Posted: 2019-07-30 11:15:33 by Alasdair Keyes
For about 10 years I've used a wiki to document everything that I learn and need to keep track of. This contains everything from walkthroughs of installing/configuring software, to lists of interview questions to ask potential hires.
When I first started working in hosting, I began collecting text files with information given to me by other colleagues. Over time this got un-wieldy so I created a MediaWiki wiki https://www.mediawiki.org/wiki/MediaWiki. I mainly picked this as it was both a wiki I was using at my workplace and it was a common interface; being the software that Wikipedia uses.
Over time I've kept Mediawiki updated but gradually I've had more and more problems with updates breaking and needing fixing so I started looking around for other wiki tools.
I eventually found Dokuwiki https://www.dokuwiki.org/. It's more lightweight and simple but seems to be up to the tasks that I need it for. It uses flat files as a back-end so I don't need to backup both files and a database and after importing all my data it's only 1/4 of the size on disk.
$ du -hs public_html.mediawiki/ 203M public_html.mediawiki/ $ du -hs public_html.dokuwiki 48M public_html.mediawiki/
I did have to install the
pagelist Dokuwiki plugins to allow me to use tags, which are the Dokuwiki version of Mediawiki's categories.
It would be nice to have been able to copy my articles directly across to the new wiki, but the Mediawiki syntax (https://www.mediawiki.org/wiki/Help:Formatting) and Dokuwiki syntax (https://www.dokuwiki.org/wiki:syntax) are different. The key differences were
<pre></pre>tags in Mediawiki that needed to be converted to
<code></code>tags in Dokuwiki that needed to be converted to
___had to be converted.
I knocked up a quick Perl script to connect to the Mediawiki DB and parse the articles into a format suitable for Dokuwiki. This was mostly done with regex replace statements to insert spaces and change tags etc.
While I was at it, I took this time to delete or update any old articles. So now I have a new wiki with refreshed info and am very pleased with Dokuwiki.
If you found this useful, please feel free to donate via bitcoin to 1NT2ErDzLDBPB8CDLk6j1qUdT6FmxkMmNz
Posted: 2019-07-17 08:47:31 by Alasdair Keyes
I upgraded my home server to Debian 10 (Buster) this week. It's running on quite an old HP Proliant Microserver so I bought a new SSD to use for the OS partitions to give it a little extra life. As such, it was a fresh install rather than an in-place upgrade.
As you would imagine 10 is much the same as 9 in most respects. But there were a couple of points of note...
The Buster Puppet install was using version
5.5.10 whereas my Puppet Master (On Debian Stretch) was using
4.8.2 when connecting to the master the new install would error with
Warning: SSL_connect returned=1 errno=0 state=error: dh key too small
The answer to this was found at another chap's blog https://blog.steve.fi/upgraded_my_first_host_to_buster.html and is to do with system-wide SSL settings, although I fixed it slightly differently.
/etc/ssl/openssl.cnf I updated the line
CipherString = DEFAULT@SECLEVEL=2
CipherString = DEFAULT@SECLEVEL=1
It turns out this is a non-standard, custom security setting made by Debian https://wiki.debian.org/ContinuousIntegration/TriagingTips/openssl-1.1.1
It doesn't appear that you can define a custom set of Diffie Hellman params for a Puppet Master as you can for other software like NGINX and Apache. As soon as I have my Puppet Master on the later version I'll be changing this setting back, assuming it doesn't interfere with anything else.
check_disk_ioNagios plugin was failing
It turns out the output of the
iostat command had changed slightly and required a tweak to continue working. Commit https://gitlab.com/alasdairkeyes/nagios-plugin-check_disk_io/commit/0708ba7b9cb0017f6f36554d54ee3e37a9b58d63
debsecanpackage is enabled by default
I wasn't aware this package existed until it started emailing me with all the system vulnerabilities. I can see a use for it, but as my systems are updated regularly, it's now purged by Puppet.
SMBus PIIX4 adapterdevice
sensors utility used by the
check_sensors Nagios plugin was erroring that I had a critical alarm.
It turns out that there is no max/critical temp information for the thermometer on this device so the reported temperature is always higher than the threshold of 0C
# sensors ... jc42-i2c-0-18 Adapter: SMBus PIIX4 adapter port 0 at 0b00 temp1: +31.0°C (low = +0.0°C) ALARM (HIGH, CRIT) (high = +0.0°C, hyst = +0.0°C) (crit = +0.0°C, hyst = +0.0°C) ...
