Matthew Ernisse

November 07, 2018 @23:30

A little over six and a half years ago I left the Linux as a desktop community for the Mac community. I replaced a Lenovo Thinkpad T500 for an Apple refurbished late 2011 MacBook Pro and honestly have not regretted it.

Over the years I maxed out the memory, went from the 500G SATA HDD to a Crucial 256GB SSD, then put the 500G HDD in the optical bay, then upgraded to a Samsung EVO 512GB SSD with the optical drive back in there. I replaced the keyboard twice, the battery twice, and had the logic board replaced out of warranty for free by Apple thanks to the recall for an issue with the discrete graphics. Through all that it quite happily chugged along and for the most part just worked. Even now it's in better shape than most of my old laptops, but the lid hinge is starting to get weak (it will often just slowly close itself on me), it needs yet another new battery, and the inability to run the latest macOS lead me to conclude that it is time to look for an upgrade.

Old Laptops

It ended up being a bit of a challenge to decide on an upgrade, though. I really like the 13" Retina MacBook Pro I have for work, I really like the portability of the MacBook, and the new MacBook Air looks like a great compromise between the two. I fought with myself for quite some time over what would come next for me and finally settled on a 15" Mid-2015 Retina MacBook Pro. Essentially the bigger brother of what I have for work.

Hello, Kitsune

Now I won't blame you if you are wondering why I'd pick a 3 year old laptop over the latest and greatest. In the end it was almost the only choice. I wanted a 15" class laptop because I spend most of my time computing sitting on the couch. The 13" is really nice and portable but it's actually a little too light and a tad too small to comfortably use on my lap. That basically ruled out the lighter and smaller MacBook and MacBook Air. As for the newer 15" MacBook Pro, I almost exclusively use the vi text editor so not having a hardware escape key is just not something I feel I can get used to. I've also heard many people at work who do have the new MacBook Pros complain loudly about the keyboard so that was another nail in the coffin of the new models.

Given all that, the last non-touchbar 15" MacBook Pro is... the Mid 2015. I found a nice example with the 2.5GHz i7 and the Radeon R9 on eBay for a real good price after a few weeks of looking and snapped it up.

Since this is the second Mac I've ever had as my primary workstation it was the first time I got to use Migration Assistant. I have previously used recovery mode to recover from Time Machine which works a treat so I had high expectations. In the end I'd say the experience lived up to them. The only real problem I had seems to be related to how I have my WiFi configured. I use WPA2 Enterprise (so username and password authentication) on my main SSID which I configure using a profile in macOS (which also serves to disable Siri, a bunch of iCloud stuff I don't use, sets up my internal certificate trust settings, and my VPN). Every time I started up Migration Assistant it would drop off the WiFi with no explanation. After flailing around a bit it looks like that was because it couldn't access the authentication information after logging me out to start the migration, so I figured I'd use Ethernet. That would have worked except that the laptop showed up on a Saturday and the only Thunderbolt to Ethernet adapter I own was at the office. Thankfully my guest WiFi uses WPA2 PSK and that actually appears to work just fine.


It took about 4 hours to transfer the 210GB or so of data, but afterwards the new Mac looked just the same as the old Mac. A quick run through my customization script to catch the few settings in the new version of macOS, the automounter, and the applications I have installed via homebrew, I have not had to go back. Sunday evening I shut off the old laptop. I do plan on re-deploying it as a test workstation if I ever get around to building a dedicated test environment, but for now it is sitting in a drawer under my desk.

It's been a good laptop and this new one has big shoes to fill.

Goodbye, Aramaki


November 06, 2018 @13:40

It's probably too late to change anyone's mind, but I saw a particularly salient Twitter come across this morning.


Also particularly poignant for me is this morning's post over at McMansion Hell

Nub Sez Vote So please, vote. This is basically the bare minimum required of all citizens in this republic other than paying your taxes (maybe). In any case, it is our only chance to directly influence the policies of this nation and is a right that thousands died to embody us with. Your voice counts, but to get it heard you have to show up.

I Voted 2018

πŸ‡ΊπŸ‡ΈπŸ» πŸŽ‰

October 31, 2018 @21:50

Getting Started

Tor in Containers

I have been looking for reasons to try Docker on one of the random stack of un-used Raspberry Pis that I have laying around and thought it might be fun to build a little travel router. Somehow that morphed into lets get Tor working on here and then well if I can get a client, and a relay, why not an onion service?

Getting the Tor relay / proxy working was pretty easy. The entrypoint script is a little bit long because I wanted to allow for a fair bit of configuration flexibility in how the container is deployed.

You can find the container in the 'tor-relay' directory of my git repo.

I chose to also put polipo in a container to provide a HTTP proxy front-end. This made it pretty easy to get on the Tor network from a machine anywhere on my LAN. I even threw together a docker-compose.yml to bring up both the Tor client and polipo. You can find that in the tor-proxy-bundle of my git repo. Then I decided to go exploring, err researching.

Tor works!

The "Dark" Web

"Onion" services, often times called "hidden" services are addresses that end in .onion and allow operators to provide services over the Tor network without having to disclose their location or IP address. There is no centralized directory of these services like there is with DNS on the 'regular' Internet so discovering what is out there is a bit tricky. After some searching I found that much like the regular Internet there are various directories of links, and search engines available. The big difference in search engine tech is that they seem to start by crawling the regular Internet looking for .onion addresses and then primed with that they can start crawling and indexing links to other .onion addresses just like any Internet search engine would.

My favorite so far is Fresh Onions because I can sort by 'last seen' and just keep poking at whatever it crawled most recently. Things seem to come and go rather frequently on the Tor network so when I was looking at the link directories I kept finding that something like 60% of the links were dead so this provided a better experience.

After a day of poking around I came to realize that as with most technology the general understanding of the dark web is pretty far from reality. The idea that the dark web is awash with black markets, and hit men for hire, and hackers appears to be about as true as it was when people talked about the Internet back in the early 1990s. In fact the reality is that the dark web even looks an awful lot like the 1990s Internet.

Some Examples

What a nice retriever we have here!

No really, that's the name of the site

There are a lot of sites like this, though this is probably the cutest... The HTML is very rudimentary, quite literally the minimum you need to get an image on the screen.

Placeholder, Placeholder, Everywhere.


Lots of placeholders too... Sometimes not even a page but an empty directory listing from a freshly configured webserver with nothing on it.

Under construction, but no .gif.. yet

Under Construction

If you remember the Internet of the 1990s you almost certainly ran across (or maybe even used) one of the many under construction animated gif images that were out there. While I have yet to see one of those pop up on the dark web, there are lots of pages that purport to be under construction.

Sign my Guestbook!

No really, please sign it!

If you remember the under construction gif you probably also remember guestbooks. A rudimentary precursor to the blog comment box, these let visitors leave public notes for the site owner. Often times these devolved into... well what you might expect from an anonymous board where anyone can post anything...

