Paperkey - an OpenPGP key archiver

by David Shaw

A reasonable way to achieve a long term backup of OpenPGP (GnuPG, PGP, etc) keys is to print them out on paper. Paper and ink have amazingly long retention qualities - far longer than the magnetic or optical means that are generally used to back up computer data.


For POSIX (Linux, Unix, *BSD, etc):
paperkey-1.3.tar.gz.sig (OpenPGP signature from my key 0x99242560)

Win32 precompiled binary: (OpenPGP signature from my key 0x99242560)

Paper? Seriously?

The goal with paper is not secure storage. There are countless ways to store something securely. A paper backup also isn't a replacement for the usual machine readable (tape, CD-R, DVD-R, etc) backups, but rather as an if-all-else-fails method of restoring a key. Most of the storage media in use today do not have particularly good long-term (measured in years to decades) retention of data. If and when the CD-R and/or tape cassette and/or USB key and/or hard drive the secret key is stored on becomes unusable, the paper copy can be used to restore the secret key.

What paperkey does

Due to metadata and redundancy, OpenPGP secret keys are significantly larger than just the "secret bits". In fact, the secret key contains a complete copy of the public key. Since the public key generally doesn't need to be escrowed (most people have many copies of it on various keyservers, web pages, etc), only extracting the secret parts can be a real advantage.

Paperkey extracts just those secret bytes and prints them. To reconstruct, you re-enter those bytes (whether by hand, OCR, QR code, or the like) and paperkey can use them to transform your existing public key into a secret key.

For example, the regular DSA+Elgamal secret key I just tested comes out to 1281 bytes. The secret parts of that (plus some minor packet structure) come to only 149 bytes. It's a lot easier to re-enter 149 bytes correctly.

Aren't CD-Rs supposed to last a long time?

They're certainly advertised to (I've seen some pretty incredible claims of 100 years or more), but in practice it doesn't really work out that way. The manufacturing of the media, the burn quality, the burner quality, the storage, etc, all have a significant impact on how long an optical disc will last. Some tests show that you're lucky to get 10 years.

For paper, on the other hand, to claim it will last for 100 years is not even vaguely impressive. High-quality paper with good ink regularly lasts many hundreds of years even under less than optimal conditions.

Another bonus is that ink on paper is readable by humans. Not all backup methods will be readable 50 years later, so even if you have the backup, you can't easily buy a drive to read it. I doubt this will happen anytime soon with CD-R as there are just so many of them out there, but the storage industry is littered with old now-dead ways of storing data.


Take the secret key in key.gpg and generate a text file to-be-printed.txt that contains the secret data:
  paperkey --secret-key my-secret-key.gpg --output to-be-printed.txt
Take the secret key data in my-key-text-file.txt and combine it with my-public-key.gpg to reconstruct my-secret-key.gpg:
  paperkey --pubring my-public-key.gpg --secrets my-key-text-file.txt --output my-secret-key.gpg
If --output is not specified, the output goes to stdout. If --secret-key is not specified, the data is read from stdin so you can do things like:
  gpg --export-secret-key my-key | paperkey | lpr
Some other useful options are:
can be "base16" or "raw". "base16" is human-readable, and "raw" is useful if you want to pass the output to another program like a bar code or QR code generator (though note that scannable codes have some of the disadvantages discussed above).
same as --output-type, but for the restore side of things. By default the input type is inferred automatically from the input data.
sets the width of base16 output
allows paperkey to continue when reconstructing even if it detects data corruption in the input.
--verbose (or -v)
be chatty about what is happening. Repeat this multiple times for more verbosity.


Note that paperkey does not change the security requirements of storing a secret key. In fact, paperkey doesn't do any crypto at all, but just saves and restores the original secret key, whether it is encryped or not. So, if your key has a passphrase on it (i.e. is encrypted), the paper copy is similarly encrypted. If your key has no passphrase, neither does the paper copy. Whatever the passphrase (or lack thereof) was on the original secret key will be the same on the reconstructed key. Also note that as with any backup or archiving system, it is prudent to verify you can restore the key from your paper copy before filing the paper away.

Universal Binaries on Apple OS X

You can build a universal ("fat") binary that will work on both PPC and Intel Macs, whether 32 or 64-bit, with the appropriate variation of:
  ./configure CFLAGS="-arch ppc -arch i386" --disable-dependency-tracking
Available architectures are any combination of "i386", "x86_64", "ppc", or "ppc64".

If you are doing the build on a OS X 10.4 (Tiger) PPC machine you may need to add the following to those CFLAGS:

  "-isysroot /Developer/SDKs/MacOSX10.4u.sdk"
The additional isysroot is not necessary on Intel Tiger boxes, or any Leopard (or later) boxes.

Note that there is little point in building a universal binary unless you plan to run the same binary on both a PPC and Intel Mac. Generally this is only useful if you are distributing the binary to other people and want it to work on older Macs.


Paperkey ships with a RPM spec file. You can build the RPM with the usual "rpmbuild -ta /path/to/the/paperkey/tarball.tar.gz".

Paperkey is Copyright © 2007, 2008, 2009, 2012 by David Shaw