Summary: splitting your key into multiple pieces and do something different with each piece is a not-too-shabby way to secure your GPG private key.
Disclaimer: I am not a security expert, and there is no such thing as “perfect security”, thus the witty title of my blog post. You are encouraged to try out what I am describing here, but I am not responsible for any misuse or data loss occurring while trying out the steps below.
I am guilty of discovering the power and versatility of GPG keys only in my late 30s (I know, I know). It happened by pure chance as it was a required step for my previous job. It’s a really nice way to bind and verify digital identities to some random bits sitting in your computer. Plus, it opens up new exciting possibilities, like securing all your passwords with something like pass. However, this has the (expected) consequence that now your GPG key becomes a single point of failure and if it gets lost or, worse, stolen, it can be quite a disaster.
To mitigate this, saving it somewhere safe is paramount. This begs the question: “how?”. I usually have (enough but not too much) trust in cloud providers like Dropbox, Google Drive & co, but ultimately you are sending your precious data over the network (hoping the connection is secure and the traffic not spoofed) and if one of these services gets compromised, the integrity and security of your data also is.
To mitigate this, I have been researching ways of securely storing my private GPG key, and I think what I came up with is decent (at least for my use case).
My first attempt was - as suggested around - to use something like
qrencode to turn my GPG key into a QR code I could then print and store in a closet. Then I could use something like zbarimg to retrive my key. However, this approach releaved to be not practical as my key is too big and it doesn’t fit into a QR code.
I ended up splitting my key into 4 parts, and I have then encoded as a QR code only the first part; the other three have been encrypted with a symmetric key (i.e.
gpg -c your-file). By doing that, I can store the encrypted parts inside Dropbox, and print the first part and store it physically in a safe place.
That’s a fair question; in principle, simply encrypting my GPG key with a symmetric key and upload it to Dropbox might have been enough, but I feel like this schema opens up more possibilities. I especially like the fact I am not dependant on just one cloud provider and that, even if Dropbox is ever compromised, and even if by some quantum-computer-miracle my symmetric key is bruteforced, the attacker would still miss the last piece of the puzzle (no pun intended).
Last but not least, here I have only used a single symmetric key for all the three parts of my key, but nobody is preventing me from using two or even three different ones. This way, even if one symmetric key is leaked somehow, the attacker couldn’t still have access to my full GPG key (provided he could get his hands on the QR code, of course).
I have two scripts that I have whipped up: the first one exports my GPG key and split it into chunks:
#!/usr/bin/env bash # First export your gpg key like so: # # gpg --export-secret-keys -a -o mykey.asc # # Then this script will generate 4 qr codes for your key. At this point # it's up to you what to do with these images. split -b 2800 $1 mykey- for file in mykey-??; do <"$file" qrencode -o "$file".png done
Running this script giving as input the filename of your exported key would first split it into multiple parts and finally for each of them run
qrencode. At this point you would be left with an
.asc file (which you might want to delete now) and a bunch of files like, for example:
mykey-aa mykey-aa.png mykey-ab mykey-ab.png ...
At this point you can delete all the
.png but the first one, and conversely delete the first
mykey-aa keeping the other around. Now, for each of the “chunks” (except the one and only
.png, of course) proceed to the gpg encryption with a symmetric key of your choice. For example:
gpg -c mykey-ab
This will spawn pinetry and allow you to insert the password. At the end of this operation, you should have multiple
.gpg files: these are the ones you can store on Dropbox. Your files should be looking somewhat like this:
mykey-aa.png mykey-ab.gpg mykey-ac.gpg mykey-ad.gpg
Once the time will come to reconstruct your key, you can use this other script:
#!/usr/bin/env bash RESULT=mykey.asc zbarimg --raw $1-aa.png | perl -pe 'chomp if eof' - > $RESULT for f in $1-ab.gpg $1-ac.gpg $1-ad.gpg; do echo $f; gpg -dq $f >> $RESULT; done # Sanity check gpg --dearmor $RESULT >/dev/null
You have to supply to this script the
basename of your files (in our example would be
mykey) and the script would first read the QR code, then decrypt the other chunks (in my case they are just 3 and are hardcoded in the script, your mileage might vary) and dulcis in fundo it will perform a sanity check on the reconstructed key.
Voilà, your key has been reconstructed! You can now import the key again inside your GPG keychain by doing
gpg --import <key>.
I have tested this approach myself when migrating the key from a laptop to another and it worked just fine, so hopefully this will be useful to you as well.
My friend Edsko pointed out a few quirks in my original scheme. In particular, assuming the possibility that Dropbox can actually be violated and that the last piece of the puzzle standing between the full key is the QR code, using the first chunk as our unencrypted QR code is a poor choice. First of all, we are still subject to bruteforce attacks as now all it takes for the attacker to get knowledge of the key is to bruteforce 1/4 of it. Not only that, but crucially the first part of the key includes a few bytes for the header (the
-----BEGIN PGPG PRIVATE KEY BLOCK-----) which means there are actually even fewer bytes to bruteforce.
A better scheme here would be to:
This way, we are not only eliminating the problem of the header, but also never storing something unencrypted. The obvious consequence is that now we need an extra step to recover our key.
Somebody also suggested to use something like paperbak, which is ultimately possible but I was put off by what seemed to be quite a careful procedure for restoring the data from the printed bitmap. In particular the readme explains how is important to have a good printer etc etc, so I didn’t want to take the risk.