As I have other temperature sensors available I disabled this one by creating the following file
chip "jc42-i2c-0-18" bus "i2c-0" "SMBus PIIX4 adapter port 0 at 0b00" ignore temp1
Other than that it was all pretty seamless.
If you found this useful, please feel free to donate via bitcoin to 1NT2ErDzLDBPB8CDLk6j1qUdT6FmxkMmNz
Posted: 2019-07-09 12:45:23 by Alasdair Keyes
I started up the Tor browser yesterday and noticed that it didn't start in it's usual time frame, 10 minutes later the browser had still not opened.
Checking top, I saw that a GPG process was using 100% CPU.
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 19330 username 20 0 78364 47020 4448 R 99.7 0.6 0:16.43 gpg 3145 username 20 0 3458164 139712 63512 R 12.6 1.7 18:43.51 cinnamon
I'd read recently about an attack on GPG where keys were being poisoned with a large number of signatures to exploit a GPG bug and corrupt GPG installs https://threatpost.com/pgp-ecosystem-targeted-in-poisoning-attacks/146240/, I wondered if this is what was occuring.
I checked what the GGP process was running.
$ ps aux | grep 19330 username 19330 64.6 0.6 82192 50980 ? RL 10:51 0:31 /usr/bin/gpg --status-fd 2 --homedir /home/username/.local/share/torbrowser/gnupg_homedir --keyserver hkps://hkps.pool.sks-keyservers.net --keyserver-options ca-cert-file /usr/share/torbrowser-launcher/sks-keyservers.netCA.pem include-revoked no-honor-keyserver-url no-honor-pka-record --refresh-keys
It seemed to be running
--refresh-keys which requests updates to keys from the key servers. I ran the following to see what keys were being refreshed.
$ /usr/bin/gpg --homedir /home/username/.local/share/torbrowser/gnupg_homedir --list-keys /home/username/.local/share/torbrowser/gnupg_homedir/pubring.kbx ---------------------------------------------------------------- pub rsa4096 2014-12-15 [C] [expires: 2020-08-24] EF6E286DDA85EA2A4BA7DE684E2C6E8793298290 uid [ unknown] Tor Browser Developers (signing key) <firstname.lastname@example.org> sub rsa4096 2018-05-26 [S] [expires: 2020-09-12]
I checked the key servers entry for the key
EF6E286DDA85EA2A4BA7DE684E2C6E8793298290 at http://pgp.mit.edu/pks/lookup?op=vindex&search=0x4E2C6E8793298290 and saw the key had received a large number of signatures on 2019-06-30, it does indeed look like it has been poisoned with excessive signatures.
I downloaded the latest Tor Browser for Linux directly from https://www.torproject.org/ and didn't receive this issue during startup which is good news.
However, my tor install is through the
torbrowser-launcher provided by the Linux Mint repos (originally provided by Ubuntu).
torbrowser-launcher doesn't contain the TOR Browser itself (as the name suggests, it's just a launcher), it is a python environment that will download the latest Tor Browser directly from Tor project. To do this, it uses the Tor Project's public GPG Key to verify the downloaded files are legitimate, during this process it does a refresh from the key servers and hits the poisoning issue.
It seems if you are affected by this, you're best off downloading tor direct from the Tor Project itself. Unfortunately, verification that the file you download from the website requires gpg, you can certainly try and ensure that the key that created the signature is correct...
$ gpg --verify tor-browser-linux64-8.5.3_en-US.tar.xz.asc Downloads/tor-browser-linux64-8.5.3_en-US.tar.xz gpg: Signature made Fri 21 Jun 2019 02:30:51 PM CEST gpg: using RSA key EB774491D9FF06E2 gpg: Can't check signature: No public key
EB774491D9FF06E2 matches the key listed at https://2019.www.torproject.org/docs/verifying-signatures.html.en and is a subkey for the Tor Project Signing key, but without the key in your keyring, this check isn't as secure as it should be.
If you found this useful, please feel free to donate via bitcoin to 1NT2ErDzLDBPB8CDLk6j1qUdT6FmxkMmNz
Posted: 2019-07-09 09:06:41 by Alasdair Keyes
I've been interested in running ZFS for a while but have always held off making the leap due to worries about features and stability. ZFS was originally developed for Solaris and has been ported over to Linux by the ZFS on Linux (ZoL) project https://zfsonlinux.org/.
Recently ZoL 0.8 was released with native encryption which is really a must. Unfortunately the latest Debian release 'Buster' only has 0.7.12 so the native encryption feature isn't available.
I've been experimenting with a Virtualbox VM to develop and test a suitable setup that I would be happy with on my production hardware.