Turns out those exist on the "Dark" web too.

"Dark" Thoughts

Design aside the "anonymity" of the dark web is very similar to the feeling back in the 1990s and early 2000s Internet. Before advertizers could track you all across the Internet it had a "wild west" feel in places. There were lots of aliases (the hacker nom-de-plume or 'handle'), and strange usernames (I was mee156 at one point thanks to a particularly uncreative corporate IT department) and often they were ephemeral. There were plenty of sites purporting the same sort of potentially illegal (often fake) products and services attributed to the dark web all because by and large you were anonymous (sorta). In a way as someone who grew up in those early days it is actually sort of heartening to see a bit of a renaissance so that maybe the kids today will get a shot at making some of the same mistakes I did and not have that follow them forever.

Hidden service in a (pair of) container(s)

Docker, Tor, Raspberry Pi

There are a lot of reasons people might run an onion service. Nefarious purposes aside if you aren't just using it for research or as a way to provide a link back into your private network then you are probably concerned about anonymity. There have been a few good DEFCON talks about anonymity and security on Tor and how people often screw it up.

While not a silver bullet it seems like putting your service and the Tor client that provides your .onion address into isolated containers are reasonable first steps towards operational security. By isolating the network to just the two containers you can reduce the attack surface and information leaked if the service is compromised. You can also develop the service in isolation (say on your laptop) and then transport it to another machine to deploy it providing an airgap. Beyond that wrapping this into containers makes it simpler to deploy just about anywhere. You could even put them up on a public cloud provider (if you can get an anonymous account), or since this works on a Raspberry Pi you could hide the Pi somewhere other than your home or work and simply snag an open Ethernet port or WiFi network (obviously with permission from the owner...).

Similar to the proxy and relay stuff, you can see an example docker-compose.yml, hidden service client, and Apache instance over in my git repo (under onion-service-bundle, onion-service, apache-rpi respectively).
The example onion service that I have on my Pi right now is available here if you are interested.


Containers bring a lot of interesting possibilities to systems like Tor, where you are essentially creating an overlay network that you are then isolating and keeping largely ephemeral. The onion service keeps a little state (public/private key pairs) but for the most part there isn't anything that needs to be kept around between container runs. There are also other ways to create tunnel connections from inside a container to the world, opening up many different possibilities.

The other interesting thing is that while there are a lot of sites claiming services like what you read in typical reports about "The Dark Web", the vast majority of what is out there are either legitimate attempts to provide anonymous services (eg: The New York Times via Tor, SecureDrop to pass sensitive tips to journalists, and publications or collections of written works like zines), or research / experimentation like the examples above (and my own test service). There is even a streaming radio service over Tor out there.

I think demystifying things helps normalize them. There are plenty of people who use Tor to be able to access the free and open Internet in ways that those of us in countries that don't censor the Internet take for granted, and people who live under regimes so oppressive that reading certain things or posting certain opinions can earn them real jail time. It is important for more people to use Tor for the usual everyday things, provide relays, and run onion services to ensure that the people who are under real threat have more noise to hide in.

Final Thoughts

Peeking around Tor onion services did leave me with one other piece of advice I'd like to pass along. If you have not already I'd urge you to do two things.

  1. Use a password manager. I generally recommend LastPass, but people I know and trust like 1Password as well. This is your best defense the next time some service gets breached and your data ends up out there (more often than not it is found on the regular Internet).
  2. Sign up for Troy Hunt's very good and free service Have I Been Pwned. This will alert you when your data has been found in a data breach.


September 27, 2018 @11:30

I was making some firewall changes last weekend and while watching the logs I discovered that every now and then some host would try to connect to on port 80. This was peculiar since I don't use the IPv4 link local addresses anywhere in my network. It seemed to be happening randomly from all of my Linux hosts, both physical and virtual.

If the processes originated on my firewall which is running OpenBSD I'd be able to track down the process that was doing this by adding a more specific rule with "log (user)" to pf.conf(5) but it seems that Linux dropped this ability from Netfilter sometime back in the 2.6 time frame. 😒

The part that makes this a bit unique is that this is a connection that will certainly fail. It trips a rule in my firewall that blocks bogons outbound on my WAN interface which means that the normal tools like netstat(1) and lsof(8) will only reveal anything if I somehow catch the process between the execution of the connect(2) system call and it failing. What I need to be able to do is log in real time what is happening, which I could do with something like strace(1) but I'd need to know the PID of the process and that is what I'm trying to find out.

So off I went looking for other things that might be helpful and stumbled upon the Linux kernel audit system. The audit system has been around for a while and lets you ask the kernel to communicate the details of syscalls as they happen. There is a filtering mechanism built in so that you don't end up dumping too much information or dramatically impacting performance and the raw data is sent to a userland program via a netlink socket. By default most distributions ship auditd, which listens on that netlink socket and dumps all the messages into a log.

Since I am looking at an attempted TCP connection the connect system call is the one I am interested in. I don't know much else about it though so it turns out a pretty simple filter rule is what I was looking for.

$ sudo auditctl -a exit,always -F arch=b64 -S connect

This asks the kernel to log upon exiting of the syscall any calls to connect(2). This immediately started flooding the audit log with entries like:

type=SYSCALL msg=audit(1538057732.986:13752): arch=c000003e syscall=42 success=yes exit=0 a0=12 a1=7ffc987e28c0 a2=6e a3=7f20abf93dd0 items=1 ppid=9803 pid=19584 auid=4294967295 uid=33 gid=33 euid=33 suid=33 fsuid=33 egid=33 sgid=33 fsgid=33 tty=(none) ses=4294967295 comm="apache2" exe="/usr/sbin/apache2" key=(null)
type=SOCKADDR msg=audit(1538057732.986:13752): saddr=01002F746D702F7...
type=CWD msg=audit(1538057732.986:13752): cwd="/"
type=PATH msg=audit(1538057732.986:13752): item=0 name="/tmp/passenger.PIT9MCV/agents.s/core" inode=1499146 dev=fe:01 mode=0140666 ouid=0 ogid=0 rdev=00:00 nametype=NORMAL
type=PROCTITLE msg=audit(1538057732.986:13752): proctitle=2F7573722F7362696E2F617...

OK, so I'm getting closer but obviously some of the data is coming out in some packed hex format and the things I want aren't all on the same line so I need to figure out how to decode this. While searching for the format of the messages in hopes of writing a quick and dirty parser I found ausearch(8) which includes the handy -i option.

