GoRTR

DEPRECATION NOTICE

This software is no longer maintained. We advise replacing your production use of this software with the swap-in replacement StayRTR

GitHub release (latest by date)

GoRTR is an open-source implementation of RPKI to Router protocol (RFC 6810) using the the Go Programming Language.

/lib contains a library to create your own server and client.
/prefixfile contains the structure of a JSON export file and signing capabilities.
/cmd/gortr/gortr.go is a simple implementation that fetches a list and offers it to a router.
/cmd/rtrdump/rtrdump.go allows copying the PDUs sent by a RTR server as a JSON file.
/cmd/rtrmon/rtrmon.go compare and monitor two RTR servers (using RTR and/or JSON), outputs diff and Prometheus metrics.

Disclaimer

This software comes with no warranty.

In the field

Cloudflare operates 200+ GoRTR globally. They provide redundancy in at the PoP level. This provides increased reliability by computing a unique prefix list and providing a secure distribution of the file over its CDN before being sent to the routers.

GoRTR also powers the public RTR server available on rtr.rpki.cloudflare.com on port 8282 and 8283 for SSH (rpki/rpki)

Telia has deployed RPKI and uses GoRTR connected with OctoRPKI and rpki-client to distribute the ROAs to its routers. Instances of the RTR servers handle around 250 sessions each.

NTT has deployed OpenBSD's rpki-client together with GoRTR to facilitate rejecting RPKI Invalid BGP route announcements towards it's Global IP Network (AS 2914). More information is available here.

GTT deployed GoRTR along with OctoRPKI. The setup currently provides 400+ RTR sessions to their routers for filtering RPKI invalids.

Cogent deployed GoRTR and OctoRPKI at the end of May 2020. 8 validators feed approximately 2500 routers.

Router vendors also used this software to develop their implementations.

Do you use this tool at scale? Let us know!

Features of the server

Refreshes a JSON list of prefixes (from either Cloudflare or a RIPE Validator)
Prometheus metrics
Lightweight
TLS
SSH
Signature verification and expiration control

Features of the extractor

Generate a list of prefixes sent via RTR (similar to Cloudflare JSON input, or RIPE RPKI Validator)
Lightweight
TLS
SSH

Features of the API

Protocol v0 of RFC6810
Protocol v1 of RFC8210
Event-driven API
TLS
SSH

To start developing

You need a working Go environment (1.10 or newer). This project also uses Go Modules.

$ git clone git@github.com:cloudflare/gortr.git && cd gortr
$ go build cmd/gortr/gortr.go

With Docker

If you do not want to use Docker, please go to the next section.

If you have Docker, you can start GoRTR with docker run -ti -p 8082:8082 cloudflare/gortr. The containers contain Cloudflare's public signing key and an testing ECDSA private key for the SSH server.

It will automatically download Cloudflare's prefix list and use the public key to validate it.

You can now use any CLI attributes as long as they are after the image name:

$ docker run -ti -p 8083:8083 cloudflare/gortr -bind :8083

If you want to build your own image of GoRTR:

$ docker build -t mygortr -f Dockerfile.gortr.prod .
$ docker run -ti mygortr -h

It will download the code from GitHub and compile it with Go and also generate an ECDSA key for SSH.

Please note: if you plan to use SSH with Cloudflare's default container (cloudflare/gortr), replace the key private.pem since it is a testing key that has been published. An example is given below:

$ docker run -ti -v $PWD/mynewkey.pem:/private.pem cloudflare/gortr -ssh.bind :8083

Install it

There are a few solutions to install it.

Go can directly fetch it from the source

$ go get github.com/cloudflare/gortr/cmd/gortr

Copy cf.pub to your local directory if you want to use Cloudflare's signed JSON file.

You can use the Makefile (by default it will be compiled for Linux, add GOOS=darwin for Mac)

$ make dist-key build-gortr

The compiled file will be in /dist.

Or you can use a package (or binary) file from the Releases page:

$ sudo dpkg -i gortr[...].deb
$ sudo systemctl start gortr

If you want to sign your list of prefixes, generate an ECDSA key. Then generate the public key to be used in GoRTR. You will have to setup your validator to use this key or have another tool to sign the JSON file before passing it to GoRTR.

$ openssl ecparam -genkey -name prime256v1 -noout -outform pem > private.pem
$ openssl ec -in private.pem -pubout -outform pem > public.pem

Run it

Once you have a binary:

$ ./gortr -tls.bind 127.0.0.1:8282

Make sure cf.pub is in the current directory. Or pass -verify.key=path/to/cf.pub

Package it

If you want to package it (deb/rpm), you can use the pre-built docker-compose file.

