Trustable "guix pull"
Christopher 11 years ago Leo 11 years ago Christopher 11 years ago Ludovic 11 years ago Ludovic 11 years ago Leo 11 years ago Thompson, 11 years ago Ludovic 11 years ago Mike 11 years ago fluxboks 11 years ago Thompson, 11 years ago Ludovic 11 years ago Ludovic 10 years ago Ludovic 10 years ago Leo 10 years ago Mike 10 years ago Ludovic 10 years ago Ludovic 10 years ago Werner 10 years ago Mike 10 years ago ng0 10 years ago Ludovic 10 years ago Mike 10 years ago Werner 10 years ago Christopher 10 years ago Leo 10 years ago Mike 10 years ago Ludovic 10 years ago ng0 10 years ago Leo 10 years ago Ludovic 10 years ago Werner 10 years ago ng0 10 years ago Ludovic 10 years ago Thompson, 10 years ago Ludovic 10 years ago Ludovic 10 years ago Bob 10 years ago Bob 10 years ago Ludovic 10 years ago Ludovic 10 years ago Mike 10 years ago Ludovic 10 years ago Ludovic 9 years ago Vagrant 8 years ago Ludovic 8 years ago Vagrant 8 years ago Ludovic 8 years ago Ludovic 7 years ago zimoun 7 years ago Ludovic 7 years ago Ricardo 7 years ago Ricardo 7 years ago Ludovic 7 years ago Ricardo 7 years ago Ludovic 7 years ago Vagrant 7 years ago Efraim 7 years ago Ludovic 7 years ago Jakub 7 years ago Ludovic 7 years ago Ludovic 7 years ago Justus 7 years ago Ludovic 7 years ago Ludovic 7 years ago Ludovic 7 years ago Ludovic 7 years ago Ludovic 7 years ago Ludovic 7 years ago Ludovic 6 years ago John 6 years ago zimoun 6 years ago Ludovic 6 years ago Ludovic 6 years ago zimoun 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Ludovic 6 years ago Trustable "guix pull" Done Details 16 participants Ludovic Courtès Christopher Allan Webber Thompson, David Efraim Flashner fluxboks John Soo Justus Winter Jakub Kądziołka Leo Famulari Ludovic Courtès Mike Gerwitz ng0 Ricardo Wurmus Vagrant Cascadian Werner Koch zimoun Owner unassigned Submitted by Christopher Allan Webber Severity serious Debbugs page Right now, when a user does a "guix pull", that pulls down the latest repository of code from git, which is kept in a tarball.
Once you receive the latest code, this has some checks: what's the hash of each package, etc. Unfortunately, it's delivered over http: (define %snapshot-url ;; " http://hydra.gnu.org/job/guix/master/tarball/latest/download" " http://git.savannah.gnu.org/cgit/guix.git/snapshot/master.tar.gz" ) At minimum we should deliver this over HTTPS, ideally with a single certificate that is trusted by the user, so the user can't be easily MITM'ed.
On top of that, even if you run from git proper what there isn't a test about is: can you trust those latest commits? Git doesn't really check, at least by default. https://mikegerwitz.com/papers/git-horror-story How about this: anyone with commit access should use "signed off by" and gpg signatures combined. We should keep some list of guix committers' gpg keys. No commit should be pushed to guix without a gpg signature.
At this point, at least, there is some possibility of auditing things. Perhaps before a master.tar.gz is made, there can be some integrity check of the commits matching the current set of "trusted" keys? On Wed, Mar 02, 2016 at 10:03:59AM -0800, Christopher Allan Webber wrote: Toggle quote (5 lines) > Right now, when a user does a "guix pull", that pulls down the latest > repository of code from git, which is kept in a tarball.
Once you > receive the latest code, this has some checks: what's the hash of each > package, etc. A discussion worth having. But, let's merge this bug into debbugs.gnu.org/22629. Also, we should read "The Update Framework" as requested there. Toggle quote (28 lines) > > Unfortunately, it's delivered over http: > > (define %snapshot-url > ;; "http://hydra.gnu.org/job/guix/master/tarball/latest/download" > "http://git.savannah.gnu.org/cgit/guix.git/snapshot/master.tar.gz" > ) > > At minimum we should deliver this over HTTPS, ideally with a single > certificate that is trusted by the user, so the user can't be easily > MITM'ed. > > On top of that, even if you run from git proper what there isn't a test > about is: can you trust those latest commits?