I fired up tcpdump(8) on the pflog(4) interface and waited for one of the packets to be dropped. That let me find what I was looking for in the audit log... the culprit.

ausearch and tcpdump to the rescue

It turns out it was a puppet agent run. Now I know none of my modules try to talk to that address but puppet does a lot of things including running facter to get information about the system and the environment it is running in. I know some cloud infrastructure has standardized on that address as a location for guest agents to pick up metadata from so I suspected some default module trying to see if we are running on a cloud provider. A quick locate(1) and grep(1) later and it turns out that the built in facter ec2 module does in fact try to pull metadata from

apollo@10:16:49 1.8T ~ >locate facter
[ ... many paths elided for brevity ...]
[ ... many more paths ...]
apollo@10:16:53 1.8T ~ >grep -R /usr/lib/ruby/vendor_ruby/facter/*
/usr/lib/ruby/vendor_ruby/facter/ec2/rest.rb:      DEFAULT_URI = ""
/usr/lib/ruby/vendor_ruby/facter/ec2/rest.rb:      DEFAULT_URI = ""
/usr/lib/ruby/vendor_ruby/facter/util/ec2.rb:      url = ""
/usr/lib/ruby/vendor_ruby/facter/util/ec2.rb:  # GET request for the URI  If the
/usr/lib/ruby/vendor_ruby/facter/util/ec2.rb:    uri = "{version}/user-data/"

So in the end, the Linux audit system is our friend. There is a lot of other cool stuff in there, I ran across a post from the slack engineering team that talks about how they use the audit system and how they leverage this information to alert on and challenge user actions in real time. It is also a cautionary tale that good network hygiene is important since you never know what random things you might leak out onto the Internet (or your ISP's network) if you aren't careful.


September 16, 2018 @15:00

I installed one of the Mojave public betas last week on the Mac Mini I have in the office. I used it as an excuse to finally tweak a script I wrote for customizing macOS out of the box.


I'll annotate inline below, you can snag the original if it looks useful to you. The first hunk is just standard shell boilerplate. I tend to write POSIX shell and eschew any bash specific nonsense for maximum compatibility.

# install-macos (c) 2017-2018 Matthew J. Ernisse <>
# All Rights Reserved.
# Customize a base macOS install.
# Redistribution and use in source and binary forms,
# with or without modification, are permitted provided
# that the following conditions are met:
#     * Redistributions of source code must retain the
#       above copyright notice, this list of conditions
#       and the following disclaimer.
#     * Redistributions in binary form must reproduce
#       the above copyright notice, this list of conditions
#       and the following disclaimer in the documentation
#       and/or other materials provided with the distribution.

set -e

I don't use iCloud, I run ownCloud instead. This just makes the directory I use to sync my files to/from.

    if [ ! -d "$HOME/Documents/cloud" ]; then
        echo "Creating ownCloud directory"
        mkdir -p "$HOME/Documents/cloud"

Disable more things I don't use or care about. Touristd is a bit annoying because they can push more crap to bother you with down the line. This at least shuts it up after initial install.

    echo "Disabling Siri"
    defaults write StatusMenuVisible -bool false
    defaults write UserHasDeclinedEnable -bool true
    defaults write 'Assistant Enabled' 0

    defaults write \
        seed- \
        -date "$(date)"

Set a whole bunch of system preferences. Disable automatic spelling and unicode quote correction. Set my preferred Finder style (list view), and enable daily update check and installation. Also try to keep finder from crapping .DS_Store folders all over the network shares.

    echo "Writing various macOS defaults"
    # I suspect I am missing some...
    defaults write NSGlobalDomain AppleKeyboardUIMode -int 3
    defaults write NSGlobalDomain AppleICUForce24HourTime -int 1
    defaults write NSGlobalDomain AppleAquaColorVariant -int 6
    defaults write NSGlobalDomain AppleInterfaceStyle "Dark"
    defaults write NSGlobalDomain \
        AppleMiniturizeOnDoubleClick -bool false
    defaults write NSGlobalDomain AppleShowScrollBars "Always"
    defaults write NSGlobalDomain ApplePressAndHoldEnabled -bool false
    defaults write NSGlobalDomain \
        NSAutomaticCapitalizationEnabled -int 0
    defaults write NSGlobalDomain \
        NSAutomaticDashSubstitutionEnabled -int 0
    defaults write NSGlobalDomain \
        NSAutomaticPeriodSubstitutionEnabled -int 0
    defaults write NSGlobalDomain \
        NSAutomaticQuoteSubstitutionEnabled -int 0
    defaults write NSGlobalDomain \
        NSAutomaticSpellingCorrectionEnabled -int 0
    defaults write NSGlobalDomain \
        NSAutomaticTextCompletionEnabled -int 1
    defaults write NSGlobalDomain NSCloseAlwaysConfirmsChanges -int 1

    defaults write -int 1
    defaults write disable-shadow -bool true
    defaults write FXPreferredViewStyle -string '"Nlsv"'

    # Check for updates automatically, daily, and auto-install
    # security updates
    defaults write \
        AutomaticCheckEnabled -bool true
    defaults write ScheduleFrequency -int 1
    defaults write AutomaticDownload -int 1
    defaults write CriticalUpdateInstall -int 1

    # Don't shit .DS_Store all over the show.
    defaults write \
        DSDontWriteNetworkStores true

As it says, install my internal CA into the trust store.

# Install and trust my local CA.
    echo "Installing CA, you will be prompted for your password"
    local tmpfile=$(mktemp)
    curl --fail \
        --silent \
        --location \
        --insecure \
        --output $tmpfile \

    security add-trusted-cert \
        -k "$HOME/Library/Keychains/login.keychain-db" \

    rm $tmpfile

This gets called to install homebrew and a bunch of applications. There is a nasty hack later as the script needs to be run with sudo, but homebrew won't work that way.

# I hate you so much homebrew for having a fucking trustmeprompt shell
# pipe to a thing installer.  Fuck you.
    echo "Installing Homebrew and sundry applications"
    /usr/bin/ruby -e "$(curl -fsSL"
    brew analytics off
    brew install caskroom/cask/docker
    brew install caskroom/cask/firefox
    brew install caskroom/cask/iterm2
    brew install caskroom/cask/owncloud
    brew install caskroom/cask/visual-studio-code
    brew install caskroom/cask/vnc-viewer
    brew install caskroom/cask/vlc
#   broken?!
#   brew install caskroom/fonts/font-inconsolata
    brew install imagemagick
    brew install flake8
    brew install ffmpeg
    brew install fdk-aac-encoder
    brew install gnupg
    brew install gnutls
# this doesn't seem to function correctly.
#   brew install opensc
    brew install pass
    brew install tmux
    brew install telnet
    brew install pwgen
    brew install wget

Setup my default shell environment.

    echo "Installing .profile"
    curl --fail \
        --silent \
        --location \
        --output "$HOME/.profile" \

Setup autofs, this gets customized a bit for different sites. It also setups the client to match my NFSv4 configuration.

    local nfs_server=""
    local shares="/vol/staff/mernisse /vol/backup /vol/media"
    echo "Setting up NFS automount."

