridgeName, err)
+	}
+	if len(netdev.CIDRs) == 0 {
+		return nil, ErrBridgeNoIP
+	}
+	if subnet != nil {
+		for _, cidr := range netdev.CIDRs {
+			if subnet.Contains(cidr.IP) {
+				return cidr.IP, nil
+			}
+		}
+	}
+	// No subnet, or none in the required subnet; just return the first one
+	return netdev.CIDRs[0].IP, nil
+}
+
+type BridgeConfig struct {
+	DockerBridgeName string
+	WeaveBridgeName  string
+	DatapathName     string
+	NoFastdp         bool
+	NoBridgedFastdp  bool
+	AWSVPC           bool
+	NPC              bool
+	MTU              int
+	Mac              string
+	Port             int
+}
+
+func (config *BridgeConfig) configuredBridgeType() Bridge {
+	switch {
+	case config.NoFastdp:
+		return bridgeImpl{}
+	case config.NoBridgedFastdp:
+		return fastdpImpl{datapathName: config.WeaveBridgeName}
+	default:
+		return bridgedFastdpImpl{fastdpImpl: fastdpImpl{datapathName: config.DatapathName}}
+	}
+}
+
+func EnsureBridge(procPath string, config *BridgeConfig, log *logrus.Logger) (Bridge, error) {
+	bridgeType, err := ExistingBridgeType(config.WeaveBridgeName, config.DatapathName)
+	if bridgeType != nil || err != nil {
+		return bridgeType, err
+	}
+
+	bridgeType = config.configuredBridgeType()
+	for {
+		if err := bridgeType.init(config); err != nil {
+			if errors.Cause(err) == errBridgeNotSupported {
+				log.Warnf("Skipping bridge creation of %q due to: %s", bridgeType, err)
+				bridgeType = bridgeImpl{}
+				continue
+			}
+			return nil, err
+		}
+		break
+	}
+
+	if err := configureIPTables(config); err != nil {
+		return bridgeType, errors.Wrap(err, "configuring iptables")
+	}
+
+	if config.AWSVPC {
+		// 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.
+		if err := sysctl(procPath, "net/ipv4/conf/"+config.WeaveBridgeName+"/proxy_arp", "1"); err != nil {
+			return bridgeType, errors.Wrap(err, "setting 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
+		if err := sysctl(procPath, "net/ipv4/neigh/"+config.WeaveBridgeName+"/proxy_delay", "0"); err != nil {
+			return bridgeType, errors.Wrap(err, "setting proxy_arp")
+		}
+	}
+
+	if err := linkSetUpByName(config.WeaveBridgeName); err != nil {
+		return bridgeType, err
+	}
+
+	if err := ConfigureARPCache(procPath, config.WeaveBridgeName); err != nil {
+		return bridgeType, errors.Wrapf(err, "configuring ARP cache on bridge %q", config.WeaveBridgeName)
+	}
+
+	return bridgeType, nil
+}
+
+func (b bridgeImpl) initPrep(config *BridgeConfig) error {
+	mac, err := net.ParseMAC(config.Mac)
+	if err != nil {
+		return errors.Wrapf(err, "parsing bridge MAC %q", config.Mac)
+	}
+
+	linkAttrs := netlink.NewLinkAttrs()
+	linkAttrs.Name = config.WeaveBridgeName
+	linkAttrs.HardwareAddr = mac
+	if config.MTU == 0 {
+		config.MTU = 65535
+	}
+	b.bridge = &netlink.Bridge{LinkAttrs: linkAttrs}
+	if err = netlink.LinkAdd(b.bridge); err != nil {
+		return errors.Wrapf(err, "creating bridge %q", config.WeaveBridgeName)
+	}
+	if err := netlink.LinkSetHardwareAddr(b.bridge, mac); err != nil {
+		return errors.Wrapf(err, "setting bridge %q mac %v", config.WeaveBridgeName, mac)
+	}
+	// Attempting to set the bridge MTU to a high value directly
+	// fails. Bridges take the lowest MTU of their interfaces. So
+	// instead we create a temporary interface with the desired MTU,
+	// attach that to the bridge, and then remove it again.
+	dummy := &netlink.Dummy{LinkAttrs: netlink.NewLinkAttrs()}
+	dummy.LinkAttrs.Name = "vethwedu"
+	if err = netlink.LinkAdd(dummy); err != nil {
+		return errors.Wrap(err, "creating dummy interface")
+	}
+	if err := netlink.LinkSetMTU(dummy, config.MTU); err != nil {
+		return errors.Wrapf(err, "setting dummy interface mtu to %d", config.MTU)
+	}
+	if err := netlink.LinkSetMasterByIndex(dummy, b.bridge.Attrs().Index); err != nil {
+		return errors.Wrap(err, "setting dummy interface master")
+	}
+	if err := netlink.LinkDel(dummy); err != nil {
+		return errors.Wrap(err, "deleting dummy interface")
+	}
+
+	return nil
+}
+
+func (b bridgeImpl) init(config *BridgeConfig) error {
+	if err := b.initPrep(config); err != nil {
+		return err
+	}
+	if _, err := CreateAndAttachVeth(BridgeIfName, PcapIfName, config.WeaveBridgeName, config.MTU, true, func(veth netlink.Link) error {
+		return netlink.LinkSetUp(veth)
+	}); err != nil {
+		return errors.Wrap(err, "creating pcap veth pair")
+	}
+	if err := EthtoolTXOff(config.WeaveBridgeName); err != nil {
+		return errors.Wrap(err, "setting tx off")
+	}
+
+	return nil
+}
+
+func (f fastdpImpl) init(config *BridgeConfig) error {
+	odpSupported, err := odp.CreateDatapath(f.datapathName)
+	if !odpSupported {
+		msg := ""
+		if err != nil {
+			msg = err.Error()
+		}
+		return errors.Wrap(errBridgeNotSupported, msg)
+	}
+	if err != nil {
+		return errors.Wrapf(err, "creating datapath %q", f.datapathName)
+	}
+	datapath, err := netlink.LinkByName(f.datapathName)
+	if err != nil {
+		return errors.Wrapf(err, "finding datapath %q", f.datapathName)
+	}
+	if config.MTU == 0 {
+		/* GCE has the lowest underlay network MTU we're likely to encounter on
+		   a local network, at 1460 bytes.  To get the overlay MTU from that we
+		   subtract 20 bytes for the outer IPv4 header, 8 bytes for the outer
+		   UDP header, 8 bytes for the vxlan header, and 14 bytes for the inner
+		   ethernet header.  In addition, we subtract 34 bytes for the ESP overhead
+		   which is needed for the vxlan encryption. */
+		config.MTU = 1376
+	}
+	if err := netlink.LinkSetMTU(datapath, config.MTU); err != nil {
+		return errors.Wrapf(err, "setting datapath %q mtu %d", f.datapathName, config.MTU)
+	}
+	return nil
+}
+
+func (bf bridgedFastdpImpl) init(config *BridgeConfig) error {
+	if err := bf.fastdpImpl.init(config); err != nil {
+		return err
+	}
+	if err := bf.bridgeImpl.initPrep(config); err != nil {
+		return err
+	}
+	if _, err := CreateAndAttachVeth(BridgeIfName, DatapathIfName, config.WeaveBridgeName, config.MTU, true, func(veth netlink.Link) error {
+		if err := netlink.LinkSetUp(veth); err != nil {
+			return errors.Wrapf(err, "setting link up on %q", veth.Attrs().Name)
+		}
+		if err := odp.AddDatapathInterface(bf.datapathName, veth.Attrs().Name); err != nil {
+			return errors.Wrapf(err, "adding interface %q to datapath %q", veth.Attrs().Name, bf.datapathName)
+		}
+		return nil
+	}); err != nil {
+		return errors.Wrap(err, "creating bridged fastdp veth pair")
+	}
+
+	if err := linkSetUpByName(bf.datapathName); err != nil {
+		return err
+	}
+
+	return nil
+}
+
+func (b bridgeImpl) attach(veth *netlink.Veth) error {