My existing production setup runs Debian Stretch using Linux software Raid with LUKS Encryption on top and running ext4 as a filesystem.
For this test setup I'm using Virtualbox with 4x 2GB disks for ZFS with Striped/Mirrored configuration, it's essentially ZFS's version of RAID 10. For a configuration like this you should ensure you have at least 2GB RAM, I did try with 1GB however the LUKS encrypted devices were failing to startup at boot with out of memory errors. Debian 'Buster' is the OS.
The Disk setup is
/dev/sdbZFS disk 1
/dev/sdcZFS disk 2
/dev/sddZFS disk 3
/dev/sdeZFS disk 4
apt update && apt upgrade -y
contribto the Debian apt repo list in
deb http://deb.debian.org/debian buster main contrib deb-src http://deb.debian.org/debian buster main contrib
apt update && apt install dpkg-dev linux-headers-amd64 cryptsetup -y
This can take some time, make a cup of tea.
apt install zfs-dkms zfsutils-linux -y
cryptsetup -y luksFormat /dev/sdb cryptsetup -y luksFormat /dev/sdc cryptsetup -y luksFormat /dev/sdd cryptsetup -y luksFormat /dev/sde
Get the UUID for each LUKS device
# ls -l /dev/disk/by-uuid/ total 0 lrwxrwxrwx 1 root root 10 Jul 9 08:55 0af47096-987d-41b5-b5a7-98827850f46d -> ../../sda1 lrwxrwxrwx 1 root root 9 Jul 9 08:55 5888dfc8-4df0-410e-8aec-992aad7abd97 -> ../../sdc lrwxrwxrwx 1 root root 9 Jul 9 08:55 abd4a557-de16-4ecd-ab73-e4d41293dcf4 -> ../../sde lrwxrwxrwx 1 root root 9 Jul 9 08:55 e2f1931b-2413-4181-9500-baad1a74c12d -> ../../sdd lrwxrwxrwx 1 root root 9 Jul 9 08:55 edc129d6-dc90-4338-bc2e-9476843ff41f -> ../../sdb lrwxrwxrwx 1 root root 10 Jul 9 08:55 fc1b09a1-41e2-4503-8c4f-d2e532dea5aa -> ../../sda5
/etc/crypttab file with your disk configuration, it should look similar to this, the target name can be any unique name that you want.
# <target name> <source device> <key file> <options> sdb_crypt UUID=edc129d6-dc90-4338-bc2e-9476843ff41f none luks sdc_crypt UUID=5888dfc8-4df0-410e-8aec-992aad7abd97 none luks sdd_crypt UUID=e2f1931b-2413-4181-9500-baad1a74c12d none luks sde_crypt UUID=abd4a557-de16-4ecd-ab73-e4d41293dcf4 none luks
As you can see the UUID mapping in
/dev/disk/by-uuid is mapped against a unique name for device mapper.
This isn't required, however it's good to ensure that your LUKS setup is correct before proceeding. You will be asked for your LUKS passwords on boot. Once you log back in again, you should be able to run the following
ls and see the LUKS devices are initialized correctly
$ ls -l /dev/mapper/ total 0 crw------- 1 root root 10, 236 Jul 9 08:55 control lrwxrwxrwx 1 root root 7 Jul 9 08:55 sdb_crypt -> ../dm-0 lrwxrwxrwx 1 root root 7 Jul 9 08:55 sdc_crypt -> ../dm-1 lrwxrwxrwx 1 root root 7 Jul 9 08:56 sdd_crypt -> ../dm-3 lrwxrwxrwx 1 root root 7 Jul 9 08:55 sde_crypt -> ../dm-2
You will sometimes get a warning that the zfs kernel module isn't loaded, just follow the instructions and run...
This will only need to be run once, once a pool is configured the module will be loaded automatically.
# zpool create pool01 mirror /dev/mapper/sdb_crypt /dev/mapper/sdc_crypt mirror /dev/mapper/sdd_crypt /dev/mapper/sde_crypt
Check the setup
# zpool status pool: pool01 state: ONLINE scan: none requested config: NAME STATE READ WRITE CKSUM pool01 ONLINE 0 0 0 mirror-0 ONLINE 0 0 0 sdb_crypt ONLINE 0 0 0 sdc_crypt ONLINE 0 0 0 mirror-1 ONLINE 0 0 0 sdd_crypt ONLINE 0 0 0 sde_crypt ONLINE 0 0 0 errors: No known data errors
Rebooting again will ensure that everything is configured and the LUKS devices are brought up before ZFS mounts the pool, otherwise you will end up with ZFS errors and the pool won't load.
zpool status again and you should see the same output as above. If the LUKS devices fail to initialize and none of the devices are available, you will see an error about
no pool available.