    if [ ! -d "$HOME/Shares" ]; then
        mkdir $HOME/Shares

    if [ ! -f /etc/auto_nfs ]; then
        touch /etc/auto_nfs
        chown root:wheel /etc/auto_nfs
        chmod 644 /etc/auto_nfs

    for share in $shares; do
        if ! grep -q "nfs://$nfs_server$share" /etc/auto_nfs; then
            echo "$(basename $share)        nfs://$nfs_server$share" \
                >> /etc/auto_nfs

    if ! grep -q "$HOME/Shares  auto_nfs" /etc/auto_master; then
        echo "$HOME/Shares  auto_nfs" >> /etc/auto_master

    if ! grep -q nfs.client.default_nfs4domain /etc/nfs.conf; then
        echo "nfs.client.default_nfs4domain = localdomain" >> /etc/nfs.conf

    automount -c

This just says hello and is called at the start of the script.

    local reset="\033[0m"
    local green="\033[32;1;m"
    local yellow="\033[33;1;m"
    local red="\033[31;1;m"
    local magenta="\033[35;1;m"
    local blue="\033[34;1;m"
    local cyan="\033[36;1;m"

    local hello=" ${green}H${yellow}e${red}l${magenta}l${blue}o${reset}"
    hello="${hello}, I am the Macintosh. "

    hello="${hello} $(sysctl -n hw.model)"
    hello="${hello} ${cyan}macOS $(sw_vers -productVersion)${reset}"

    printf "$hello\n"

The default uid/gid doesn't match my network so I change it here. This is why the script needs to be run as root. You do have to be careful with this since it can do wonky things to your login session once it does what its thing.

    if [ "$(id -u mernisse)" -eq 1000 ]; then

    echo "Setting uid to 1000 and creating media group"
    dscl . -change $HOME UniqueID $SUDO_UID 1000
    dseditgroup -o create -i 1042 media
    dseditgroup -o edit -a mernisse media
    echo "Changing ownership of $HOME to reflect new uid"
    chown -R 1000 $HOME

This is the start of execution. It checks to see if you are running as root or doing the homebrew step.

# ublock origin?
if [ ! "$UID" -eq 0 ] && [ ! "$1" = "homebrew" ]; then
    echo "Please run this script with sudo(8)."
    exit 1

Catch the homebrew install which needs to be run as the user, not as root.

if [ "$1" = "homebrew" ]; then
    echo "Returning to sudo session..."

Call all the stuff above.


I replaced all the spinning disks with SSDs a while ago. So I don't need the sudden motion sensor..

echo "Disabling SuddenMotionSensor"
pmset -a sms 0

Some finder related things here. I don't like things being hidden.

echo "Unhiding /Volumes and ~/Library"
chflags nohidden ~/Library
chflags nohidden /Volumes

locate is a good thing to have.

echo "Enabling locatedb"
launchctl load -w /System/Library/LaunchDaemons/

This re-executes the script as the user that ran sudo. This is done to make homebrew happy.

echo "Dropping privs to $SUDO_USER to install homebrew"
echo "************************************************"
# This is a hack...
sudo -u $SUDO_USER $0 homebrew

Finally, change my UID and GID if needed.

# Do this late, because my window session will still have the old UID cached
# it gets... wonky.

And that's it. There is still a few things I have to do manually, including installing a Configuration Profile with my WiFi and VPN settings but it really reduces the amount of things that I have to do to get a new macOS install up and running. πŸ‘ πŸ₯ƒ

September 15, 2018 @16:40

For a while now I've used a Yubikey Neo as a PIV card to authenticate to my public facing hosts. This is fairly straightforward but requires a host with OpenSC on it. In my .profile I have a function called add_smartcard which will add the PIV driver to the ssh-agent. This means I actually authenticate with the key that was generated in the Yubikey and not my password.

Yubikey in my laptop

# add_smartcard - Add the PIV key to the current ssh-agent if available.
# Requires a opensc compatible smartcard and associated libs and binaries.
    # Set to a string in the opensc-tool(1) -l output for your card.
    local _card_name="Yubikey"

    # set to the installed location of the opensc libraries.
    # on OSX with HomeBrew this is /usr/local/lib
    local _lib_dir="/usr/local/lib"

    if ! quiet_which opensc-tool; then

    if [ -z "$SSH_AUTH_SOCK" ]; then

    if opensc-tool -l | grep -q "$_card_name"; then
        if ssh-add -l | grep -q opensc-pkcs11; then

        ssh-add -s "$_lib_dir/"


    # If card is no longer present, remove the key.
    if ssh-add -l | grep -q opensc-pkcs11; then
        ssh-add -e "$_lib_dir/" > /dev/null

Yubikey PIV Authentication

This is all well and good but I wanted to have stronger authentication for scenarios when I'm not on one of my computers. I also wanted to ensure other users of my systems were protected since I can't force them to use PIV cards for authentication. I did a little research and found pam_oath which supports both sequence based and time based one time passwords. This means it is compatible with the OTP profile on the Yubikey and authenticator based apps like Google Authenticator.

The parts I did with Puppet

The first part is pretty straightforward so I setup a Puppet module to do it for me. You need to have the PAM module installed, add it to the sshd pam.d policy, and update your sshd_config.

I am using Debian 9 but you should be able to adapt the following to most Puppet setups and distributions.

# Setup OATH (HTOP) modules/oath/manifests/init.pp
class oath {
  package { 'libpam-oath':
    ensure => latest,

  package { 'oathtool':
    ensure => latest,

  service { 'sshd':

  file { '/etc/users.oath':
    ensure => present,
    owner => root,
    group => root,
    mode => '0600',

  augeas { 'add':
    context => "/files/etc/pam.d/sshd",
    changes => [
      'ins 01 after include[. = "common-auth"]',
      'set 01/type auth',
      'set 01/control required',
      'set 01/module',
      'set 01/argument[last()+1] usersfile=/etc/users.oath',
      'set 01/argument[last()+1] window=20',
    onlyif => 'match /files/etc/pam.d/sshd/*/module[. = ""] size == 0',

  augeas { 'set ChallengeResponseAuthentication':
    context => '/files/etc/ssh/sshd_config',
    changes => [
      'set ChallengeResponseAuthentication yes',
    onlyif => 'match /files/etc/ssh/sshd_config/ChallengeResponseAuthentication != "yes"',
    notify => Service['sshd'],

Things I didn't do with Puppet

The last thing you need is to initalize your shared secrets. I didn't want to do this with Puppet since I felt the need to control where the secrets were and minimize their exposure. The way I have pam_oath configured they will ultimately live in a file called /etc/users.oath. Make sure this is owned by root and has mode 0600. There are two examples that follow for creating the secret. One is for the OATH-HOTP used in the Yubikey, the other is for TOTP which most authenticator apps use (I use the One Time Password feature built into the Hurricane Electric Network Tools app on iOS, but I tested this with Google Authenticator as well).