+	return netlink.LinkSetMasterByIndex(veth, b.bridge.Attrs().Index)
+}
+
+func (bf bridgedFastdpImpl) attach(veth *netlink.Veth) error {
+	return bf.bridgeImpl.attach(veth)
+}
+
+func (f fastdpImpl) attach(veth *netlink.Veth) error {
+	return odp.AddDatapathInterface(f.datapathName, veth.Attrs().Name)
+}
+
+func configureIPTables(config *BridgeConfig) error {
+	ipt, err := iptables.New()
+	if err != nil {
+		return errors.Wrap(err, "creating iptables object")
+	}
+	if config.DockerBridgeName != "" {
+		if config.WeaveBridgeName != config.DockerBridgeName {
+			err := ipt.Insert("filter", "FORWARD", 1, "-i", config.DockerBridgeName, "-o", config.WeaveBridgeName, "-j", "DROP")
+			if err != nil {
+				return err
+			}
+		}
+
+		dockerBridgeIP, err := FindBridgeIP(config.DockerBridgeName, nil)
+		if err != nil {
+			return err
+		}
+
+		// forbid traffic to the Weave port from other containers
+		if err = ipt.AppendUnique("filter", "INPUT", "-i", config.DockerBridgeName, "-p", "tcp", "--dst", dockerBridgeIP.String(), "--dport", fmt.Sprint(config.Port), "-j", "DROP"); err != nil {
+			return err
+		}
+		if err = ipt.AppendUnique("filter", "INPUT", "-i", config.DockerBridgeName, "-p", "udp", "--dst", dockerBridgeIP.String(), "--dport", fmt.Sprint(config.Port), "-j", "DROP"); err != nil {
+			return err
+		}
+		if err = ipt.AppendUnique("filter", "INPUT", "-i", config.DockerBridgeName, "-p", "udp", "--dst", dockerBridgeIP.String(), "--dport", fmt.Sprint(config.Port+1), "-j", "DROP"); err != nil {
+			return err
+		}
+
+		// let DNS traffic to weaveDNS, since otherwise it might get blocked by the likes of UFW
+		if err = ipt.AppendUnique("filter", "INPUT", "-i", config.DockerBridgeName, "-p", "udp", "--dport", "53", "-j", "ACCEPT"); err != nil {
+			return err
+		}
+		if err = ipt.AppendUnique("filter", "INPUT", "-i", config.DockerBridgeName, "-p", "tcp", "--dport", "53", "-j", "ACCEPT"); err != nil {
+			return err
+		}
+	}
+
+	if config.NPC {
+		// Steer traffic via the NPC
+		_ = ipt.NewChain("filter", "WEAVE-NPC") // ignore error because it might already exist
+		// If WEAVE-NPC chain doesn't exist then next line will fail
+		if err = ipt.AppendUnique("filter", "FORWARD", "-o", config.WeaveBridgeName, "-j", "WEAVE-NPC"); err != nil {
+			return err
+		}
+		if err = ipt.AppendUnique("filter", "FORWARD", "-o", config.WeaveBridgeName, "-m", "state", "--state", "NEW", "-j", "NFLOG", "--nflog-group", "86"); err != nil {
+			return err
+		}
+		if err = ipt.AppendUnique("filter", "FORWARD", "-o", config.WeaveBridgeName, "-j", "DROP"); err != nil {
+			return err
+		}
+	} else {
+		// Work around the situation where there are no rules allowing traffic
+		// across our bridge. E.g. ufw
+		if err = ipt.AppendUnique("filter", "FORWARD", "-i", config.WeaveBridgeName, "-o", config.WeaveBridgeName, "-j", "ACCEPT"); err != nil {
+			return err
+		}
+	}
+	// Forward from weave to the rest of the world
+	if err = ipt.AppendUnique("filter", "FORWARD", "-i", config.WeaveBridgeName, "!", "-o", config.WeaveBridgeName, "-j", "ACCEPT"); err != nil {
+		return err
+	}
+	// and allow replies back
+	if err = ipt.AppendUnique("filter", "FORWARD", "-o", config.WeaveBridgeName, "-m", "conntrack", "--ctstate", "RELATED,ESTABLISHED", "-j", "ACCEPT"); err != nil {
+		return err
+	}
+
+	// create a chain for masquerading
+	if err = ipt.ClearChain("nat", "WEAVE"); err != nil {
+		return errors.Wrap(err, "clearing WEAVE chain")
+	}
+	if err = ipt.AppendUnique("nat", "POSTROUTING", "-j", "WEAVE"); err != nil {
+		return err
+	}
+
+	return nil
+}
+
+func linkSetUpByName(linkName string) error {
+	link, err := netlink.LinkByName(linkName)
+	if err != nil {
+		return errors.Wrapf(err, "setting link up on %q", linkName)
+	}
+	return netlink.LinkSetUp(link)
+}
+
+func AddBridgeAddr(bridgeName string, addr *net.IPNet, removeDefaultRoute bool) error {
+	link, err := netlink.LinkByName(bridgeName)
+	if err != nil {
+		return errors.Wrapf(err, "AddBridgeAddr finding bridge %q", bridgeName)
+	}
+	newAddresses, err := AddAddresses(link, []*net.IPNet{addr})
+	if err != nil {
+		return errors.Wrapf(err, "adding address %v to %q", addr, bridgeName)
+	}
+	for _, ipnet := range newAddresses {
+		arping.GratuitousArpOverIfaceByName(ipnet.IP, bridgeName)
+	}
+	if removeDefaultRoute {
+		routeFilter := &netlink.Route{
+			LinkIndex: link.Attrs().Index,
+			Dst:       &net.IPNet{IP: addr.IP.Mask(addr.Mask), Mask: addr.Mask},
+			Protocol:  2, // RTPROT_KERNEL
+		}
+		filterMask := netlink.RT_FILTER_OIF | netlink.RT_FILTER_DST | netlink.RT_FILTER_PROTOCOL
+		routes, err := netlink.RouteListFiltered(netlink.FAMILY_V4, routeFilter, filterMask)
+		if err != nil {
+			return errors.Wrapf(err, "failed to get route list for bridge %q", bridgeName)
+		}
+		for _, r := range routes {
+			if err = netlink.RouteDel(&r); err != nil {
+				return errors.Wrapf(err, "failed to delete default route %+v", r)
+			}
+		}
+	}
+	return nil
+}
@@ -0,0 +1,60 @@
+package net
+
+import "fmt"
+import "syscall"
+import "unsafe"
+
+const (
+	SIOCETHTOOL     = 0x8946     // linux/sockios.h
+	ETHTOOL_GTXCSUM = 0x00000016 // linux/ethtool.h
+	ETHTOOL_STXCSUM = 0x00000017 // linux/ethtool.h
+	IFNAMSIZ        = 16         // linux/if.h
+)
+
+// linux/if.h 'struct ifreq'
+type IFReqData struct {
+	Name [IFNAMSIZ]byte
+	Data uintptr
+}
+
+// linux/ethtool.h 'struct ethtool_value'
+type EthtoolValue struct {
+	Cmd  uint32
+	Data uint32
+}
+
+func ioctlEthtool(fd int, argp uintptr) error {
+	_, _, errno := syscall.RawSyscall(syscall.SYS_IOCTL, uintptr(fd), uintptr(SIOCETHTOOL), argp)
+	if errno != 0 {
+		return errno
+	}
+	return nil
+}
+
+// Disable TX checksum offload on specified interface
+func EthtoolTXOff(name string) error {
+	if len(name)+1 > IFNAMSIZ {
+		return fmt.Errorf("name too long")
+	}
+
+	socket, err := syscall.Socket(syscall.AF_INET, syscall.SOCK_DGRAM, 0)
+	if err != nil {
+		return err
+	}
+	defer syscall.Close(socket)
+
+	// Request current value
+	value := EthtoolValue{Cmd: ETHTOOL_GTXCSUM}
+	request := IFReqData{Data: uintptr(unsafe.Pointer(&value))}
+	copy(request.Name[:], name)
+
+	if err := ioctlEthtool(socket, uintptr(unsafe.Pointer(&request))); err != nil {
+		return err
+	}
+	if value.Data == 0 { // if already off, don't try to change
+		return nil
+	}
+
+	value = EthtoolValue{ETHTOOL_STXCSUM, 0}
+	return ioctlEthtool(socket, uintptr(unsafe.Pointer(&request)))
+}
@@ -5,28 +5,35 @@ import (
 	"net"
 	"syscall"

-	"github.com/vishvananda/netlink/nl"
+	"github.com/vishvananda/netlink"
 )

 // Wait for an interface to come up.