$ docker-compose -f docker-compose-pkg.yml up

You can find both files in the dist/ directory.

Usage with a proxy

This was tested with a basic Squid proxy. The User-Agent header is passed in the CONNECT.

You have to export the following two variables in order for GoRTR to use the proxy.

export HTTP_PROXY=schema://host:port
export HTTPS_PROXY=schema://host:port

With SSL

You can run GoRTR and listen for TLS connections only (just pass -bind "").

First, you will have to create a SSL certificate.

$ openssl ecparam -genkey -name prime256v1 -noout -outform pem > private.pem
$ openssl req -new -x509 -key private.pem -out server.pem

Then, you have to run

$ ./gortr -ssh.bind :8282 -tls.key private.pem -tls.cert server.pem

With SSH

You can run GoRTR and listen for SSH connections only (just pass -bind "").

You will have to create an ECDSA key. You can use the following command:

$ openssl ecparam -genkey -name prime256v1 -noout -outform pem > private.pem

Then you can start:

$ ./gortr -ssh.bind :8282 -ssh.key private.pem -bind ""

By default, there is no authentication.

You can use password and key authentication:

For example, to configure user rpki and password rpki:

$ ./gortr -ssh.bind :8282 -ssh.key private.pem -ssh.method.password=true -ssh.auth.user rpki -ssh.auth.password rpki -bind ""

And to configure a bypass for every SSH key:

$ ./gortr -ssh.bind :8282 -ssh.key private.pem -ssh.method.key=true -ssh.auth.key.bypass=true -bind ""

Configure filters and overrides (SLURM)

GoRTR supports SLURM configuration files (RFC8416).

Create a json file (slurm.json):

{
    "slurmVersion": 1,
    "validationOutputFilters": {
     "prefixFilters": [
       {
        "prefix": "10.0.0.0/8",
        "comment": "Everything inside will be removed"
       },
       {
        "asn": 65001,
       },
       {
        "asn": 65002,
        "prefix": "192.168.0.0/24",
       },
     ],
     "bgpsecFilters": []
    },
    "locallyAddedAssertions": {
     "prefixAssertions": [
       {
        "asn": 65001,
        "prefix": "2001:db8::/32",
        "maxPrefixLength": 48,
        "comment": "Manual add"
       }
     ],
     "bgpsecAssertions": [
     ]
    }
  }

When starting GoRTR, add the -slurm ./slurm.json argument.

The log should display something similar to the following:

INFO[0001] Slurm filtering: 112214 kept, 159 removed, 1 asserted
INFO[0002] New update (112215 uniques, 112215 total prefixes).

For instance, if the original JSON fetched contains the ROA: 10.0.0.0/24-24 AS65001, it will be removed.

The JSON exported by GoRTR will contain the overrides and the file can be signed again. Others GoRTR can be configured to fetch the ROAs from the filtering GoRTR: the operator manages one SLURM file on a leader GoRTR.

Debug the content

You can check the content provided over RTR with rtrdump tool

$ ./rtrdump -connect 127.0.0.1:8282 -file debug.json

You can also fetch the re-generated JSON from the -export.path endpoint (default: http://localhost:8080/rpki.json)

Data sources

Use your own validator, as long as the JSON source follows the following schema:

{
  "roas": [
    {
      "prefix": "10.0.0.0/24",
      "maxLength": 24,
      "asn": "AS65001"
    },
    ...
  ]
}

Cloudflare (list curated, signed, compressed and cached in +160 PoPs)
Third-party JSON formatted VRP exports:
- NTT (based on OpenBSD's rpki-client)
- RIPE (based on RIPE NCC's RPKI Cache Validator)

To use a data source that do not contain signatures or validity information, pass: -verify=false -checktime=false

Note: for boolean flags, it requires the equal sign

Cloudflare's prefix list removes duplicates and entries that are not routed on the Internet (>/24 IPv4 and >/48 IPv6).

By default, the session ID will be randomly generated. The serial will start at zero.

You can define a serial to start with the following way:

the JSON must contain a serial field in metadata; and
the flag -useserial must be set to 1 or 2

When flag is set to 1, every change of file will increment the serial regardless of the current serial field. Make sure the refresh rate of GoRTR is more frequent than the refresh rate of the JSON.

When flag is set to 2, GoRTR will set the value of the serial in the JSON. If an ID is missed or not updated, it will cause discrepancies on the client.

Configurations

Compatibility matrix

A simple comparison between software and devices. Implementations on versions may vary.

