TeX Live security improvements
Today I committed a set of changes to the TeX Live subversion repository that should pave the way for better security handling in the future. Work is underway to use strong cryptographic signatures to verify that packages downloaded and installed into a TeX Live installation have not been tinkered with.
While there is still a long way to go and to figure out, the current changes already improve the situation considerably.
Status up to now
Although we did ship size and checksum information within the TeX Live database, these information were only considered by the installer when re-starting an installation to make sure that the downloaded packages are the ones we should use.
Neither the installer nor tlmgr did use the checksum to verify that the downloaded packages is correct, relying mostly on the fact that the packages are xz-compressed and would create rubbish when there is a transfer error.
Although none of us believes that there is a serious interest in tinkering with the TeX Live distribution – maybe to steal just another boring scientific paper? – the door was still open.
The changes committed to the repository today, which will be in a testing phase to get rid of bugs, consists of the following:
- unification of installation routines: it didn’t make sense to have duplication of the actual download and unpack code in the installer and in tlmgr, so the code base was simplified and unified, and both the installer and tlmgr now use the same code paths to obtain, unpack, and install a package.
- verification of size and checksum data: hand-in-hand with the above change, verification of downloaded packages based on both the size as well as the checksum is now carried out.
Together these two changes will allow install-tl/tlmgr to verify that a package (.tar.xz) is according to the information in the accompanying texlive.tlpdb. This still leaves the option of tampering with a package and updating the texlive.tlpdb for the fixes checksums/sizes.
For the future we do plan mostly two things:
- switch to stronger hashing algorithm: till now we use md5, but we will switch to sha512 instead.
- GnuPG signing of the checksum file of the texlive.tlpdb, that is detached signature of texlive.tlpdb.checksum
The last step above will give a very high level of security, as it will be not practically possible to alter the information in the texlive.tlpdb, thus no tampering with the checksum information of the containers, and in turn no tampering with the actual containers will be possible.
Due to the wide range of supported architectures and operating systems, we will not make verification obligatory, but if a gpg binary is found, it will be used to verify the downloaded texlive.tlpdb.checksum.
Details have to be hammered out and the actual programming has to be done, but we are on the way.