 func EnsureInterface(ifaceName string) (*net.Interface, error) {
-	s, err := nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_LINK)
-	if err != nil {
-		return nil, err
-	}
-	defer s.Close()
-	iface, err := ensureInterface(s, ifaceName)
+	iface, err := ensureInterface(ifaceName)
 	if err != nil {
 		return nil, err
 	}
 	return iface, err
 }

-func ensureInterface(s *nl.NetlinkSocket, ifaceName string) (*net.Interface, error) {
+func ensureInterface(ifaceName string) (*net.Interface, error) {
+	ch := make(chan netlink.LinkUpdate)
+	// NB: We do not supply (and eventually close) a 'done' channel
+	// here since that can cause incorrect file descriptor
+	// re-usage. See https://github.com/weaveworks/weave/issues/2120
+	if err := netlink.LinkSubscribe(ch, nil); err != nil {
+		return nil, err
+	}
+	// check for currently-existing interface after subscribing, to avoid race
 	if iface, err := findInterface(ifaceName); err == nil {
 		return iface, nil
 	}
-	netlinkReceiveUntil(s, matchIfName(ifaceName))
+	for update := range ch {
+		if ifaceName == update.Link.Attrs().Name && update.IfInfomsg.Flags&syscall.IFF_UP != 0 {
+			break
+		}
+	}
 	iface, err := findInterface(ifaceName)
 	return iface, err
 }
@@ -35,21 +42,35 @@ func ensureInterface(s *nl.NetlinkSocket, ifaceName string) (*net.Interface, err
 // This matches the behaviour in 'weave attach', which is the only context in which
 // we expect this to be called.  If you change one, change the other to match.
 func EnsureInterfaceAndMcastRoute(ifaceName string) (*net.Interface, error) {
-	s, err := nl.Subscribe(syscall.NETLINK_ROUTE, syscall.RTNLGRP_LINK, syscall.RTNLGRP_IPV4_ROUTE)
+	iface, err := ensureInterface(ifaceName)
 	if err != nil {
 		return nil, err
 	}
-	defer s.Close()
-	iface, err := ensureInterface(s, ifaceName)
-	if err != nil {
+	ch := make(chan netlink.RouteUpdate)
+	if err := netlink.RouteSubscribe(ch, nil); err != nil {
 		return nil, err
 	}
 	dest := net.IPv4(224, 0, 0, 0)
-	if CheckRouteExists(ifaceName, dest) {
-		return iface, err
+	check := func(route netlink.Route) bool {
+		return route.Dst != nil && route.Dst.IP.Equal(dest)
 	}
-	netlinkReceiveUntil(s, matchRoute(ifaceName, dest))
-	return iface, err
+	// check for currently-existing route after subscribing, to avoid race
+	routes, err := netlink.RouteList(nil, netlink.FAMILY_V4)
+	if err != nil {
+		return nil, err
+	}
+	for _, route := range routes {
+		if check(route) {
+			return iface, nil
+		}
+	}
+	for update := range ch {
+		if check(update.Route) {
+			return iface, nil
+		}
+	}
+	// should never get here
+	return iface, nil
 }

 func findInterface(ifaceName string) (iface *net.Interface, err error) {
@@ -62,39 +83,24 @@ func findInterface(ifaceName string) (iface *net.Interface, err error) {
 	return
 }

-func netlinkReceiveUntil(s *nl.NetlinkSocket, f func(syscall.NetlinkMessage) (bool, error)) error {
-	for {
-		msgs, err := s.Receive()
-		if err != nil {
-			return err
-		}
-		for _, m := range msgs {
-			if done, err := f(m); done {
-				return err
-			}
-		}
+func LocalAddresses() ([]*net.IPNet, error) {
+	links, err := netlink.LinkList()
+	if err != nil {
+		return nil, err
 	}
-}

-func matchIfName(ifaceName string) func(m syscall.NetlinkMessage) (bool, error) {
-	return func(m syscall.NetlinkMessage) (bool, error) {
-		switch m.Header.Type {
-		case syscall.RTM_NEWLINK: // receive this type for link 'up'
-			ifmsg := nl.DeserializeIfInfomsg(m.Data)
-			attrs, err := syscall.ParseNetlinkRouteAttr(&m)
-			if err != nil {
-				return true, err
-			}
-			name := ""
-			for _, attr := range attrs {
-				if attr.Attr.Type == syscall.IFA_LABEL {
-					name = string(attr.Value[:len(attr.Value)-1])
-				}
-			}
-			if ifaceName == name && ifmsg.Flags&syscall.IFF_UP != 0 {
-				return true, nil
-			}
+	retval := []*net.IPNet{}
+	for _, link := range links {
+		if _, isBridge := link.(*netlink.Bridge); isBridge || (link.Attrs().Flags&net.FlagLoopback) != 0 {
+			continue
+		}
+		addrs, err := netlink.AddrList(link, netlink.FAMILY_V4)
+		if err != nil {
+			return nil, err
+		}
+		for _, addr := range addrs {
+			retval = append(retval, addr.IPNet)
 		}
-		return false, nil
 	}
+	return retval, nil
 }
@@ -0,0 +1,562 @@
+// package IPsec provides primitives for establishing IPsec in the fastdp mode.
+package ipsec
+
+import (
+	"crypto/rand"
+	"crypto/sha256"
+	"encoding/binary"
+	"fmt"
+	"io"
+	"net"
+	"strconv"
+	"sync"
+	"syscall"
+
+	"github.com/Sirupsen/logrus"
+	"github.com/coreos/go-iptables/iptables"
+	"github.com/pkg/errors"
+	"github.com/vishvananda/netlink"
+	"golang.org/x/crypto/hkdf"
+
+	"github.com/weaveworks/mesh"
+)
+
+const (
+	keySize   = 36 // AES-GCM key 32 bytes + 4 bytes salt
+	nonceSize = 32 // HKDF nonce size
+
+	mark    = uint32(0x1) << 17 // iptables marks
+	markStr = "0x20000/0x20000" // update if the above mark changes
+
+	tableMangle  = "mangle"
+	tableFilter  = "filter"
+	chainIn      = "WEAVE-IPSEC-IN"
+	chainInMark  = "WEAVE-IPSEC-IN-MARK"
+	chainOut     = "WEAVE-IPSEC-OUT"
+	chainOutMark = "WEAVE-IPSEC-OUT-MARK"
+)
+
+type SPI uint32
+
+// Used to identify:
+// - directional SPIs,
+// - ipsec establishments.