Git doesn't really check, > at least by default. > > https://mikegerwitz.com/papers/git-horror-story > > How about this: anyone with commit access should use "signed off by" and > gpg signatures combined. We should keep some list of guix committers' > gpg keys. No commit should be pushed to guix without a gpg signature. > At this point, at least, there is some possibility of auditing things. > > Perhaps before a master.tar.gz is made, there can be some integrity > check of the commits matching the current set of "trusted" keys? > > > Leo Famulari writes: Toggle quote (9 lines) > On Wed, Mar 02, 2016 at 10:03:59AM -0800, Christopher Allan Webber wrote: >> Right now, when a user does a "guix pull", that pulls down the latest >> repository of code from git, which is kept in a tarball.
Once you >> receive the latest code, this has some checks: what's the hash of each >> package, etc. > > A discussion worth having. But, let's merge this bug into > debbugs.gnu.org/22629. I'm not sure they should be merged, though they're related. That thread doesn't deal at all with security, though it provides some other good ideas. It even says: PS: I do not mention the issue of authenticating code here, which is obviously very important and deserves to be treated separately.
However I have no objections to merging them if others think we should Toggle quote (2 lines) > Also, we should read "The Update Framework" as requested there. This? https://theupdateframework.github.io/ There seem to be quite a few papers there! severity 22883 serious Hello! Christopher Allan Webber <cwebber@dustycloud.org> skribis: Toggle quote (11 lines) > On top of that, even if you run from git proper what there isn't a test > about is: can you trust those latest commits?
Git doesn't really check, > at least by default. > > https://mikegerwitz.com/papers/git-horror-story > > How about this: anyone with commit access should use "signed off by" and > gpg signatures combined. We should keep some list of guix committers' > gpg keys. No commit should be pushed to guix without a gpg signature. > At this point, at least, there is some possibility of auditing things. To make progress on this front, I’ve decided to start signing all my commits, so: Toggle snippet (6 lines) $ git config commit.gpgsign true $ git config --global user.signingkey 090B11993D9AEBB5 I invite everyone to do the same.
Hopefully, within a few weeks, we can add a commit hook to reject unsigned commits. Note that we’ll be signing patches we push on behalf of contributors who do not have commit access (reviewer’s responsibility). Also, rebasing, amending, and cherry-picking code signed by someone else would lose the original signature, which isn’t great and should be avoided, if possible. What remains to be seen, among other things, is how we’ll maintain a keyring of the committers, and how we’ll distribute it to users of ‘guix pull’; the TUF spec has clever ideas about it, but we need to see how they map to our setup.
Thoughts? Ludo’. -----BEGIN PGP SIGNATURE----- Version: GnuPG v2 iQIcBAEBCAAGBQJXHplLAAoJEAkLEZk9muu1UcEP/RykKMk+zQP/f1Wm4/TRGlje IRTV9HXlLk8jPosxBGARfuKPyN+9lLgzU9kd7xlO/usgEZhSk9dyFY8JgpbAvWUH G5+Z2ZlgDwQJbjNBz4qLvqpXVeFyh1Wys4zd80k+CA4dbC/OgumaJf76f3B1GRAa gd1+IWQEcaLzKy+As2A0cRs5GWjsnBR5Sss7GSnFHx0jP2JIm5Z1n7JJ+aj4r9iF DDgrQn01/gT04bVbnQ7UO4Oze4L4i1YoVagO5jj5KEMff/YmP53hiD5jeIEuTzer /mFDGzwfgHdyGIPR3/dXL50in+Wcml52ig0oIacRYewBkF5cjT3p/VOx88lRhZ/S Yo2Rp6NgBCP+F/yzCU2OgZQfpw2/Y01wz7/ChvhwtmYYR/aCbfrDuxO9yYUI2Ffg RF9ai6iefzJFNRU9Ld97ksUL94JOgJEOoBigNs79hC+iNJ9ap9a958dkYbYL1N+W lcDZx7YuBcufmOIKxmvzzSNMmhXi9gsb+fPhRP7LYozze4fNVYF1OSZ/qF3PCjYv S8cnKr16L
The rest of this page is on the live site.