Create a secret for the Yubikey

> dd if=/dev/urandom count=1 bs=1k 2>/dev/null | sha256sum

Put the returned hexadecimal string into your Yubikey as the shared secret.

Yubico Personalization Tool

Create a secret for an authenticator app

> dd if=/dev/urandom count=1 bs=1k 2>/dev/null | sha256sum | cut -b 1-30
> oathtool --totp -v <hexadecimal key from above>

So if your randomly generated key is 01ab5d053493a266172b16248a8377 then you would see:

imladris@15:27:29 ~ >oathtool --totp -v 01ab5d053493a266172b16248a8377
Hex secret: 01ab5d053493a266172b16248a8377
Digits: 6
Window size: 0
Step size (seconds): 30
Start time: 1970-01-01 00:00:00 UTC (0)
Current time: 2018-09-15 19:37:58 UTC (1537040278)
Counter: 0x30DC773 (51234675)


I used the Python module qrcode which includes a command line utility called qr to generate the configration QR code for the authenticator app. Using the above output of oathtool as an example this is how I made the QR code.

qr "otpauth://totp/<user>@<host>?secret=AGVV2BJUSORGMFZLCYSIVA3X" > qr.png

You can find more information about the otpauth:// uri format on Google's GitHub wiki.


Regardless of which mode you are using you'll need to add the secret to the users.oath file (I am using the example secret from above).

# Yubikey
HOTP    user1   -   01ab5d053493a266172b16248a8377
# Authenticator App
HOTP/T30/6  user2   -   01ab5d053493a266172b16248a8377    

The second line automatically increments the code every 30 seconds. In both cases they will expect a 6 digit code but in the second form it is explicit.

SSH with One Time Password

You can have different users with different methods. There are more complex PAM methods available if you don't want everyone to be required to use MFA or key based authentication, for example this is a good writeup that includes group or host exclusions. I feel like that provides an attacker the ability to work around your MFA.

In the end this probably took me long to write about than actually do and it's enhanced the security of my systems without any negative impact with one exception.

I use Panic's Transmit from time to time and I'm cheap so I have not upgraded to 5.0 yet. It turns out that they don't support the OTP prompt (in 4.0 at least). You can use a custom SSH key, and I believe you can restrict that key to sftp only so I may look into that as a workaround.

In any case there is really no reason you can't secure your servers now.


Edited: September 04, 2018 @23:30

So, I mentioned a while back that I watch Acquisitions Inc on the yubtubs. Well through there I also started watching Dice, Camera, Action. During the Stream of Many Eyes event there was a DCA episode featuring Travis McElroy and that reminded me of the fact that I have had The Adventure Zone languishing away on my iPhone for a while now, un-listened to. Now I'm pretty terrible about keeping up with podcasts (there is so much good stuff out there to listen to and watch these days) so I just wanted to toss out a few words about what happened next.

The Adventure Zone: Balance

I started listening on July 23rd (according to and have basically been blowing through several episodes per day since. There are something north of 90 episodes in the feed and I am within 20 of being current. If that isn't a glowing enough review to interest you then here is a brief synopsis.

The Adventure Zone starts out as a comedy roundtable / actual play ish podcast as three brothers and their father take a stab at playing a starter module for D&D. It's light and funny and clearly a learning experience for all involved. I stuck with it and after the first story 'Here There Be Gerblins' wraps up the first hints of what it will become are unveiled. By the end of the Balance adventure I was hooked. I very literally laughed, cried, and cheered aloud while listening. The production quality went through the roof somewhere about 1/3rd of the way through and the story very quickly left the known world and became something special and unique unto itself. The chemistry of the McElroy family is something delightful to behold and they do an amazing job of morphing D&D into something more consumable in podcast form. It really focuses on the collaborative story telling aspect with often hilarious and serious consequences of perpetually unpredictable dicerolls. I think almost anyone would take a shine to the Tres Horny Boys (as they named their group, accidentally).

As I said am not entirely caught up, I am just up to the start of the new 'season' which was preceded by several short story arcs written by each of the McElroys using different RPG systems and settings as they tried to figure out what they wanted to do for 'season two' but all the mini-stories so far have been really really enjoyable.

The Adventure Zone: Amnesty

I implore anyone who might be reading this to check this out.

Apple Podcasts, RSS Feed


I also forgot to mention that the music in this podcast is completely off the chain.

September 02, 2018 @12:45

Recently I had a rental VW with the fancy new radio in it and I figured I'd give CarPlay a shot.


Welp, I guess I won't be needing that feature when I buy a new car.

August 29, 2018 @09:20

I've been stewing about this for a while and have not yet found an alternative so this is part rant part dear lazyweb plea.

Goodbye, Sonos.

Sonos recently released the 9.0 version of their software which now requires you to have a Sonos account. I have zero desire to sign up for an account or be be in a situation where my home stereo equipment needs to connect to the Internet just to work so I'm actively looking to replace all the Sonos equipment in my home with something else. At the moment the leading idea is to just sprinkle Bluetooth speakers around the house. I don't see any drawback to this approach. with the exception of the Since you need to use a phone or tablet to control the Sonos system there isn't any real drawback to just using Bluetooth audio streaming directly to a speaker.

Honestly since they never got AirPlay or the Android audio streaming equivalent working (for no clear reason since both have been available on Raspberry Pis for YEARS now), nor did they ever support anything other than optical Dolby Digital on the Play:Bar and Play:Base TV speaker products, and since their controller applications just keep getting worse and worse, I am not sad about leaving them. For me, the only nice thing about their hardware that I have found over the years that is missing from most modern network speakers is the inclusion of Ethernet.

So if anyone out there dear lazyweb has an idea of a replacement that doesn't need the cloud to provide base functionality I'd be interested in hearing about it.

πŸ”Š 🍸

August 27, 2018 @17:10

For a long time now the core of my ad blocking strategy has been squid and privoxy running on my OpenBSD routers. Mobile devices VPN into the network and receive a proxy.pac which routes all traffic to these proxies which reject connections to known ad hosts. With the growing adoption of HTTPS (thankfully) privoxy is becoming less and less useful so I have been trying to find better ways to block ads at the networking level.

I'm not going to get into the ethics of ad blocking, it's my choice to make but I will leave this here.

Tay Tay says block ads (source)

Around the same time CloudFlare announced, a privacy focused anycast DNS service. I've been using the Level 3 anycast DNS resolvers for a while now but that's not exactly optimal. With CloudFlare's resolvers you get not only a geographically distributed DNS resolver cluster but DNS-over-TLS and DNS-over-HTTPS support.

Now I run ISC BIND for resolvers, which at this point doesn't support either encrypted DNS method. I do support and validate DNSSEC but that doesn't keep people from eavesdropping on me.

Enter unbound

For a while now OpenBSD has had unbound as the recursive resolver in the base installation so I've been aware of it and trust it. Since I do both recursive and authorative DNS on the same servers I have not had a reason to introduce it. Until CloudFlare.