+type spiID [24]byte
+
+func getSPIId(srcPeer, dstPeer mesh.PeerName, connUID uint64) (id spiID) {
+	binary.BigEndian.PutUint64(id[:], uint64(srcPeer))
+	binary.BigEndian.PutUint64(id[8:], uint64(dstPeer))
+	binary.BigEndian.PutUint64(id[16:], connUID)
+	return
+}
+
+type spiInfo struct {
+	spi      SPI
+	isDirOut bool
+}
+
+// IPSec
+
+type IPSec struct {
+	sync.RWMutex
+	ipt *iptables.IPTables
+	log *logrus.Logger
+
+	spiInfo map[spiID]spiInfo
+	// A reference to spiInfo; spiInfo might be of an expired SPI.
+	spis map[SPI]*spiInfo
+}
+
+func New(log *logrus.Logger) (*IPSec, error) {
+	ipt, err := iptables.New()
+	if err != nil {
+		return nil, errors.Wrap(err, "iptables new")
+	}
+
+	ipsec := &IPSec{
+		ipt:     ipt,
+		log:     log,
+		spiInfo: make(map[spiID]spiInfo),
+		spis:    make(map[SPI]*spiInfo),
+	}
+
+	return ipsec, nil
+}
+
+// InitSALocal initializes inbound ipsec from remotePeer and triggers
+// the initialization on remotePeer.
+func (ipsec *IPSec) InitSALocal(localPeer, remotePeer mesh.PeerName, connUID uint64, localIP, remoteIP net.IP, udpPort int, sessionKey *[32]byte, initRemote func([]byte) error) error {
+	// ID of inbound SPI
+	spiID := getSPIId(remotePeer, localPeer, connUID)
+
+	ipsec.Lock()
+	defer ipsec.Unlock()
+
+	// Derive SA key
+	nonce, err := genNonce()
+	if err != nil {
+		return errors.Wrap(err, "generate nonce")
+	}
+	key, err := deriveKey(sessionKey[:], nonce, localPeer)
+	if err != nil {
+		return errors.Wrap(err, "derive key")
+	}
+
+	// Allocate SA
+	sa, err := netlink.XfrmStateAllocSpi(xfrmAllocSpiState(remoteIP, localIP))
+	if err != nil {
+		return errors.Wrap(err, fmt.Sprintf("ip xfrm state allocspi (in, %s, %s)", remoteIP, localIP))
+	}
+
+	// Use SPI generated by the kernel. The kernel ensures {dstIP, spi} to
+	// be unique.
+	spi := SPI(sa.Spi)
+
+	ipsec.log.Infof("ipsec: InitSALocal: %s -> %s :%d 0x%x", remoteIP, localIP, udpPort, spi)
+
+	// Create SA
+	if sa, err := xfrmState(remoteIP, localIP, spi, false, key); err == nil {
+		if err := netlink.XfrmStateUpdate(sa); err != nil {
+			return errors.Wrap(err, fmt.Sprintf("xfrm state update (in, %s, %s, 0x%x)", sa.Src, sa.Dst, sa.Spi))
+		}
+	} else {
+		return errors.Wrap(err, "new xfrm state (in)")
+	}
+
+	// Install iptables rules
+	if err := ipsec.installDropNonEncrypted(localIP, remoteIP, udpPort, spi); err != nil {
+		return errors.Wrap(err, fmt.Sprintf("install protecting rules (%s, %s, %d, 0x%x)", localIP, remoteIP, udpPort, spi))
+	}
+
+	si := spiInfo{spi: spi, isDirOut: false}
+	ipsec.spiInfo[spiID] = si
+	ipsec.spis[spi] = &si
+
+	// Trigger the initialization on the remote peer
+	msg := &msgInitSARemote{nonce, spi}
+	if err := initRemote(msg.serialize()); err != nil {
+		return errors.Wrap(err, "send InitSARemote")
+	}
+
+	return nil
+}
+
+// InitSARemote initializes outbound ipsec to remotePeer.
+// Triggered by remotePeer.
+func (ipsec *IPSec) InitSARemote(msgInitSARemote []byte, localPeer, remotePeer mesh.PeerName, connUID uint64, localIP, remoteIP net.IP, udpPort int, sessionKey *[32]byte) error {
+	// ID of outbound SPI
+	spiID := getSPIId(localPeer, remotePeer, connUID)
+
+	msg, err := deserializeMsgInitSARemote(msgInitSARemote)
+	if err != nil {
+		return errors.Wrap(err, "deserialize InitSARemote")
+	}
+	spi := msg.spi
+
+	ipsec.Lock()
+	defer ipsec.Unlock()
+
+	ipsec.log.Infof("ipsec: InitSARemote: %s -> %s :%d 0x%x", localIP, remoteIP, udpPort, spi)
+
+	// Derive SA key by using the received nonce
+	key, err := deriveKey(sessionKey[:], msg.nonce, remotePeer)
+	if err != nil {
+		return errors.Wrap(err, "derive key")
+	}
+
+	// Create SA
+	if sa, err := xfrmState(localIP, remoteIP, spi, true, key); err == nil {
+		if err := netlink.XfrmStateAdd(sa); err != nil {
+			return errors.Wrap(err, fmt.Sprintf("xfrm state update (out, %s, %s, 0x%x)", sa.Src, sa.Dst, sa.Spi))
+		}
+	} else {
+		return errors.Wrap(err, "new xfrm state (out)")
+	}
+
+	// Create or update SP
+	sp := xfrmPolicy(localIP, remoteIP, spi)
+	if err := netlink.XfrmPolicyUpdate(sp); err != nil {
+		return errors.Wrap(err, fmt.Sprintf("xfrm policy update (%s, %s, 0x%x)", localIP, remoteIP, spi))
+	}
+
+	si := spiInfo{spi: spi, isDirOut: true}
+	ipsec.spiInfo[spiID] = si
+	ipsec.spis[spi] = &si
+
+	return nil
+}
+
+// Destroy destroys any (inbound / outbound) ipsec establishment between the peers.
+func (ipsec *IPSec) Destroy(localPeer, remotePeer mesh.PeerName, connUID uint64, localIP, remoteIP net.IP, udpPort int) error {
+	outSPIID := getSPIId(localPeer, remotePeer, connUID)
+	inSPIID := getSPIId(remotePeer, localPeer, connUID)
+
+	ipsec.Lock()
+	defer ipsec.Unlock()
+
+	// Destroy inbound
+
+	if inSPIInfo, ok := ipsec.spiInfo[inSPIID]; ok {
+		ipsec.log.Infof("ipsec: destroy: in %s -> %s 0x%x", remoteIP, localIP, inSPIInfo.spi)
+
+		inSPI := inSPIInfo.spi
+
+		inSA := &netlink.XfrmState{
+			Src:   remoteIP,
+			Dst:   localIP,
+			Proto: netlink.XFRM_PROTO_ESP,
+			Spi:   int(inSPI),
+		}
+		if err := netlink.XfrmStateDel(inSA); err != nil {
+			ipsec.log.Warnf("ipsec: xfrm state del (in, %s, %s, 0x%x) failed: %s", inSA.Src, inSA.Dst, inSA.Spi, err)
+		}
+
+		if err := ipsec.removeDropNonEncrypted(localIP, remoteIP, udpPort, inSPI); err != nil {
+			ipsec.log.Warnf("ipsec: remove protecting rules (%s, %s, %d, 0x%x) failed: %s", localIP, remoteIP, udpPort, inSPI, err)
+		}
+
+		delete(ipsec.spiInfo, inSPIID)
+		delete(ipsec.spis, inSPI)
+	}
+
+	// Destroy outbound
+
+	if outSPIInfo, ok := ipsec.spiInfo[outSPIID]; ok {
+		ipsec.log.Infof("ipsec: destroy: out %s -> %s 0x%x", localIP, remoteIP, outSPIInfo.spi)
+
+		if err := netlink.XfrmPolicyDel(xfrmPolicy(localIP, remoteIP, outSPIInfo.spi)); err != nil {
+			ipsec.log.Warnf("ipsec: xfrm policy del (%s, %s, 0x%x) failed: %s", localIP, remoteIP, outSPIInfo.spi, err)
+		}
+
+		outSA := &netlink.XfrmState{
+			Src:   localIP,
+			Dst:   remoteIP,
+			Proto: netlink.XFRM_PROTO_ESP,
+			Spi:   int(outSPIInfo.spi),
+		}
+		if err := netlink.XfrmStateDel(outSA); err != nil {
+			ipsec.log.Warnf("ipsec: xfrm state del (out, %s, %s, 0x%x) failed: %s", outSA.Src, outSA.Dst, outSA.Spi, err)
+		}
+
+		delete(ipsec.spiInfo, outSPIID)
+		delete(ipsec.spis, outSPIInfo.spi)
+	}
+
+	return nil
+}
+
+// Flush removes all policies/SAs established by us. Also, it removes chains and
+// rules of iptables.
