[Savannah-hackers] Savannah hackers update

[Elfyn McBratney]

Hello,

This is a merge of a report on the back-end problems we've been having at
Savannah and our, now regular, weekly/when we have time status update. :)

Elfyn

---

On Monday (13th), while investigating why a seemingly-active project had
no administrator, Elfyn discovered that a number of users (in /etc/passwd)
and groups (in /etc/group) didn't exist, but were either owner's (here,
owner means the person who registered the project on Savannah) or members
of a project.

This caused error's in a project's chroot /etc/passwd and /etc/group
(no gr_gid, all empty pw_ fields, etc) and broke developer CVS access for
a lot of projects.

Each user/project that was affected by this was fixed by hand and the
chroot's were updated.

On Tuesday, Elfyn made vast changes to Savannah's back-end scripts.  The
reworking was necessary to ensure the stability of Savannah, since the
old scripts were generating errors and were poorly hacked/kludged (and
needless to say, hard to maintain, too).

At the same time, and in effort to make it easier for us to manage
cronjobs sanely, Elfyn also added the back-end code to administration's
CVS repo.

The scripts that were rewritten were, sv_users and sv_groups.  Both went
active on Savannah on Tuesday night.  The rewritten scripts had a few bugs
(which is to be expected) of their own, and when run from the cronjob,
after Elfyn retired for the night, denied access to a large group of users
on Savannah (for developer CVS & webcvs).

Michael, then disabled the savannah-scripts cronjob, to prevent more files
from being tampered with.  He restored access manually for users, as they
submitted support requests, for most of Tuesday night and the following
day.

The next day (Wednesday 15th), Elfyn fixed Savannah's scripts and they went
active, once again.  This time they were working correctly, with only a few
minor details needing attention, some of which were unseen and are being
handled on a case by case basis via Savannah's support tracker and mailing
lists.

As far as we are aware, the back-end if now functioning normally, but we will
no doubt be spending some time in the coming weeks and months improving the
code.

We also kept the Savannah front-page updated with status on the problems
that were going on between the 13th and the 15th.


Lastly, our status update:

 - Developing a patch to viewcvs to catch the rlog error's that have been
 happening recently for the gcc and automake projects.  (Elfyn)

 - Imported the Savannah back-end cronjob (sv_users & sv_groups run) into
 administration's CVS repo.  (Elfyn)

 - Added error handling/recovery code to every [perl] system() call to chmod,
 chown, mv, groupadd, useradd and usermod.  (Elfyn)

 - Added advisory locking code (flock) wrapping around chroot /etc/passwd and
 /etc/group creation and ~gatekpr/project/email file creation.  (Elfyn)

 - Fixed Savane backend to not fork and exec gpg if no keys existed in
 the database for a given user.  Before, gpg would be run for all
 ~30,000 users, which would slow down the system a lot.  Now, gpg only
 runs for user's that have uploaded a key.  (Elfyn)

 - Delete all projects from the pending queue, that have not responded to our
 ``Do you still want your project'' emails.  There are now 50 projects left in
  the queue.  (Sylvain)

 - Patch a remote code execution vulnerability in Savane that could allow
 code execution with the privileges of the web server (www-data).  (Sylvain)

--
Elfyn McBratney
beu on irc.freenode.net/savannah.[non]gnu.org

PGP Key ID: 0x456548B4
PGP Key Fingerprint:
  29D5 91BB 8748 7CC9 650F 31FE 6888 0C2A 4565 48B4

pgpRoRS14B8hs.pgp
Description: PGP signature