If only some of the LUKS devices fail to initialize you will see the state being something other than
ONLINE and you can check
/var/log/kern.log for information as to why.
Posted: 2019-07-02 09:15:22 by Alasdair Keyes
I recently registered a new .uk domain, and setup a basic website on behalf of a client. I checked the logs to see how long it took for it to be accessed without me having to advertise it's presence. The timeline is
X.X.X.X - - [28/Jun/2019:17:09:47 +0100] "GET / HTTP/1.1" 301 178 "-" "Mozilla/5.0 (Macintosh; Intel Mac OS X 10.12; rv:52.0) Gecko/20100101 Firefox/33.0"
The site uses a name-based virtualhost so the visitor had to specifically request the domain rather than just hitting port 80/443 on the server IP. Within ~6 hours of registration, the domain was already being scanned. What's of further interest is that at 17:09 at the same second two separate IPs both hit the index page for the first time, indicating it was likely a bot doing a coordinated scan of new sites.
As far as I know the domain hadn't been registered for a while (if ever) and as .uk domains don't release new registrations, the most likely way for bots/people to be aware of the new website was from the HTTPS Certificate Transparency logs. If you're unaware, every new secure certificate that's issued is published to a public log, these can be searched via a number of sites such as https://crt.sh/ (and you can see all certificates issued for akeyes.co.uk here https://crt.sh/?q=akeyes.co.uk).
The take-away from this is that you should be aware that nothing goes unnoticed on the web anymore, if you're setting up a new website, ensure that it is secure from the get-go. Make sure passwords are changed from defaults and are secure and ensure software is up-to-date as bots will be looking to exploit it, this is especially important for popular CMS apps like Wordpress.
As an aside, I found it interesting that Bing had crawled the domain within 24 hours, and Google has yet still to visit.
Posted: 2019-06-26 08:38:57 by Alasdair Keyes
Last year I made a post on Wireguard and wrote a Nagios plugin to allow monitoring of connected peers. I mentioned that I would likely do a post about my thoughts on Wireguard later, and here it is...
Before I begin, this isn't a copy and past piece about how it's a slim code base and listing off the encryption algos, that is all important, but is covered in-depth in every article about Wireguard on the internet. This is viewed from a more user/admin point of view.
I will also refer to Server/Client paradigm however Wireguard seems to only operate on the idea of Peers, essentially, a "Server" would be a server with lots of peers connecting and routing traffic through it and a "Client" would be a Peer that connects to a single (or limited number) of peers and routes some/all of it's traffic across the interface.
It should be noted that this is tested using Debian Linux. Wireguard is available for lesser operating systems.
Moving on from the fact that I won't just list off the protocols it uses internally, Wireguard's use of limited encryption algorithms, ciphers etc means that as a sysadmin, I know that I can not actively downgrade or harm my VPN's security.
With tools like OpenVPN, having a range of ciphers and ability to choose different key lengths is good, but at some point I will forget to update these and eventually be running it with a key size that's too small or a cipher that has a known flaw. Large companies may security review their setups regularly but small companies or personal users will most likely not.
This does lead to the potential problem of one vulnerability potentially affecting all Wireguard installs due to similar configuration, however this can occur with any software and I don't have to worry that my lack of knowledge or ability are actively making the tunnel less secure than it should be.
Other nice extras are that Wireguard operates on asymmetric cryptography with public/private keys but also gives the option of a pre-shared key per-client for extra security (especially for say post-quantum world) and it also offers Perfect Forward Secrecy (PFS) so even if private keys are leaked previous session data is still secure.
The client (or 'peer' in Wireguard parlance) configuration file is very light weight. Often less than 10 lines of config, Private/Public and optional pre-shared keys are all included in-line in the file and are very small. No more need to hand out CA certs, private keys etc on top of config files to users.
The config file can also contain
PostUp etc. type commands to enable firewall changes or other relevant tasks that should be performed so you don't have to find ingenious ways of hooking it in with other things on your system.
Versions of Gnome Network Manager have support for Wireguard making configuration even easier for the non-tech savvy.
This part is quite impressive and well thought out. Wireguard config is stored in a single file and can either be edited directly in the file if the interface is down, or configured in realtime using the
wg tool when the interface is up (You have the option as to whether these changes are persisted or temporary until the interface is brought down).