+//
+// If destroy is true, the chains and the rules won't be re-created.
+func (ipsec *IPSec) Flush(destroy bool) error {
+	ipsec.Lock()
+	defer ipsec.Unlock()
+
+	policies, err := netlink.XfrmPolicyList(syscall.AF_INET)
+	if err != nil {
+		return errors.Wrap(err, "xfrm policy list")
+	}
+	for _, p := range policies {
+		if p.Mark != nil && p.Mark.Value == mark && len(p.Tmpls) != 0 {
+			spi := SPI(p.Tmpls[0].Spi)
+			if err := netlink.XfrmPolicyDel(&p); err != nil {
+				return errors.Wrap(err, fmt.Sprintf("xfrm policy del (%s, %s, 0x%x)", p.Src, p.Dst, spi))
+			}
+		}
+	}
+
+	states, err := netlink.XfrmStateList(syscall.AF_INET)
+	if err != nil {
+		return errors.Wrap(err, "xfrm state list")
+	}
+	for _, s := range states {
+		if _, ok := ipsec.spis[SPI(s.Spi)]; ok {
+			if err := netlink.XfrmStateDel(&s); err != nil {
+				return errors.Wrap(err, fmt.Sprintf("xfrm state list (%s, %s, 0x%x)", s.Src, s.Dst, s.Spi))
+			}
+		}
+	}
+
+	if err := ipsec.resetIPTables(destroy); err != nil {
+		return errors.Wrap(err, "reset ip tables")
+	}
+
+	return nil
+}
+
+// iptables
+
+type chain struct {
+	table string
+	chain string
+}
+type rule struct {
+	table    string
+	chain    string
+	rulespec []string
+	unique   bool
+}
+
+func (ipsec *IPSec) clearChains(chains []chain) error {
+	for _, c := range chains {
+		if err := ipsec.ipt.ClearChain(c.table, c.chain); err != nil {
+			return errors.Wrap(err, fmt.Sprintf("iptables clear chain (%s, %s)", c.table, c.chain))
+		}
+	}
+	return nil
+}
+
+func (ipsec *IPSec) deleteChains(chains []chain) error {
+	for _, c := range chains {
+		if err := ipsec.ipt.DeleteChain(c.table, c.chain); err != nil {
+			return errors.Wrap(err, fmt.Sprintf("iptables delete chain (%s, %s)", c.table, c.chain))
+		}
+	}
+	return nil
+}
+
+func (ipsec *IPSec) resetRules(rules []rule, destroy bool) error {
+	for _, r := range rules {
+		ok, err := ipsec.ipt.Exists(r.table, r.chain, r.rulespec...)
+		if err != nil {
+			return errors.Wrap(err, fmt.Sprintf("iptables exists rule (%s, %s, %s)", r.table, r.chain, r.rulespec))
+		}
+		switch {
+		case !destroy && !ok:
+			if err := ipsec.ipt.Append(r.table, r.chain, r.rulespec...); err != nil {
+				return errors.Wrap(err, fmt.Sprintf("iptables append rule (%s, %s, %s)", r.table, r.chain, r.rulespec))
+			}
+		case destroy && ok:
+			if err := ipsec.ipt.Delete(r.table, r.chain, r.rulespec...); err != nil {
+				return errors.Wrap(err, fmt.Sprintf("iptables delete rule (%s, %s, %s)", r.table, r.chain, r.rulespec))
+			}
+		}
+	}
+	return nil
+}
+
+func (ipsec *IPSec) resetIPTables(destroy bool) error {
+	chains := []chain{
+		{tableMangle, chainIn},
+		{tableMangle, chainInMark},
+		{tableFilter, chainIn},
+		{tableMangle, chainOut},
+		{tableMangle, chainOutMark},
+	}
+	rules := []rule{
+		{tableMangle, "INPUT", []string{"-j", chainIn}, true},
+		{tableMangle, chainInMark, []string{"-j", "MARK", "--set-xmark", markStr}, true},
+		{tableFilter, "INPUT", []string{"-j", chainIn}, true},
+		{tableMangle, "OUTPUT", []string{"-j", chainOut}, true},
+		{tableMangle, chainOutMark, []string{"-j", "MARK", "--set-xmark", markStr}, true},
+		{tableFilter, "OUTPUT",
+			[]string{
+				"!", "-p", "esp",
+				"-m", "policy", "--dir", "out", "--pol", "none",
+				"-m", "mark", "--mark", markStr,
+				"-j", "DROP"}, true},
+	}
+
+	if err := ipsec.clearChains(chains); err != nil {
+		return err
+	}
+
+	if err := ipsec.resetRules(rules, destroy); err != nil {
+		return err
+	}
+
+	if destroy {
+		if err := ipsec.deleteChains(chains); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
+
+func ruleMarkInboundESP(srcIP, dstIP net.IP, inSPI SPI) rule {
+	return rule{tableMangle, chainIn,
+		[]string{
+			"-s", dstIP.String(), "-d", srcIP.String(),
+			"-p", "esp",
+			"-m", "esp", "--espspi", "0x" + strconv.FormatUint(uint64(inSPI), 16),
+			"-j", chainInMark,
+		}, true}
+}
+
+func rulesDropNonEncrypted(srcIP, dstIP net.IP, udpPort int, inSPI SPI) []rule {
+	udpPortStr := strconv.FormatUint(uint64(udpPort), 10)
+	return []rule{
+		ruleMarkInboundESP(srcIP, dstIP, inSPI),
+		{tableFilter, chainIn,
+			[]string{
+				"-s", dstIP.String(), "-d", srcIP.String(),
+				"-p", "udp", "--dport", udpPortStr,
+				"-m", "mark", "!", "--mark", markStr,
+				"-j", "DROP",
+			}, false},
+		{tableMangle, chainOut,
+			[]string{
+				"-s", srcIP.String(), "-d", dstIP.String(),
+				"-p", "udp", "--dport", udpPortStr,
+				"-j", chainOutMark,
+			}, false},
+	}
+}
+
+func (ipsec *IPSec) installDropNonEncrypted(srcIP, dstIP net.IP, udpPort int, inSPI SPI) error {
+	rules := rulesDropNonEncrypted(srcIP, dstIP, udpPort, inSPI)
+	for _, r := range rules {
+		appendFunc := ipsec.ipt.Append
+		if r.unique {
+			appendFunc = ipsec.ipt.AppendUnique
+		}
+		if err := appendFunc(r.table, r.chain, r.rulespec...); err != nil {
+			return errors.Wrap(err, fmt.Sprintf("iptables append (%s, %s, %s)", r.table, r.chain, r.rulespec))
+		}
+	}
+	return nil
+}
+
+func (ipsec *IPSec) removeDropNonEncrypted(srcIP, dstIP net.IP, udpPort int, inSPI SPI) error {
+	rules := rulesDropNonEncrypted(srcIP, dstIP, udpPort, inSPI)
+	if err := ipsec.resetRules(rules, true); err != nil {
+		return err
+	}
+	return nil
+}
+
+func (ipsec *IPSec) removeDropNonEncryptedInbound(srcIP, dstIP net.IP, inSPI SPI) error {