I added the unbound packages to my DNS server's puppet manifest so the default Debian package got installed. I then added the following configuration to /etc/unbound/unbound.conf.d/cloudflare.conf. Since I'm going to have BIND actually listen to and respond to queries from clients I bind only to localhost (::1 is the IPv6 loopback address) and listen on a non-standard DNS port (5300 since it was open and semi-obvious). This does mean that I have two layers of cache to worry about if I need to clear the DNS cache for any reason but I almost never have to do that so I will worry about that later.

unbound configuration

# This file is managed by Puppet.
# Forward DNS requests to CloudFlare using DNS over TLS.
    verbosity: 1
    use-syslog: yes
    do-tcp: yes
    prefetch: yes
    port: 5300
    interface: ::1
    do-ip4: yes
    do-ip6: yes
    prefer-ip6: yes
    rrset-roundrobin: yes
    use-caps-for-id: yes
    name: "."
    forward-ssl-upstream: yes

I then switched the forwarders section of my named.conf from:

    forwarders {;;


    // Unbound listens on [::1]:5300 and forwards to CloudFlare
    forwarders {
        ::1 port 5300;

After letting puppet apply the new configuration I checked the outbound WAN interface of my router with tcpdump(8) and verified that all DNS resolution was heading off to CloudFlare.

Adding adblocking

unbound(8) has a really nice feature where you can override recursion fairly easily. This can be leveraged to block malicious sites at the DNS layer. I found a couple lists that I was able to plug in that so far have worked really well for me.

The first one is a malware block list that is already provided in the unbound config format. So I just used puppet-vcsrepo to ensure an up-to-date copy is always checked out in /usr/local/etc/unbound/blocks. I was then able to add include: "/usr/local/etc/unbound/blocks/blocks.conf" to the server: section of my unbound config.

Since I also wanted ad blocking I continued my search and came across Steven Black's curated list that consildates a number of difference sources into a hosts.txt format file. Since this isn't exactly the format unbound wants I had to do a little more work.

  1. Checked that repository out with puppet-vcsrepo into /usr/local/etc/unbound/stevenblack.
  2. Wrote the script below to convert the list format from a hosts file to an unbound configuration file.
  3. Configured puppet to exec that script when the vcsrepo pulls an update and then notify (restart) the unbound service.
  4. Added include: /usr/local/etc/unbound/stevenblack.conf to my unbound configuration.

unbound-blocklist script

# unbound-blacklist (c) 2018 Matthew J Ernisse <>
# Generate an unbound style config from a hosts list.

set -e


if [ ! -f "$SRC" ]; then
    echo "Could not open $SRC"
    exit 1

awk '/^0\.0\.0\.0/ {
    print "local-zone: \""$2"\" redirect"
    print "local-data: \""$2" A\""
}' "$SRC" > "$OUTPUT"

The entire puppet manifest for the unbound configuration is as follows. It is included by the rest of the manifests that setup BIND on my name servers.

unbound Puppet manifest

# Unbound - This is the caching recursor.  Uses DNS-over-TLS
# to CloudFlare to provide secure and private DNS resolution.
class auth_dns::unbound {
    package { 'unbound':
        ensure => latest,

    service { 'unbound':
        ensure => running,

    file { '/etc/unbound/unbound.conf.d/cloudflare.conf':
        source => 'puppet:///modules/auth_dns/unbound.conf',
        owner => 'root',
        group => 'root',
        mode => '0644',
        require => [
        notify => [

    exec { 'rebuild unbound blacklist':
        command => '/usr/bin/unbound-blacklist',
        refreshonly => true,
        require => [
        notify => Service['unbound'],

    file { '/usr/bin/unbound-blacklist':
        ensure => present,
        source => 'puppet:///modules/auth_dns/unbound-blacklist',
        owner => root,
        group => root,
        mode => '0755',

    file { '/usr/local/etc/unbound':
        ensure => directory,
        owner => root,
        group => root,
        mode => '0755',

    vcsrepo { '/usr/local/etc/unbound/blocks':
        ensure => present,
        provider => git,
        source => '',
        revision => 'master',
        require => [
        notify => Service['unbound'],

    vcsrepo { '/usr/local/etc/unbound/stevenblack':
        ensure => present,
        provider => git,
        source => '',
        revision => 'master',
        require => [
        notify => Exec['rebuild unbound blacklist'],


So far it feels like a lot of things load faster. I am noticing less requests being blocked by privoxy and squid, to the point that I'm thinking I may be able to completely depricate them. It is also nice that devices on the network that don't listen to proxy.pac files are now being protected from malware and malvertizing as well.


August 26, 2018 @11:30

iPictureFrame and XCode

I know I'm not 'average' when it comes to my opinions about technology. I imagine this has to do with growing up with technology that was much more simplistic than it is today. Compared to modern software and hardware the NEC PowerMate 286 running DOS 6.0 that I learned to program on was extremely simple. Not that it wasn't powerful, but it didn't have any designs to hide things from you. You had access to the hardware directly, and all the memory, and all the peripheral I/O space. You were able to completely control the system, and even understand exactly what was going on.

Not today.

Don't get me wrong, this isn't a bad thing. The protections in modern operating systems are required for the interconnected (and hostile) world we live in. Computers are also powerful enough that you can afford to give the user an API instead of direct access to the hardware (with all the risks that come along with that). The real problem I have is when vendors decide to lock down their consumer hardware to prevent the user from running whatever software they would like on it.

I could easily go off on a rant about Android devices with locked boot loaders, or "smart" TVs with the unnecessary, non-removable, and often poorly supported, and under powered guts, or a myriad of the unfortunate decisions manufacturers are making these days. But that's not what has been bugging me lately. I, like many people if their quarterly filings and trillion dollar valuation is to be believed, have spent a fair amount of money on iOS powered hardware. I expect when I buy a thing that I can basically do whatever I want with it. Now I really do love the security by default stance of iOS but I also believe firmly that as the owner of the device, if I want to shoot myself in the foot, I should be allowed to peel off the warranty void if removed sticker and fire away.

Fucking Apple...

Of course the worst part is that it's not that I'm not allowed to run my own code on my iOS devices. If I have a Mac, and install XCode, and sign up for an Apple Developer account, then for 6 days at a time I can run something I wrote on the thing I bought. To be clear, I'm 100% fine with that being the App Store development experience, however what I want to do is write code for my own personal use on my own personal devices. I don't want any of this software to be transmitted to Apple to be put on the store, or sent in binary form to another person. All I want to do is run my own stuff on things I own.

