SKS (Synchronising Key Server) Setup
The documentation wiki for SKS has been down for a while and building it on Ubuntu Dapper
turned out to require a couple of minor tweaks, so this is my crib sheet for how to get it
up and running. It's a compilation of wiki content extracted from Archive.org, the README,
and my own notes.
Note: Thanks to Andy Ruddock there is now a guide to installing from the Ubuntu packages
Installing the OpenSKS keyserver.
On Debian/Ubuntu there are a few dependencies to install first:
apt-get install ocaml-dev
tar zxf sks-
mv Makefile.local.au Makefile.local
Edit Makefile.local: change "libdb-4.1" to "libdb4.2";
change "BDB=-L/usr/lib" to "BDB=-L/usr"
Rather than starting with an empty database and attempting to populate it by syncing with
other keyservers (a bad idea because it loads up your peers with lots of traffic and will probably
fail anyway with deadlocks in the conflict resolution system) we'll grab a static dump from an
existing SKS server. Currently the only known source is:
http://keyserver.borgnet.us/dump - Generated every Sunday
The keydump is about 4.3GB as of May 2011, so fetching it will take a long time. It's
divided into a bunch of individual numbered files so you'll need to fetch all of them. Because
I'm too lazy to spend 8 hours sitting there doing it manually I did it like this:
wget --recursive --timestamping --level=1 --cut-dirs=3 --no-host-directories ftp://key-server.de/dump/
Many hours later, check that all the pieces downloaded correctly by comparing their checksums
against the list published by the dump provider:
md5sum -c MD5SUMS
Build Local Database
There are two ways to do this: either a full build (which reads in the dump you just downloaded
and leaves you with a complete, self-contained database) or a fastbuild (which just references
the dump and requires it to be left in place after the fastbuild is complete). I started doing
a full build, it looked like it was going to take forever so I aborted it and switched to a
quickbuild. On the 4-processor machine I was using it still took in the order of 40 minutes to
run so this might take a while.
You need to be in the basedir when running this and the dumps have to be in a sub-directory
called "dump" (which they should be if you followed the steps above), so:
If you edit the sks_build.sh script you'll discover it's just a shell script which calls SKS
itself to do the heavy lifting. If you have trouble with lack of memory you may need to tweak the
script a bit: in particular the "-n 10" flag used in the fastbuild call is a multiple of 15,000
keys to load at a time. The default setting therefore loads 150,000 keys at a time which could
cause your machine to go into swap, and changing to something like "-n 2" will cause it to load
only 30,000 at a time instead and possibly complete the job faster. The trick is to load as
many as possible in each pass without hitting swap - if that happens, performance falls through
the floor and you may as well abort it and start again (after deleting the KDB and PTree
directories created by the aborted import).
If all goes smoothly you'll end up with KDB and PTree directories in /sks.
Create SKS Config File
When it starts up SKS looks for several config files in the current directory. I just put
everything in /sks for simplicity, although it should probably be split up with config in /etc/sks
and the database in /var/lib/sks.
/etc/sksconf is the main config file which is read on startup. This is my
actual config file with my values, so change to suit your requirements:
# sks configuration for keys.keysigning.org
# Set the logfile to $basedir/log.[db|recon]
# debuglevel 4 is default (max debug level is 10)
# Set the hostname of the server
# Set the email address we use to sync with other keyservers
Set Up Peer Communications
It's important that all database changes made on individual SKS and PKS
keyservers propogate not only to other servers but also across both networks.
Keyservers (generally) should not operate in isolation: each one needs to
be part of the mesh so it will receive all changes made on other servers and its
own changes will be distributed as well.
Internally the SKS network uses a "gossip" protocol to propagate
changes around, so as long as your keyserver is gossiping with other servers
(your "peers") which in turn gossip with others etc etc then you'll be fine internally
and all changes will eventually make it to and from your server. You don't need to
be gossiping with every other SKS server, just a couple: as long as a chain
can be traced from your keyserver through to any other keyserver, you're set. Changes
can then "ripple" through the keyserver network until they touch every server.
You need to find other servers for your new server to gossip with, and put an entry
for them in your
/sks/membership file like this:
# List of other keyservers to peer with
keys.keysigning.org 11370 # Jonathan Oxer <email@example.com> 0x64011A8B
In the example snippet above you'll see that most of the lines have only a hostname
and a port. That's the absolute minimum information you need to include for SKS to
operate correctly. However, you'll see that one of the entries has a trailing comment
that includes the name, email address, and key ID of the admin for that particular
keyserver: that's the "best practice" format for a complete entry, and provides you
a handy reference for how to get in touch with the operator of that keyserver if you
ever have problems with peering.
Once again the snippet above is part of an actual "membership" file with real values
in it, so it's only to provide an example of the format. It's up to you to find other
keyserver operators to agree to peer with you so make sure you use their hostnames
instead and don't just copy the example entries above.
The best approach for finding peers is to ask on the
sks-devel mailing list
with an email something along these lines:
I have a new keyserver running and would like to peer with other
servers. Please add me to your 'membership' file with the following
entry and provide your details in return so I can do the same:
keys.example.com 11370 # Joe Bloggs 0xDEADBEEF
Of course you need to substitute your own details in the email.
