The PEERING client controller is a set of scripts to ease configuration and operation of a PEERING client, able to connect to PEERING routers and announce PEERING prefixes.
After cloning this repository, follow the instructions below to install software dependencies and set up your PEERING client.
The client runs OpenVPN to connect your machine directly to PEERING routers. We then run the BIRD software router to establish BGP sessions over the OpenVPN tunnels. BIRD is used to control prefix announcements. You will also need the socat tool that scripts use to interact with the OpenVPN socket. On Debian, you'll need apt install bird openvpn socat psmisc ipcalc.
After installing OpenVPN and BIRD, you may want to disable these services on your machine, which some distributions may enable by default. On Debian, run systemctl disable bird bird6 openvpn to accomplish this.
The provided BIRD configurations are compatible with BIRD 1.6 (but not 2+). BIRD is available for most distributions, but you can compile Bird from source.
To establish OpenVPN tunnels with PEERING routers, you will need PEERING-issued certificates. You can get certificates by submitting a project proposal on our website. Copy your certificate files into certs/ and rename them as client.crt, and client.key. Then chmod 400 all files in certs/ to prevent unauthorized access to your keys.
You will also need to explicitly create a prefixes.txt file containing the prefixes you are going to announce. This is an extra safety net. Specify one prefix per line, in the usual format, e.g., 184.164.236.0/24.
Run ./peering openvpn to get a description of command-line parameters for interacting with OpenVPN. The status command shows the status of OpenVPN tunnels. Tunnels can be either up or down. If a tunnel is up, we will also list the device (the interface) it is running on and the local IP address. You can use ip route to identify the IP address of the remote end as the gateway associated with each tunnel.
You can pass the special value all to operate on all PEERING routers simultaneously.
Creating BGP sessions with many routers simultaneously will significantly increase the amount of memory needed by BIRD.
Run ./peering bgp or ./peering bgp6 to get a description of command-line parameters for interacting with BIRD. You need to start the v4 and v6 versions separately. When you start BIRD, it will continuously attempt to establish BGP sessions with all PEERING routers (and will succeed to establish a session if the OpenVPN tunnel is up). Starting or stopping BIRD will establish and close all BGP sessions automatically.
When checking the status of BGP connections, note that sessions in the "Idle" state are waiting for their respective OpenVPN tunnels to come up. This is not an issue. Sessions in the Established state are up and exchanging routes.
Warning: The PEERING client does not import any routes by default. You should edit the import filters (
import none;in theconfigs/bird*/bird*.conffiles) to import routes. However, consider that if you connect to many muxes, the routing tables will take up a lot of RAM. Suggestions to keep memory utilization in check include (1) establishing OpenVPN tunnels only with the muxes you plan to use or selectively choose which muxes to import routes from.
Run ./peering prefix to get a description of command-line parameters available to control prefix announcements. The scripts supports prepending, changing the origin AS, poisoning an AS, and attaching communities. Note that BGP poisoning and communities require special capabilities that must be assigned to your account by PEERING staff before you can use them.
Once your client is configured you should be able to run experiments. Here are some steps to get you started:
-
cdinto the client directory -
Use the
./peeringscript to establish an OpenVPN connection to a PEERING router, e.g.,./peering openvpn up <mux>. -
Use the
./peeringscript to establish a BGP session for exchanging routes with the router, e.g.,./peering bgp start.Warning: When using IPv6, you need to edit the
client/bird6/bird6.conffile and set a valid router ID (search for the line starting withrouter id) and use a unique IP address (e.g., one allocated to your experiment)Warning: The BIRD configuration has BGP sessions configured with all PEERING muxes. You can control which sessions get established by bringing the OpenVPN tunnels up or down. However, the BIRD configuration also imports all routes received from the muxes. If you establish many sessions and import all routes, RAM utilization will be high. Start slow and provision accordingly.
Read the output of ./peering prefix to find out how to make and control announcements. For example, to announce your prefix out of all PEERING routers you are connected to, use ./peering prefix announce <prefix>.
You can check that your prefix is propagating by using Looking Glass servers from multiple providers:
Troubleshooting: If the prefix does not seem to be propagating, check that the OpenVPN tunnel is up, that the BGP session is established, and that your client is exporting the prefix to the desired router:
./peering openvpn status ./peering bgp status ./peering bgp adv <router>
To shut everything up, stop the BGP server and OpenVPN tunnels:
./peering bgp stop
./peering openvpn down allFollow these guidelines when using your PEERING client:
-
Do not announce prefixes that are not allocated to your experiment. Do not announce prefixes outside PEERING address space. (The PEERING prefix control script will print a list of valid PEERING prefixes if you input an incorrect one.)
-
Similarly, do not spoof packets with source IP addresses outside the PEERING address space allocated to your experiment.
-
Do not change announcements more than once every 10 minutes. For best results, prefer to leave announcements up for 90 minutes to avoid route-flap dampening.