Now I do understand that my particular use-case might be a bit outside the middle of the bell curve, but I think this is an expectation that isn't unreasonable. I would also point out that if you want to encourage people to learn to code, it might be a good idea to let them actually run their code, and live with it before trying to make a buck off it. In this world of launch early, release often and fix it in a patch release we really do need more people who are used to living with the choices they make. In my case I wrote a silly streaming audio player to help me fall asleep at night that requires a fair amount of infrastructure behind it, so I would never distribute it as a compiled binary, but I'd really like to not have to reload it on my device every 6 days. Similarly I have an iPad 1 and an iPad 2 that are basically useless but would make nice digital picture frames... if only I could run the app that I wrote for more than a few days without having to reload the code on them.

If anyone out there at Apple is listening, I'd really like a way to make my iOS devices trust my internal CA for code signing. Is that really so much to ask?


August 25, 2018 @12:00

UniFi Switches in the NMS

Since I installed the first bits of the Ubiquiti UniFi family of products in my network I have been impressed. They have never failed to meet my expectations. I have written several articles about some rather advanced configuration and implementation details:

And of course I have written several generally glowing reviews about the product line.

The only thing I've been missing is the UniFi switch products. Until recently I have not had a burning need for switches, I had plenty and they worked fine. However, similar to the story of the UniFi USG that replaced a wonky MikroTik RouterBoard I started to have problems with the switch in my detached garage.

Old Garage Linksys

I did actually look around a bit when choosing a switch, there are 3 devices in the garage that are powered by PoE that I really wanted the switch to provide. The Linksys did 802.3af PoE, but two of the 3 devices use the "passive PoE" which isn't compatible so I really wanted to either find a dual mode switch or a source for 802.3af to passive PoE adapters.

Well, the UniFi Switch 8-150W fit the bill perfectly. It has the ability to provide 802.3af, 802.3at (PoE+), and passive PoE over its 8 RJ-45 ports and sports 2 SFP cages as well. As with all the other UniFi stuff installation and configuration was a breeze (in fact I did it using the iOS app while standing in my garage during a downpour), and it has been completely problem free since installation (despite several 100+ degree days where I'm sure the poor thing has roasted).

New Garage USW8-150W

In fact, it worked so well that I kept poking around the UniFi switch line and discovered that there was another switch that scratched a rather odd itch.

At the office, the LAN is provided by the landlord. It is a combination of cable Internet and a bunch of Cisco Meraki gear. I have a PoE+ powered jack in my office that I use for my personal equipment but have not been able to take advantage of it so I have an unsightly mess of power cables and PoE injectors hanging about. Turns out the UniFi Switch 8 (USW8) has just the thing. Port 1 can consume PoE and power the device while port 8 can provide PoE to a downstream device. I was able to eliminate a bunch of crap by dropping in one of these as it was powered from the Meraki switch and in turn powered the downstream access point.

USW8 at work

I think it all came out rather smart. Chalk up another well designed product from Ubiquiti. I actually have another USW8 sitting and waiting to be deployed at home, but I have several holes to cut before that goes in.


July 25, 2018 @20:00

UniFi Security Gateway in the NMS

A couple days ago I wrote a bit about setting up a new Ubiquiti UniFi Security Gateway, and after living with it for a bit I have a few additional notes.

/config/user-data is preserved through resets

I'm not exactly sure why this happened but I fat fingered the JSON and during a provisioning cycle the USG wiped the certificates from /config/auth (where it seems to want you to put them) and while rebuilding I noticed that /config/user-data doesn't get wiped. When you run the restore-default command it seems to have set -x in it somewhere and emits this:

mernisse@ubnt:~$ set-default
+ cmd=restore-default
+ shift
+ case $cmd in
+ exit_if_fake restore-default
++ uname -a
++ grep mips
+ '[' 'Linux ubnt 3.10.20-UBNT #1 SMP Fri Nov 3 15:45:37 MDT 2017 mips64 GNU/Linux' = '' -o -f /tmp/FAKE ']'
+ exit_if_busy restore-default
+ '[' -f /var/run/system.state ']'
++ cat /var/run/system.state
+ state=ready
+ '[' ready '!=' ready ']'
+ state_lock
+ lockfile /var/run/system.state
+ TEMPFILE=/var/run/system.state.4478
+ LOCKFILE=/var/run/system.state.lock
+ ln /var/run/system.state.4478 /var/run/system.state.lock
+ rm -f /var/run/system.state.4478
+ return 0
+ echo 120
+ echo 3
+ rm -f /config/mgmt
+ apply_restore_default
++ cut -c -8
++ echo 7080 27092 31310 11976 31941
++ /usr/bin/md5sum
+ local R=eb2c7606
+ prune_old_config
+ find / -type d -iname 'w.????????' -exec rm -rf '{}' ';'
+ rm -f /config/config.boot
+ rm -f /config/unifi
+ rm -f /config/auth/ca.crt /config/auth/server.crt /config/auth/server.key
+ mv / /
+ state_unlock
+ /bin/rm -f /var/run/system.state.lock
+ reboot

I made a copy of the certificates for the VPN in /config/user-data to ensure that if this happens again I can simply copy them back into place.

You can load a local JSON config file

The core of the UniFi system is the integration to the NMS, otherwise it would just be an EdgeRouter LITE. It appears that the provisioning process causes the controller's configuration to be merged with your config.gateway.json file and sent to the device. The downside is that you can't just push the JSON down to the USG, you need the entire rendered payload. Luckily you do have access to the underlying commands to import and export the configuration.

Once you have the USG up and working you can backup the JSON from the ssh console by running:

mca-ctrl -t dump-cfg > /config/user-data/backup.json

If for some reason the configuration gets messed up and you can no longer talk to the controller because the VPN is down you can simply reload it with:

 mca-ctrl -t apply -c /config/user-data/backup.json

All in all I'm still happy with it minus two things that I've sent to Ubiquiti using their feedback form:

  1. Would really like to have the PEM encoded certificates in the config.gateway.json. This would certainly help if you need to reload the device.
  2. Would like to have a checkbox to bridge eth1 and eth2. Almost everything at the office is wireless, but I do have a Synology NAS that I want wired, thankfully the UniFi UAP-AC-IW that is there has a built in 2 port switch but if I wanted to use a different AP it seems like it would be really handy to be able to easily use the WAN 2 port as a switched LAN port.

🍺 πŸ‘

July 20, 2018 @16:45


I have several physical locations linked together with VPN tunnels. The central VPN server runs OpenBSD with iked(8). I also have several roaming clients (iOS and macOS) that terminate client access tunnels to this system so I am loathe to make breaking changes to it. The site to site tunnels run a gif(8) tunnel in IP-over-IP mode to provide a layer 3 routable interface on top of the IKEv2 tunnel. My internal tunnels run ospfd(8) and ospf6d(8) to exchange routes and my external site to site tunnels run bgpd(8). Most of my internal sites use OpenBSD as endpoints so configuration is painfully simple, however in my office at work I have been using a MikroTik RouterBoard RB951-2HnD. This has worked well enough but lately it has been showing its age, randomly requiring manual intervention to re-establish tunnels and flirting with periods of unexplainable high latency.