Note: Once SKS is running it will automatically pick up changes to the
membership file, so you don't need to restart the daemon if you add or remove
gossip peers. By default SKS re-reads the membership file every 6 hours but you
can override that by putting an entry in your
sksconf file like this:
to change the interval to (for example) 2 hours. My understanding (possibly wrong!)
is that SKS performs a DNS lookup to resolve hosts in the membership file only on
each reload, so having it reload the membership file regularly is a good idea even if
you aren't making frequent changes to it. That way SKS will be able to find any
peers that have changed their IP address.
PKS Email Sync
The PKS network operates differently: it uses email notifications for internal
updates, so to make sure all changes made on your server reach the PKS
network you need to send out sync emails to a cooperative PKS server. SKS servers
will only send sync emails for changes that were initiated locally (otherwise the PKS
network would get a mail-storm of notifications for every change that propagates
through the SKS network!) so having
other SKS servers to gossip with is *not* enough to get changes out to the PKS
network. You must also set up mailsync.
Put something like this in
# Send an email notification to this server when a key is added so that
# all new keys propagate to the PKS network
You'll need to add the (uncommented) email addresses of your PKS peers, but don't
just do this without their permission or they might get upset. You'll probably only
need one PKS peer: I can't actually remember why I ended up with two.
Start DB Server
Finally time to start it up! Amazingly this is the point where I really came
unstuck because nowhere in the README or on the archived wiki does it tell you
what you need to do to start the actual server. I don't know if this is technically
correct, but some poking around in the source led me to this and it seems to work
so please let me know if this is not correct.
The main sks binary wraps a whole bunch of operations including the DB build
stuff we just did. One of the operations is "db", which starts the main db server:
/usr/local/bin/sks db &
So that you can see what's going on, put a tail on the db logfile:
tail -f /sks/db.log
By default your new server will be running on port 11371 (the normal PGP/GPG
server port) so edit your ~/.gnupg/options file on your workstation and
select your keyserver with an entry like:
where 184.108.40.206 is the IP address or hostname of your keyserver.
Then do something like:
gpg --recv-keys 64011A8B
and you should successfully pull down the key, as well as see the action appear
in the DB log. Success! You now have a (partly) functioning keyserver.
Start Reconciliation Server
Once again this is just my interpretation so I may be wrong, but the next step
seems to be starting the recon server that handles synchronising with other
servers (ie: the "gossip" functionality mentioned previously).
/usr/local/bin/sks recon &
Check the operation with a tail on the recon logfile:
tail -f /sks/recon.log
Process Incoming Sync Email
Changes made on the PKS network also need to find their way back to the SKS network
via email. However, it's less critical to handle incoming PKS email because as long
as at least one keyserver on the SKS network receives a PKS update it will be
propagated through the rest of the SKS network via the usual "gossip" mechanism.
There are two common methods for processing incoming PKS updates, both relying on
a script called
sks_add_mail which is distributed as part of SKS.
Most Linux mail transport agents (MTAs) use the
/etc/aliases file (or
similar) as a system-level mapping of virtual email addresses to recipients. You can
use the aliases file to redirect all incoming email for a specific address to the
sks_add_mail script by adding an entry similar to this:
pgp-public-keys: "|/usr/local/bin/sks_add_mail /sks"
That entry will cause any email sent to the address "pgp-public-keys@yourdomain"
to be redirected to the "sks_add_mail" script, which is also given the path to your
SKS directory so that it can find the database to perform the update.
If you already have a user on your server with the correct name and your MTA is
configured to use Procmail, you can simply put a
.procmailrc file in
that user's home directory to have all mail with a subject starting with "incremental"
(which PKS updates do by convention) redirected to the sks_add_mail script. For example:
* ^Subject: incremental
| sks_adde_mail /sks
SKS generates database statistics internally and makes a report available with details
of all peers and the internal stats. Because generating the report puts quite a heavy
load on the server it's only done once each day.
You can access the stats for your server by pointing a browser at "/pks/lookup?op=stats"
on port 11371, similar to this:
Because the stats are generated daily you may find it annoying not to have them
available for a while if you restart the DB daemon. You can get around this by forcing the
daemon to do a full stats run immediately on startup with the
option, like this:
/usr/local/bin/sks db -initial_stat &
Starting up will take longer, but you'll then have a stats report available right away.
If you run a firewall in front of your keyserver (or iptables rules directly on it) you
need to make sure certain ports are open. The critical ports are:
TCP 11370 (for reconciliation with other keyservers)
TCP 11371 (for HTTP connections)
Port 11371 is a well-known registered port, but use of 11370 for reconciliation is
only a convention and is configured in the
membership file on a per-peer
basis. I've never seen a peer use anything except 11370 for reconciliation, but keep in
mind that it is possible and if so you'll need to open the appropriate port through
SKS Network Status
The status of all known hosts in the SKS network is polled every 15 minutes and a
report put online at:
New servers are discovered automatically if they are peering with others that are
already known, so you don't need to request your new server to be added. Just set up
peering and eventually your server will appear in the report.