+	r := ruleMarkInboundESP(srcIP, dstIP, inSPI)
+	if err := ipsec.ipt.Delete(r.table, r.chain, r.rulespec...); err != nil {
+		return errors.Wrap(err, fmt.Sprintf("iptables delete unique (%s, %s, %s)", r.table, r.chain, r.rulespec))
+	}
+	return nil
+}
+
+// xfrm
+
+func xfrmAllocSpiState(srcIP, dstIP net.IP) *netlink.XfrmState {
+	return &netlink.XfrmState{
+		Src:          srcIP,
+		Dst:          dstIP,
+		Proto:        netlink.XFRM_PROTO_ESP,
+		Mode:         netlink.XFRM_MODE_TRANSPORT,
+		ReplayWindow: 256,
+		ESN:          true,
+	}
+}
+
+func xfrmState(srcIP, dstIP net.IP, spi SPI, isDirOut bool, key []byte) (*netlink.XfrmState, error) {
+	if len(key) != keySize {
+		return nil, fmt.Errorf("key should be %d bytes long", keySize)
+	}
+
+	state := xfrmAllocSpiState(srcIP, dstIP)
+
+	state.Spi = int(spi)
+	state.Aead = &netlink.XfrmStateAlgo{
+		Name:   "rfc4106(gcm(aes))",
+		Key:    key,
+		ICVLen: 128,
+	}
+
+	return state, nil
+}
+
+func xfrmPolicy(srcIP, dstIP net.IP, spi SPI) *netlink.XfrmPolicy {
+	ipMask := []byte{0xff, 0xff, 0xff, 0xff} // /32
+
+	return &netlink.XfrmPolicy{
+		Src:   &net.IPNet{IP: srcIP, Mask: ipMask},
+		Dst:   &net.IPNet{IP: dstIP, Mask: ipMask},
+		Proto: syscall.IPPROTO_UDP,
+		Dir:   netlink.XFRM_DIR_OUT,
+		Mark: &netlink.XfrmMark{
+			Value: mark,
+			Mask:  mark,
+		},
+		Tmpls: []netlink.XfrmPolicyTmpl{
+			{
+				Src:   srcIP,
+				Dst:   dstIP,
+				Proto: netlink.XFRM_PROTO_ESP,
+				Mode:  netlink.XFRM_MODE_TRANSPORT,
+				Spi:   int(spi),
+			},
+		},
+	}
+}
+
+// Key derivation
+
+func genNonce() ([]byte, error) {
+	buf := make([]byte, nonceSize)
+	n, err := rand.Read(buf)
+	if err != nil {
+		return nil, fmt.Errorf("crypto rand failed: %s", err)
+	}
+	if n != nonceSize {
+		return nil, fmt.Errorf("not enough of random data: %d", n)
+	}
+	return buf, nil
+}
+
+func deriveKey(sessionKey []byte, nonce []byte, peerName mesh.PeerName) ([]byte, error) {
+	key := make([]byte, keySize)
+
+	info := make([]byte, 8)
+	binary.BigEndian.PutUint64(info, uint64(peerName))
+
+	hkdf := hkdf.New(sha256.New, sessionKey, nonce, info)
+
+	n, err := io.ReadFull(hkdf, key)
+	if err != nil {
+		return nil, err
+	}
+	if n != keySize {
+		return nil, fmt.Errorf("derived too short key: %d", n)
+	}
+
+	return key, nil
+}
+
+// Protocol Messages
+
+type msgInitSARemote struct {
+	nonce []byte
+	spi   SPI
+}
+
+func deserializeMsgInitSARemote(b []byte) (*msgInitSARemote, error) {
+	if len(b) == 0 {
+		return nil, fmt.Errorf("empty msg")
+	}
+
+	msg := &msgInitSARemote{}
+	if len(b) != msg.size() {
+		return nil, fmt.Errorf("invalid payload size: %d", len(b))
+	}
+
+	msg.nonce = make([]byte, nonceSize)
+	copy(msg.nonce, b[:nonceSize])
+	b = b[nonceSize:]
+
+	msg.spi = SPI(binary.BigEndian.Uint32(b))
+
+	return msg, nil
+}
+
+func (msg *msgInitSARemote) size() int {
+	return nonceSize + 32 // SPI
+}
+
+func (msg *msgInitSARemote) serialize() []byte {
+	b := make([]byte, msg.size())
+
+	copy(b[:nonceSize], msg.nonce)
+	binary.BigEndian.PutUint32(b[nonceSize:], uint32(msg.spi))
+
+	return b
+}
@@ -0,0 +1,18 @@
+package net
+
+import (
+	"fmt"
+	"net"
+	"os"
+)
+
+func ListenUnixSocket(pathname string) (net.Listener, error) {
+	if err := os.Remove(pathname); err != nil && !os.IsNotExist(err) {
+		return nil, fmt.Errorf("Error removing existing unix socket: %s", err)
+	}
+	listener, err := net.Listen("unix", pathname)
+	if err != nil {
+		return nil, fmt.Errorf("ListenUnixSocket failed: %s", err)
+	}
+	return listener, nil
+}
@@ -0,0 +1,38 @@
+package net
+
+import (
+	"crypto/rand"
+	"crypto/sha256"
+	"net"
+)
+
+func RandomMAC() (net.HardwareAddr, error) {
+	mac := make([]byte, 6)
+	if _, err := rand.Read(mac); err != nil {
+		return nil, err
+	}
+
+	setUnicastAndLocal(mac)
+
+	return net.HardwareAddr(mac), nil
+}
+
+func MACfromUUID(uuid []byte) net.HardwareAddr {
+	hash := sha256.New()
+	// We salt the input just as a precaution to avoid clashes with
+	// other applications who might have had the bright idea of
+	// generating MACs in the same way.
+	hash.Write([]byte("9oBJ0Jmip-"))
+	hash.Write(uuid)
+	sum := hash.Sum(nil)
+
+	setUnicastAndLocal(sum)
+
+	return net.HardwareAddr(sum[:6])
+}
+
+// In the first byte of the MAC, the 'multicast' bit should be
+// clear and 'locally administered' bit should be set.
+func setUnicastAndLocal(mac []byte) {
+	mac[0] = (mac[0] & 0xFE) | 0x02
+}
@@ -0,0 +1,116 @@
+package net
+
+import (
+	"encoding/json"
+	"fmt"
+	"net"
+	"os"
+	"strconv"
+
+	"github.com/vishvananda/netlink"
+	"github.com/vishvananda/netns"
+)
+
+type Dev struct {
+	Name  string           `json:"Name,omitempty"`
+	MAC   net.HardwareAddr `json:"MAC,omitempty"`
+	CIDRs []*net.IPNet     `json:"CIDRs,omitempty"`
+}
+
+func LinkToNetDev(link netlink.Link) (Dev, error) {
+	addrs, err := netlink.AddrList(link, netlink.FAMILY_V4)
+	if err != nil {
+		return Dev{}, err
+	}
+
+	netDev := Dev{Name: link.Attrs().Name, MAC: link.Attrs().HardwareAddr}
+	for _, addr := range addrs {
+		netDev.CIDRs = append(netDev.CIDRs, addr.IPNet)
+	}
+	return netDev, nil
+}
+
+// ConnectedToBridgeVethPeerIds returns peer indexes of veth links connected to
+// the given bridge. The peer index is used to query from a container netns
+// whether the container is connected to the bridge.
