GnuPG this Past Summer


As usual, Werner has made a cornucopia of contributions. He improved --quick-addkey and --quick-gen-key, he changed gpg-agent and dirmngr to exit if their sockets disappear, he added an assuan logging monitor, he implemented new export and import filters, he did some work on g13, he added /run/user/UID/gnupg sockets, he introduced an option (--recipient-file) to work directly with keys stored in a file, and he made a number of improvements to GPGME including adding TOFU support.

The filtering changes allow controlling what packets are imported or exported. For instance, if you want to only keep a single user id when exporting a key, you could use:

gpg --no-options --import-options import-export       \
    --import-filter keep-uid='mbox = joe at' \
    --import  < >

More information about this feature is available in his note to the GnuPG mailing list or gpg's documentation.

The --recipient-file option is an oft-requested feature, which allows working with keys without importing them.

Werner also fixed a critical bug in the way the mixer in the random number generator stirred the pool. Specifically, the bug allowed an attacker who obtains 580 bytes from the standard random number generator (RNG) to trivially predict the next 20 bytes of output. Fortuitously, this bug does not affect the default generation of keys (more details).

Justus continued to improve our new test suite for GnuPG. The improvements included not only fixes to the new scheme-based driver, but also a bunch of new tests. A couple of the changes included bug fixes to TinySCHEME. Unfortunately, the upstream developers don't appear to be interested in the fixes.

Most of Justus' time recently has been focused not on the test suite, but on improving the Python bindings for GPGME. This work was started by Ben McGinnes, who contributed an initial port of the PyME bindings to Python 3. Justus finished this port, restored Python 2 compatibility, and added more pythonic interfaces (e.g., making everything work with objects implementing the buffer protocol like byte strings). The low-level interface has, however, been retained and existing applications should continue to work (if not, this is a bug, please file a bug report). He also ported the GPGME test suite to the Python bindings. This uncovered a number of latent bugs in the bindings, which he fixed. From our perspective, these are now the official Python bindings for GPGME: we've added them to the GPGME repository, and we will continue to maintain them in the foreseeable future. Nevertheless, to be more compatible with Python developers' work flow, we are also packaging pyme3 for pypi, which means that the bindings can be installed using pip install pyme3. More information is available in Justus' blog post.

Justus also set up a Jenkins host for continuous integration. In addition to running make check for each commit under several configurations, it also runs the checks with various sanitizers enabled. This has already prevented a number of minor bugs from making it into releases.

Andre has made a number of end-user facing contributions. The most notable is for users of Kleopatra, which now has new dialogs for File Encryption and Decryption / Verification. These greatly reduce the number of required interactions to perform these operations. He also worked on the new file type registration on Windows so that decrypting a file only requires a double click. Additionally, he has continued his work on the GnuPG plugin for Outlook, which should be released with gpg3win-3 this fall. The code is already in good form, and testers are encouraged to check it out together with the new Kleopatra (see Test version of Gpg4win-3.) Andre has also been working on improving KMail's gpg support. One of the focuses of this work has been adding TOFU support to the libraries used by KMail. Andre also merged the C++ and Qt bindings for GPGME from KDE into the official GPGME repository. This included a port of the C++ API to pure standard C++ without boost, and the removal of some KDE-Framework use in the Qt bindings so that the bindings now only require Qt 5 base. This should make working with gpg in a Qt application even more convenient. In particular, executing operations asynchronously is very easy. Finally, Andre fixed some CRL-related bugs in dirmngr.

Kai's recent work has focused on porting Mailpile to use GPGME rather than its own wrapper, which only works with GnuPG 1.4. Unfortunately, many projects decide to take a similar approach to Mailpile, and write their own code to interact with gpg. As a reminder, we strongly encourage all developers to not directly interact with gpg, but to use GPGME, which is not only more complete, but also has seen a lot of testing. We realize that GPGME's interface's are not always ideal, however, we are open to suggestions for improvements, and feature requests. Similarly, if you don't understand how to do what you want using GPGME, we encourage you to ask for help on the gnupg-devel mailing list.

Jussi Kivilinna has continued his work optimizing libgcrypt. In the recent past, most of his effort was spent on implementing assembly versions of various cryptographic functions for the ARMv8/AArch32 architecture.

Niibe worked on mitigating the recently published Flip Feng Shui exploit. Flip Feng Shui uses a cross-VM, row hammer-based exploit to change the trusted.gpg file, which is used by Debian's package manager apt to verify downloads, and apt's sources.list file, which determines where packages are downloaded from, in a controlled manner. This allows attackers to replace packages that are installed with their own versions. The fix is to make sure that gpgv always checks that self-signatures are valid.

Niibe also spent time improving GnuPG's smartcard support. This has primarily consisted of many small, but important improvements including smartcard support for ECC keys and various bug fixes. Further, Niibe investigated adding signature verification for ssh keys stored in the authorizedkeys file. This would allow detecting corrupted keys, which could happen via a Flip Feng Shui-type attack. Although there is some support for signature verification in ssh, Niibe discovered that this particular mode of operation is not yet supported by ssh-agent.

Finally, Niibe has released a new version of GnuK (1.2.1). GnuK is a fully free cryptographic token (hardware and software). Not only is GnuK based on free software, but the entire hardware specification is open, and the parts are relatively easy to buy and assemble. The GnuK token can be ordered from seeed or the FSF.

As usual, dkg contributed various clean ups and bug fixes. He contributed a patch to avoid publishing the GnuPG version by default, and another to improve --quick-revuid. He also provided a patch to reenable exporting secret keys without a passphrase, which was possible in gpg 1.4 and 2.0, but, due to various technicalities, was not possible in 2.1. dkg also started a discussion about having systemd manage gpg's daemons. This would ensure that GnuPG's daemons are stopped when the user logs out. He provided patches, but so far these changes have not yet been accepted.

Ben Kibbey made a number of contributions. Among his bug fixes and clean ups, he fixed the OpenIndiana (Solaris) builds.

I (Neal) returned from a several month sabbatical. My first order of business was to tie up some loose ends with the TOFU support in GnuPG. Among other things, I added several checks to reduce the number of gratuitous conflicts. In particular, if two keys have the same email address and are cross signed, then they are almost certainly controlled by the same person. In fact, this is a usual way of indicating key rotation. I also set the default policy to "good" for keys that the user has directly signed.


There have been several GnuPG releases since the last status update: 2.1.13, 2.1.14, 2.1.15, and 1.4.21; and two releases of libgcrypt 1.7.1 and 1.7.2. Finally, a new version of GPGME is available, 1.7.0, which includes the newly upstreamed Python, C++ and Qt bindings as well as a number of bugs fixes and various improvements.

Public Appearances

Werner held a keynote at GUADEC, "We Want More Centralization, Do We?." His talk was covered by LWN.

In May, Neal held his "An Advanced Introduction to GnuPG" talk at INRIA, and again at GHM in August. Neal will hold the same talk on October 3rd at 18:00 at the ACM chapter at Johns Hopkins University in Baltimore, and again on October 5th at 18:30 at the NYLUG (you need to RSVP for this event).

In August, we took part in the GUUG-hosted OpenPGP.conf. I've already posted a report to our blog.

Note: We are looking to interview representatives from organizations who rely on GnuPG, e.g., journalists, lawyers, NGOs, governmental organizations, software distributors, companies, etc., for some publicity material that we are producing. If you fall into this category, or know someone who does, and would be willing to be interviewed, please get in touch with me!