The tools and man pages have great detail and are easy to follow and the general amount of limited options mean that there's not too much to get wrong with configuration.
There is also a
wg-quick tool which will bring up interfaces and configure default routing for you too.
Having the functionality for editing config via CLI is great for automation. I built a puppet module this weekend to configure a Wireguard server and the
wg-quick tools were invaluable.
As far as I know Wireguard hasn't been security reviewed. This is not surprising, it's still in development and it takes a lot of time and effort for software to be reviewed but it will be interesting to see the results when it finally does happen.
Due to the connection-less way Wireguard works there is not defined list of peers that are connected/unconnected. The server knows how long it has been since a handshake has occurred and started a new PFS session with peers but not if a peer is actually connected. This is also in part due to the use of UDP (Tunnelling TCP over TCP has some problems so UDP is best here). Connection information can be extrapolated (as the Nagios plugin does), but it would be nice to know how many connections there are. Connection numbers can be a good way of knowing early on if there are any problems.
When viewing the output of Wireguard's configuration all peers are defined only by their public key. This is good for providing some level of anonymity, but if you were running a large organisation with a lot of Wireguard peers, it would be handy to have a
nice-name field to indicate either a particular real-world person or perhaps the data-centre that is on the other end of the interface. This can be added into the config file as a comment, but it would be nice to see it added as an optional extra in the config.
Wireguard is touted as being very fast due to both it's slim code and the way it's designed to operate.
My VPN servers generally don't have too many users so I can't make a direct useful comparison. The Client's network speed seemed neither faster or slower than an OpenVPN connection. If I had done some in-depth checks I may have seen a reduction in CPU/RAM/Network use, but really, who has the time?
All in all I like Wireguard and plan on moving to it soon. Maybe in tandem with my existing VPN software until I have confidence that it is suitable.
I've written a puppet configuration to roll out once I'm ready. The only thing holding me back is waiting for my desktop OS to ship Network Manager with a Wireguard plugin so I can play nicely with my general network configuration.
There are a million articles on line for how to get Wireguard up and running and if you use VPNs I would suggest at least looking in to it.
Posted: 2019-03-15 22:06:15 by Alasdair Keyes
I was refreshing my memory on the Bridge pattern (https://en.wikipedia.org/wiki/Bridge_pattern) for some code I was writing and I came across this Github repo https://github.com/domnikl/DesignPatternsPHP with PHP implementations of many common design patterns.
It's well worth bookmarking for when you need to brush up or even use as a framework for implementing them.
Posted: 2019-02-15 15:51:23 by Alasdair Keyes
Anyone who works in the modern corporate environment is well aware that the landscape changes quickly and that it's very easy to become complacent, resting on previous experience and not updating your skill-set for the future.
This is why I have undertaken the Montana Fish, Wildlife and Parks bear identification test to hone my ursine identification skills. With a quite staggering 86.7% accuracy, I am able to identify the difference between the types of bear found in Montana and surrounding states.
I have never been to Montana, I have no plans to visit Missoula or Great Falls. But I know that should I ever get caught in the North-Western wilderness of the United States I will be able to fall back on my above-par identification skills.
Posted: 2019-01-19 16:01:55 by Alasdair Keyes
I think I've linked to Computerphile videos in the past, but they definitely deserve it's own post.
Computerphile is a Youtube channel created (I believe) some staff/faculty at Nottingham University.
They release regular videos for both beginners and experts, explaining everything from how recently announced vulnerabilities work, through to low-level assembly and CPU operations. It's taught me a lot e.g. how Face ID works and how double-ratchet message encryption is employed to keep tools like Whatsapp and Signal secure, to name just two.
If it's the kind of thing that tickles your fancy it's quite a rabbit hole so wait until you have a spare Sunday afternoon and get to it.
Posted: 2019-01-17 17:01:47 by Alasdair Keyes
A while back I wrote a library to easily allow creation of redirect URLs on the fly called redirecttoken https://gitlab.com/alasdairkeyes/redirecttoken/.
It was a cleaner implementation of a setup I used on this site to see what links visotors are clicking.
After I migrated my site from Slim to Laravel about 2 months ago, I created a Laravel Service Provider to easily port this functionality into my new site. This has also just been released at https://gitlab.com/alasdairkeyes/redirecttoken-laravel/.
A basic config is configurable through the
.env file so no need for any additional coding to add it to your Laravel install.
Both packages are on Packagist and can be installed using composer.
I'm now available for IT consultancy and software development services - Cloudee LTD.
Happy user of Digital Ocean (Affiliate link)