+func ConnectedToBridgeVethPeerIds(bridgeName string) ([]int, error) {
+	var ids []int
+
+	br, err := netlink.LinkByName(bridgeName)
+	if err != nil {
+		return nil, err
+	}
+	links, err := netlink.LinkList()
+	if err != nil {
+		return nil, err
+	}
+
+	for _, link := range links {
+		if _, isveth := link.(*netlink.Veth); isveth && link.Attrs().MasterIndex == br.Attrs().Index {
+			peerID := link.Attrs().ParentIndex
+			if peerID == 0 {
+				// perhaps running on an older kernel where ParentIndex doesn't work.
+				// as fall-back, assume the peers are consecutive
+				peerID = link.Attrs().Index - 1
+			}
+			ids = append(ids, peerID)
+		}
+	}
+
+	return ids, nil
+}
+
+// Lookup the weave interface of a container
+func GetWeaveNetDevs(processID int) ([]Dev, error) {
+	peerIDs, err := ConnectedToBridgeVethPeerIds("weave")
+	if err != nil {
+		return nil, err
+	}
+
+	return GetNetDevsByVethPeerIds(processID, peerIDs)
+}
+
+func GetNetDevsByVethPeerIds(processID int, peerIDs []int) ([]Dev, error) {
+	// Bail out if this container is running in the root namespace
+	netnsRoot, err := netns.GetFromPid(1)
+	if err != nil {
+		return nil, fmt.Errorf("unable to open root namespace: %s", err)
+	}
+	defer netnsRoot.Close()
+	netnsContainer, err := netns.GetFromPid(processID)
+	if err != nil {
+		// Unable to find a namespace for this process - just return nothing
+		if os.IsNotExist(err) {
+			return nil, nil
+		}
+		return nil, fmt.Errorf("unable to open process %d namespace: %s", processID, err)
+	}
+	defer netnsContainer.Close()
+	if netnsRoot.Equal(netnsContainer) {
+		return nil, nil
+	}
+
+	var netdevs []Dev
+	peersStr := make([]string, len(peerIDs))
+	for i, id := range peerIDs {
+		peersStr[i] = strconv.Itoa(id)
+	}
+	nds, err := WithNetNSByPid(processID, "list-netdevs", peersStr...)
+	if err != nil {
+		return nil, fmt.Errorf("list-netdevs failed: %s", err)
+	}
+	err = json.Unmarshal(nds, &netdevs)
+
+	return netdevs, err
+}
+
+// Get the weave bridge interface.
+// NB: Should be called from the root network namespace.
+func GetBridgeNetDev(bridgeName string) (Dev, error) {
+	link, err := netlink.LinkByName(bridgeName)
+	if err != nil {
+		return Dev{}, err
+	}
+
+	return LinkToNetDev(link)
+}
@@ -0,0 +1,89 @@
+package net
+
+import (
+	"bytes"
+	"errors"
+	"fmt"
+	"os/exec"
+	"path/filepath"
+	"runtime"
+
+	"github.com/vishvananda/netlink"
+	"github.com/vishvananda/netns"
+)
+
+var ErrLinkNotFound = errors.New("Link not found")
+
+// NB: The following function is unsafe, because:
+//     - It changes a network namespace (netns) of an OS thread which runs
+//       the function. During execution, the Go runtime might clone a new OS thread
+//       for scheduling other go-routines, thus they might end up running in
+//       a "wrong" netns.
+//     - runtime.LockOSThread does not guarantee that a spawned go-routine on
+//       the locked thread will be run by it. Thus, the work function is
+//       not allowed to spawn any go-routine which is dependent on the given netns.
+
+//     Please see https://github.com/weaveworks/weave/issues/2388#issuecomment-228365069
+//     for more details and make sure that you understand the implications before
+//     using the function!
+func WithNetNSUnsafe(ns netns.NsHandle, work func() error) error {
+	runtime.LockOSThread()
+	defer runtime.UnlockOSThread()
+
+	oldNs, err := netns.Get()
+	if err == nil {
+		defer oldNs.Close()
+
+		err = netns.Set(ns)
+		if err == nil {
+			defer netns.Set(oldNs)
+
+			err = work()
+		}
+	}
+
+	return err
+}
+
+func WithNetNSLinkUnsafe(ns netns.NsHandle, ifName string, work func(link netlink.Link) error) error {
+	return WithNetNSUnsafe(ns, func() error {
+		link, err := netlink.LinkByName(ifName)
+		if err != nil {
+			if err.Error() == errors.New("Link not found").Error() {
+				return ErrLinkNotFound
+			}
+			return err
+		}
+		return work(link)
+	})
+}
+
+var WeaveUtilCmd = "weaveutil"
+
+// A safe version of WithNetNS* which creates a process executing
+// "nsenter --net=<ns-path> weaveutil <cmd> [args]".
+func WithNetNS(nsPath string, cmd string, args ...string) ([]byte, error) {
+	var stdout, stderr bytes.Buffer
+
+	args = append([]string{"--net=" + nsPath, WeaveUtilCmd, cmd}, args...)
+	c := exec.Command("nsenter", args...)
+	c.Stdout = &stdout
+	c.Stderr = &stderr
+	if err := c.Run(); err != nil {
+		return nil, fmt.Errorf("%s: %s", string(stderr.Bytes()), err)
+	}
+
+	return stdout.Bytes(), nil
+}
+
+func WithNetNSByPid(pid int, cmd string, args ...string) ([]byte, error) {
+	return WithNetNS(NSPathByPid(pid), cmd, args...)
+}
+
+func NSPathByPid(pid int) string {
+	return NSPathByPidWithRoot("/", pid)
+}
+
+func NSPathByPidWithRoot(root string, pid int) string {
+	return filepath.Join(root, fmt.Sprintf("/proc/%d/ns/net", pid))
+}
@@ -3,18 +3,17 @@ package net
 import (
 	"fmt"
 	"net"
-	"syscall"
+	"os"

 	"github.com/vishvananda/netlink"
-	"github.com/vishvananda/netlink/nl"
 )

 // A network is considered free if it does not overlap any existing
 // routes on this host. This is the same approach taken by Docker.
 func CheckNetworkFree(subnet *net.IPNet, ignoreIfaceNames map[string]struct{}) error {
-	return forEachRoute(ignoreIfaceNames, func(name string, route netlink.Route) error {
+	return forEachRoute(ignoreIfaceNames, func(route netlink.Route) error {
 		if route.Dst != nil && overlaps(route.Dst, subnet) {
-			return fmt.Errorf("Network %s overlaps with existing route %s on host.", subnet, route.Dst)
+			return fmt.Errorf("Network %s overlaps with existing route %s on host", subnet, route.Dst)
 		}
 		return nil
 	})
@@ -28,88 +27,45 @@ func overlaps(n1, n2 *net.IPNet) bool {
 // For a specific address, we only care if it is actually *inside* an
 // existing route, because weave-local traffic never hits IP routing.