Old Work Network


This is not meant to be a comprehensive HOWTO. I doubt your setup will be close enough to mine to translate directly but hopefully you will find some useful information since this isn't a particularly well documented use case for the Ubiquiti UniFi USG product line.

It is also worth noting that under the covers the USG runs the same EdgeOS as their EdgeRouter line of products with the caveat that the controller will overwrite the configuration any time it provisions the device. Fortunately Ubiquiti has foreseen this and provides a way to provide advanced configuration via a JSON file on the controller.

I manage all of my sites from a centralized UniFi controller instance, so I need the VPN to work before I can swap out the RouterBoard for the USG. This is an overview of how I did that.


Since I already have a working VPN tunnel at the site I already had all the X.509 certificates and IP addresses needed to configure the new router. Starting at home, where the controller is located I plugged in the USG WAN port to my LAN and connected my laptop to the USG LAN port. I was able to adopt the gateway into the controller with no trouble.

I fiddled around with the config until I got it working and stuffed the changes into the config.gateway.json file. Finally I blew the device away and forgot it from the controller. It is important at this point to reload the certificates into the factory defaulted router (put them in /config/auth) before adopting the gateway in the controller. The gateway will go into a reboot loop much the same way as if you typo-ed the config.gateway.json file if it cannot find the files. Once the certificates were loaded, I re-adopted the gateway and the configuration was applied.

I was then able to take it into work and swap the MicroTik.


I will simply annotate the config.gateway.json file inline to explain how this all ended up going together.

    "service": {
        "dns": {
            "forwarding": {
                "options": [

Set the DNS domain name handed out by the gateway, not strictly needed in this context, but handy.

        "nat": {
            "rule": {
                "6004": {
                    "description": "VPN Link Local NAT",
                    "destination": {
                        "address": "!"
                    "log": "disable",
                    "outbound-interface": "tun0",
                    "outside-address": {
                        "address": ""
                    "source": {
                        "address": ""
                    "type": "source"

NAT any traffic coming from the tunnel or IPSec endpoint addresses to the canonical address of the router. This prevents local daemons from selecting the wrong source IP (most frequently done by syslogd).

    "interfaces": {
        "loopback": {
            "lo": {
                "address": [

This is the IPSec endpoint, I use policy based IPSec so this needs to exist somewhere so the traffic can get picked up by the kernel and sent across the tunnel.

        "tunnel": {
            "tun0": {
                "address": [
                "description": "ub3rgeek vpn",
                "encapsulation": "ipip",
                "ip": {
                    "ospf": {
                        "network": "point-to-point"
                "local-ip": "",
                "mtu": "1420",
                "multicast": "enable",
                "remote-ip": "",
                "ttl": "255"

This sets up the IP-over-IP tunnel. Note I could not get the OSPF session to come up for the life of me using my normal /32 addressed tunnel so I switched to a /30. After that OSPF came right up. If you debug ospf events and get complaints that the peer address of tun0 is not an ospf address, then you might be hitting this too.

    "protocols": {
        "ospf": {
            "area": {
                "": {
                    "network": [
            "parameters": {
                "abr-type": "cisco",
                "router-id": ""
            "passive-interface": [

This is rather straightforward, I'm redistributing the local networks and the tunnel address. This is a pretty simple OSPF configuration. Since I have no routers on the Ethernet end of things I set both interfaces to passive.

    "vpn": {
        "ipsec": {
            "auto-firewall-nat-exclude": "enable",
            "esp-group": {
                "ub3rgeek": {
                    "compression": "disable",
                    "lifetime": "3600",
                    "mode": "tunnel",
                    "pfs": "dh-group14",
                    "proposal": {
                        "1": {
                            "encryption": "aes256",
                            "hash": "sha256"
            "ike-group": {
                "ub3rgeek": {
                    "ikev2-reauth": "no",
                    "key-exchange": "ikev2",
                    "lifetime": "28800",
                    "proposal": {
                        "1": {
                            "dh-group": "14",
                            "encryption": "aes256",
                            "hash": "sha256"
            "site-to-site": {
                "peer": {
                    "": {
                        "authentication": {
                            "id": "",
                            "mode": "x509",
                            "remote-id": "",
                            "x509": {
                                "ca-cert-file": "/config/auth/ca.crt",
                                "cert-file": "/config/auth/server.crt",
                                "key": {
                                    "file": "/config/auth/server.key"
                        "connection-type": "initiate",
                        "ike-group": "ub3rgeek",
                        "ikev2-reauth": "inherit",
                        "local-address": "default",
                        "tunnel": {
                            "0": {
                                "allow-nat-networks": "disable",
                                "allow-public-networks": "disable",
                                "esp-group": "ub3rgeek",
                                "local": {
                                    "prefix": ""
                                "protocol": "all",
                                "remote": {
                                    "prefix": ""

This is the real meat and potatoes of the configuration. It corresponds to the following configuration on the OpenBSD side of things.

ikev2 "work" passive esp \
        from to \
        peer $PEER_WORK \
        ikesa enc aes-256 \
                auth hmac-sha2-256 \
                prf hmac-sha2-256 \
                group modp2048 \
        childsa enc aes-256 \
                auth hmac-sha2-256 \
                group modp2048 \
        srcid dstid \
        lifetime 360m bytes 32g


In the end I am very happy about the whole thing. The USG is pretty slick and for simple configurations I imagine it is super easy to get going, and other than the lack of documentation for some of the things that aren't exposed in the controller UI it was not too hard to figure out. I would suggest if you are stuck trying to figure out the cli, you might want to explore the EdgeOS or Vyatta (the upstream Open Source project the EdgeOS is based on) documentation. I found those helpful.

New Work Network


July 12, 2018 @20:47

I enabled HTTPS on this website just under a year ago. If you follow my blog you know that this is a static website, and since there appears to be a bit of an uproar in the web community over HTTPS right now I figured I'd simply weigh in.

Do you need HTTPS for your website?


There are lots of good reasons for this, and not many reasons not do it but the major point that resonates with me is not the risks to your website, but the risks to the general Internet at large. Actors (both malicious and benign) can inject content into any HTTP served site and cause the web browser of their visitors site to do... essentially whatever they want. This doesn't have to be targeted at your site, anyone in the middle can simply target ALL HTTP traffic out there, regardless of the content.

This isn't a user agent (browser) problem, this isn't a server problem, anyone with access to ANY part of the network between the server and the user agent can inject anything they want without the authenticity provided by TLS.

HTTPS is Easy, and for most it is free. It also allows HTTP/2 which is faster (even for static sites like this one which uses HTTP/2). Really it is. If you aren't convinced let me also point you at Troy Hunt's excellent demo of what people can do to your static website.

Subscribe via RSS. Send me a comment.