Device/software	Plaintext	TLS	SSH	Notes
RTRdump	Yes	Yes	Yes
RTRlib	Yes	No	Yes	Only SSH key
Juniper	Yes	No	No
Cisco	Yes	No	Yes	Only SSH password
Nokia SR OS	Yes	No	No	Since 9.0.R1(2010)
Arista	Yes	No	No
FRRouting	Yes	No	Yes	Only SSH key
Bird2	Yes	No	Yes	Only SSH key
Quagga	Yes	No	No

Configure on Juniper

Configure a session to the RTR server (assuming it runs on 192.168.1.100:8282)

louis@router> show configuration routing-options validation
group TEST-RPKI {
    session 192.168.1.100 {
        port 8282;
    }
}

Add policies to validate or invalidate prefixes

louis@router> show configuration policy-options policy-statement STATEMENT-EXAMPLE
term RPKI-TEST-VAL {
    from {
        protocol bgp;
        validation-database valid;
    }
    then {
        validation-state valid;
        next term;
    }
}
term RPKI-TEST-INV {
    from {
        protocol bgp;
        validation-database invalid;
    }
    then {
        validation-state invalid;
        reject;
    }
}

Display status of the session to the RTR server.

louis@router> show validation session 192.168.1.100 detail
Session 192.168.1.100, State: up, Session index: 1
  Group: TEST-RPKI, Preference: 100
  Port: 8282
  Refresh time: 300s
  Hold time: 600s
  Record Life time: 3600s
  Serial (Full Update): 1
  Serial (Incremental Update): 1
    Session flaps: 2
    Session uptime: 00:25:07
    Last PDU received: 00:04:50
    IPv4 prefix count: 46478
    IPv6 prefix count: 8216

Show content of the database (list the PDUs)

louis@router> show validation database brief
RV database for instance master

Prefix                 Origin-AS Session                                 State   Mismatch
1.0.0.0/24-24              13335 192.168.1.100                           valid
1.1.1.0/24-24              13335 192.168.1.100                           valid

Configure on Cisco

You may want to use the option to do SSH-based connection.

On Cisco, you can have only one RTR server per IP.

To configure a session for 192.168.1.100:8282: Replace 65001 by the configured ASN:

router bgp 65001
 rpki server 192.168.1.100
  transport tcp port 8282
 !
!

For an SSH session, you will also have to configure router bgp 65001 rpki server 192.168.1.100 password xxx where xxx is the password. Some experimentations showed you have to configure the username/password first, otherwise it will not accept the port.

router bgp 65001
 rpki server 192.168.1.100
  username rpki
  transport ssh port 8282
 !
!
ssh client tcp-window-scale 14
ssh timeout 120

The last two SSH statements solved an issue causing the connection to break before receiving all the PDUs (TCP window full problem).

To visualize the state of the session:

RP/0/RP0/CPU0:ios#sh bgp rpki server 192.168.1.100

RPKI Cache-Server 192.168.1.100
  Transport: SSH port 8282
  Connect state: ESTAB
  Conn attempts: 1
  Total byte RX: 1726892
  Total byte TX: 452
  Last reset
    Timest: Apr 05 01:19:32 (04:26:58 ago)
    Reason: protocol error
SSH information
  Username: rpki
  Password: *****
  SSH PID: 18576
RPKI-RTR protocol information
  Serial number: 15
  Cache nonce: 0x0
  Protocol state: DATA_END
  Refresh  time: 600 seconds
  Response time: 30 seconds
  Purge time: 60 seconds
  Protocol exchange
    ROAs announced:  67358 IPv4   11754 IPv6
    ROAs withdrawn:     80 IPv4      34 IPv6
    Error Reports :      0 sent       0 rcvd
  Last protocol error
    Reason: response timeout
    Detail: response timeout while in DATA_START state

To visualize the accepted PDUs:

RP/0/RP0/CPU0:ios#sh bgp rpki table

  Network               Maxlen          Origin-AS         Server
  1.0.0.0/24            24              13335             192.168.1.100
  1.1.1.0/24            24              13335             192.168.1.100

Configure on Arista

router bgp <asn>
   rpki cache <name>
      host <ipv4|ipv6|hostname> [vrf <vrfname>] [port <1-65535>] # default port is 323
      local-interface <interface>
      preference <1-10>                    # the lower the value, the more preferred
                                           # default is 5
      refresh-interval <1-86400 seconds>   # default is 3600
      expire-interval <600-172800 seconds> # default is 7200
      retry-interval <1-7200 seconds>      # default is 600

If multiple caches are configured, the preference controls the priority.
Caches which are more preferred will be connected to first, if they are not reachable then connections will be attempted to less preferred caches.
If caches have the same preference value, they will all be connected to and the ROAs that are synced from them will be merged together.