 func CheckAddressOverlap(addr net.IP, ignoreIfaceNames map[string]struct{}) error {
-	return forEachRoute(ignoreIfaceNames, func(name string, route netlink.Route) error {
+	return forEachRoute(ignoreIfaceNames, func(route netlink.Route) error {
 		if route.Dst != nil && route.Dst.Contains(addr) {
-			return fmt.Errorf("Address %s overlaps with existing route %s on host.", addr, route.Dst)
+			return fmt.Errorf("Address %s overlaps with existing route %s on host", addr, route.Dst)
 		}
 		return nil
 	})
 }

-func forEachRoute(ignoreIfaceNames map[string]struct{}, check func(name string, r netlink.Route) error) error {
+func forEachRoute(ignoreIfaceNames map[string]struct{}, check func(r netlink.Route) error) error {
+	ignoreIfaceIndices := make(map[int]struct{})
+	for ifaceName := range ignoreIfaceNames {
+		if iface, err := net.InterfaceByName(ifaceName); err == nil {
+			ignoreIfaceIndices[iface.Index] = struct{}{}
+		}
+	}
 	routes, err := netlink.RouteList(nil, netlink.FAMILY_V4)
 	if err != nil {
 		return err
 	}
 	for _, route := range routes {
-		link, err := netlink.LinkByIndex(route.LinkIndex)
-		if err == nil {
-			if _, found := ignoreIfaceNames[link.Attrs().Name]; found {
-				continue
-			}
+		if _, found := ignoreIfaceIndices[route.LinkIndex]; found {
+			continue
 		}
-		if err := check(link.Attrs().Name, route); err != nil {
+		if err := check(route); err != nil {
 			return err
 		}
 	}
 	return nil
 }

-func CheckRouteExists(ifaceName string, dest net.IP) bool {
-	found := false
-	forEachRoute(map[string]struct{}{}, func(name string, route netlink.Route) error {
-		if name == ifaceName && route.Dst.IP.Equal(dest) {
-			found = true
-		}
-		return nil
+func AddRoute(link netlink.Link, scope netlink.Scope, dst *net.IPNet, gw net.IP) error {
+	err := netlink.RouteAdd(&netlink.Route{
+		LinkIndex: link.Attrs().Index,
+		Scope:     scope,
+		Dst:       dst,
+		Gw:        gw,
 	})
-	return found
-}
-
-func matchRoute(ifaceName string, dest net.IP) func(m syscall.NetlinkMessage) (bool, error) {
-	return func(m syscall.NetlinkMessage) (bool, error) {
-		switch m.Header.Type {
-		case syscall.RTM_NEWROUTE:
-			route, err := deserializeRoute(m.Data)
-			if err != nil {
-				return true, err
-			}
-			link, _ := netlink.LinkByIndex(route.LinkIndex)
-			if link.Attrs().Name == ifaceName && route.Dst.IP.Equal(dest) {
-				return true, nil
-			}
-		}
-		return false, nil
+	if os.IsExist(err) { // squash duplicate route errors
+		err = nil
 	}
-}
-
-// This code proposed for addition to the vishvananda/netlink library
-// deserializeRoute decodes a binary netlink message into a Route struct
-func deserializeRoute(m []byte) (netlink.Route, error) {
-	route := netlink.Route{}
-	msg := nl.DeserializeRtMsg(m)
-	attrs, err := nl.ParseRouteAttr(m[msg.Len():])
-	if err != nil {
-		return route, err
-	}
-	route.Scope = netlink.Scope(msg.Scope)
-
-	native := nl.NativeEndian()
-	for _, attr := range attrs {
-		switch attr.Attr.Type {
-		case syscall.RTA_GATEWAY:
-			route.Gw = net.IP(attr.Value)
-		case syscall.RTA_PREFSRC:
-			route.Src = net.IP(attr.Value)
-		case syscall.RTA_DST:
-			route.Dst = &net.IPNet{
-				IP:   attr.Value,
-				Mask: net.CIDRMask(int(msg.Dst_len), 8*len(attr.Value)),
-			}
-		case syscall.RTA_OIF:
-			routeIndex := int(native.Uint32(attr.Value[0:4]))
-			route.LinkIndex = routeIndex
-		}
-	}
-	return route, nil
+	return err
 }
@@ -0,0 +1,83 @@
+package net
+
+import (
+	"errors"
+	"io/ioutil"
+	"net"
+	"os"
+
+	"github.com/weaveworks/mesh"
+	"github.com/weaveworks/weave/db"
+)
+
+func getOldStyleSystemUUID() ([]byte, error) {
+	uuid, err := ioutil.ReadFile("/sys/class/dmi/id/product_uuid")
+	if os.IsNotExist(err) {
+		uuid, err = ioutil.ReadFile("/sys/hypervisor/uuid")
+	}
+	return uuid, err
+}
+
+func getSystemUUID(hostRoot string) ([]byte, error) {
+	uuid, err := getOldStyleSystemUUID()
+	if err != nil && !os.IsNotExist(err) {
+		return nil, err
+	}
+	machineid, err := ioutil.ReadFile(hostRoot + "/etc/machine-id")
+	if os.IsNotExist(err) {
+		machineid, err = ioutil.ReadFile(hostRoot + "/var/lib/dbus/machine-id")
+	}
+	if err != nil && !os.IsNotExist(err) {
+		return nil, err
+	}
+	if len(uuid) == 0 && len(machineid) == 0 {
+		return nil, errors.New("All system IDs are blank")
+	}
+	return append(machineid, uuid...), nil
+}
+
+func getPersistedPeerName(dbPrefix string) (mesh.PeerName, error) {
+	d, err := db.NewBoltDBReadOnly(dbPrefix)
+	if err != nil {
+		return mesh.UnknownPeerName, err
+	}
+	defer d.Close()
+	var peerName mesh.PeerName
+	nameFound, err := d.Load(db.NameIdent, &peerName)
+	if err != nil || !nameFound {
+		return mesh.UnknownPeerName, err
+	}
+	return peerName, nil
+}
+
+// GetSystemPeerName returns an ID derived from concatenated machine-id
+// (either systemd or dbus), the system (aka bios) UUID and the
+// hypervisor UUID.  It is tweaked and formatted to be usable as a mac address
+func GetSystemPeerName(dbPrefix, hostRoot string) (string, error) {
+	// Check if we have a persisted name that matches the old-style ID for this host
+	if oldUUID, err := getOldStyleSystemUUID(); err == nil {
+		if _, err := os.Stat(db.Pathname(dbPrefix)); err == nil {
+			persistedPeerName, err := getPersistedPeerName(dbPrefix)
+			if err != nil && !os.IsNotExist(err) {
+				return "", err
+			}
+			if persistedPeerName == mesh.PeerNameFromBin(MACfromUUID(oldUUID)) {
+				return persistedPeerName.String(), nil
+			}
+		}
+	} else if !os.IsNotExist(err) {
+		return "", err
+	}
+	var mac net.HardwareAddr
+	if uuid, err := getSystemUUID(hostRoot); err == nil {
+		mac = MACfromUUID(uuid)
+	} else if !os.IsNotExist(err) {
+		return "", err
+	} else {
+		mac, err = RandomMAC()
+		if err != nil {
+			return "", err
+		}
+	}
+	return mac.String(), nil
+}
				`@@ -0,0 +1 @@`
				`8f313a9df867c207ae9ceac6e0935d1d2a0360aafce780dd5d1e0687ee98e0c3 docker-1.10.3.tgz`
				`@@ -0,0 +1 @@`
				`7541b5490cee1183d911f909c0e91d3d9d57b2b775e40f14a1a7a6de216ca4a8 docker-1.10.3.tgz`
				`@@ -0,0 +1 @@`
				`c02d861383d035ba020b361ef93bf14beb3964a42a6ae645b505b6148493e7ec qemu-aarch64-static`
				`@@ -0,0 +1 @@`
				`0ff63682b8868620d771aa812d3053dc3c579dc8633ce62d3d7ae7eee83beb06 qemu-arm-static`