-
Be conservative. Routers are often running close to their limits and our first priority is to not disrupt Internet operation. In particular, do not announce AS-paths with more than 5 AS-hops, do not announce paths containing AS-sets with more than 5 ASes, and do not announce paths with more than 5 attached communities.
The app module configures a network namespace with a single
interface, and routes the network namespace through PEERING OpenVPN
tunnels. Run ./peering app for a list of available parameters.
Each application operates on a PEERING prefix (either v4 or v6). We support isolating applications in their own network namespace or with a Linux virtual bridge. The virtual bridge approach is useful to run Docker containers attached to the bridge.
By default, the namespace or bridge is called pappX, where X is an ID to identify the namespace. Users that need multiple applications will need to set a different ID to each application. By default, the application's egress traffic will be routed using table 20000, which is populated by BIRD. An option allows the user to choose a specific upstream to route egress traffic out of (-u). When deleting an application (-d), pass all the other parameters identically to when the application was created.
In case sending traffic out of the namespace does not work, here are a list of things to check:
- Check prefix propagation on Looking Glasses. AT&T and HE have telnet-accessible LGs:
route-server.ip.att.netandroute-views.he.net - Check that PEERING OpenVPN tunnels and BGP sessions are up; announce a prefix and check reachability from the host.
- Check that IP forwarding is enabled (e.g., run
sysctl -w net.ipv4.ip_forward=1). - Check that the
FORWARDchain iniptablesis set toACCEPT, and change it if needed (iptables -P FORWARD ACCEPT). - Check that the DNS resolver replies to requests from PEERING space.
The "unreachable" printed by BIRD when printing exported routes is unrelated to prefix propagation. It just means that BIRD itself doesn't know the route to your prefix, which is fine as the destination is the host itself and no further routing is necessary.
PEERING offers a way to make announcements via REST API, without the need for a VPN connection and prior approval of an experiment on the platform. These announcements are scheduled and made according to the availability of prefixes. Each sequence of announcements is considered an experiment. Each announcement lasts 90 minutes (to allow for BGP convergence and avoid route-flap dampening).
To deploy an experiment we can use the peering.py script in this directory. Just pass the experiment as parameter and the URL of the page. An example of an experiment can be seen at utils/experiment-examples/experiment-1.json. An experiment can be generated by using the following python code utils/experiment-generator.py.
Access to the API is controlled by a permission token that becomes available to you in the user dashboard on the PEERING site. Note that the API is very limited; if you need fine-grained control over the announcements, consider submitting a proposal on the Website to receive full access and use the functionality above. The token must be in the certificate directory certs/token.json like shown below. The refresh token is currently not required. More information on the wiki page
{
"access": "<token>",
"refresh": ""
}We use TinyProxy to provide a proxy for HTTP access from inside your container on PEERING routers. If your experiment does not run containers on PEERING routers, then you have no need to use TinyProxy.
Run ./peering proxy to get a description of command-line parameters available to interact with TinyProxy. The proxy for a given router's container needs to start after the OpenVPN tunnel is established, or TinyProxy will be unable to bind to the right IP address.
This script reads your container's allocated prefix ID from a file named container.txt. This information is necessary to compute prefixes and install routes, you can find your container's ID on the PEERING website dashboard. Generate the file with echo ID > container.txt.
Bringing up a proxy prints relevant information to access and interact with that router's container.
TinyProxy addresses for isi01 (tap2, 2)
local address: 100.66.128.6:8802
subnet: 100.125.16.8/30
pidfile: /home/cunha/git/peering/client/var/tinyproxy.isi01.pid
logfile: /home/cunha/git/peering/client/var/tinyproxy.isi01.log
updating (add) 100.125.16.8/30 via 100.66.128.1 dev tap2
updating (add) 2804:269c:ff03:2:2::/80 via 2804:269c:ff00:2:1::1 dev tap2
You can SSH into your container by using the second host in the /30
subnet. In other words, SSH to the third address in the /30. You should log in as root using your private key:
ssh -i ~/.ssh/peering_id_rsa [email protected]
You can use the proxy on the container by setting the http_proxy
environment variable to match TinyProxy's local address (above).
export http_proxy=http://100.66.128.6:8802/
apt update
apt install lighttpd
Containers have limited RAM and disk space. The amount of RAM available on containers is insufficient to run a PEERING client on IXP sites (e.g., amsterdam01 and seattle01). We recommend users run the PEERING client remotely (e.g., on the cloud or at a server in their institution), and route traffic into the container by rewriting the BGP next-hop field (see the -M parameter to ./peering prefix).
The control scripts allow you to quickly start using PEERING. They do not cover all possible uses of PEERING. If you need to perform more complex announcements (e.g., make BGP announcements with BGP communities attached), these scripts provide a useful starting point.
More information about PEERING configuration:
- Client data plane.
- Mux data plane.
- Additional information can be found in the Wiki.
The peering.py module can also be imported into Python programs to programmatically control announcements. Announcements are specified in JSON, following the JSON schema in configs/announcement_schema.json. You should edit the allocatedPrefixes entry in the JSON schema to the prefixes allocated to your experiment.