sgrubb at redhat.com
Mon Aug 15 23:42:51 UTC 2011
On Thursday, August 04, 2011 07:53:33 AM Kamil Paral wrote:
> > > Honestly, I don't know. On the one hand, I have some scripts that
> > > are good for fedora
> > > QE in general. For example, the shell error test...why would anyone
> > > purposely write
> > > shell script that does not work? This can always be fixed before a
> > > release. Some tests
> > > are still under development like the ELF binary well known tmp file
> > > test. This can make
> > > some false positives, but there are enough good things in it to
> > > start asking real
> > > questions about packages...like.../home/cagney/tmp/a.out...why is
> > > that in any program?
> > > But the chroot tests are solid. As are the exec stack tests. So, yes
> > > there are things
> > > that can be automated so problems are not shipped.
> > Awesome. CCing autoqa-devel on the reply, then; is anyone from AutoQA
> > willing to work with Steve to take a look at his tests and identify
> > good
> > candidates for bringing into AutoQA? Thanks!
> I'll reply by describing the current state of AutoQA. I'm afraid we're not
> there yet.
> I don't know how Steve's tests behave, but:
> 1. We can't run destructive tests (uninstalling packages, deleting system
> files, stopping services).
All are read only. I have several more that do stuff like start each servoce and scan
the audit logs for AVCS, then stop all services. But this is disruptive and require
running apps. So, I have saved some of those for a later release.
> 2. We can help Steve create the AutoQA wrappers
> for those tests, but we can't maintain the very tests themselves,
> obviously. He has to do that.
If you look at the test source, you can pull them in and maintain them yourself. Most
are very simple ideas.
> 3. Unfortunately we don't have an
> infrastructure for third-party test maintainers. Currently the tests have
> to be in our git, that means he has to send any changes in patches. We
> deploy new version only once in several weeks at best. 4. I suppose these
> tests would run after each koji build. The only way of reporting results
> right now is to send emails for those maintainers that opted-in for this,
> nothing else.
> That said, we would love to execute more tests for Fedora. But until the
> proper support is ready, it takes quite some effort. The first approach is
> go through the tests, select some appropriate ones and do that now. The
> second approach is wait some time until we are ready and then Steve can
> maintain these tests independently and we just execute them. We will of
> course create a ticket about that and follow on it when that time comes.
How do we pull some of these in? I am interested in making sure we do not regress on
the executable stack portion at a minimum. Between NX, FORTIFY_SOURCE, and stack-
protector, we have pretty much won the stack overflow/inject shell code battle
(excluding ROP possibilities) and I would hate to see that make a return.
The find-hidden-exec test is also something people should run, but maybe during testing
of a release after using it. It found some interesting problems over the years.
Usually due to file creation with very bad permissions.
The find_chroot test could run each build. It recently found a security problem in
libcap. We really don't want apps that can escape the intended chroot. There is also a
python equivalent. It finds lots of problems. But I think that part of the problem is
python devs think python takes care of things, or maybe they don't know how to move
all modules in the chroot and changing into it makes the program crash. Either way,
there are some problems in python that need addressing.
The find-sh4errors script can run each build. The idea is that bash has a mode, -n,
that it simply parses the script and does not execute it. This finds many problems. And
we really don't want to ship scripts that bash won't parse.
The other scripts are harder to say they should be automated and run for each build.
But I guarantee that if you want to find bugs in the distribution, start trying some of
these scripts. There are lots of problems - enough glory to go around for everyone. :)
More information about the test