On Tue, Feb 25, 2020 at 02:04:51PM +1030, Alex Murray wrote:
> > On Fri, Feb 21, 2020 at 02:04:37PM -0800, Kees Cook wrote:
> >> On Thu, Feb 20, 2020 at 03:45:39AM +0000, Seth Arnold wrote:
> >> > I'm worried that turning this flag on for the first time in an LTS release
> >> > may be breaking too many expectations.
> >> > Adapting applications may be too much effort; because I don't know what
> >> > exactly apport is doing here it is hard to estimate how much effort it
> >> > will take to adapt it. Possibly the user-launched apport instances need
> >> > to create their own directory on launch. Possibly apport needs a
> >> > more invasive redesign.
> >> > [...]
> >> > Source code searching is not practical. The combination of working
> >> > with files in a world-writable sticky directory and not already using
> >> > O_EXCL with O_CREAT is not feasible to search for.
> >> FWIW, I think that the scope of the change is small enough (only in
> >> world-writable stick directories) and dramatic enough (usually total
> >> failure), that the risk is worth the benefit. Excepting the very few
> >> special directories (like /var/crash, where the software using them
> >> is likely enumerable), I would also argue that breaking stuff in
> >> "standard" temp directories (like /tmp) that isn't using O_EXCL is
> >> actually _desirable_, given that it is expressly risky to operate in
> >> that condition.
> >> And, I would suggest that doing this in an LTS is the right thing to do,
> >> otherwise you wait 2 years before gaining this defense that would be
> >> actively _disabled_ compared to all other distros with a modern version
> >> of systemd. And if this is the first noticed problem, that seems to be a
> >> reasonably good indication of how rare the case is of it creating "real"
> >> problems.
> > As an additional wrinkle, procps in focal-proposed is now setting
> > fs.protected_regular=2 by default, overriding the systemd setting. This
> > hasn't made it out of proposed yet because the additional restriction broke
> > the postgresql-common autopkgtest (fix for this is in progress):
> > https://bugs.launchpad.net/ubuntu/+source/procps/+bug/1864423
> > Kees, it sounds like you were advocating for setting the level to
> > fs.protected_regular=1, but NOT raising it to 2. Should we back out this
> > procps change for 20.04?
> (I'll let Kees respond for himself but figured this was a good point to
> reply)
> From the Ubuntu Security team's view, it is not clear how much breakage
> will result from setting this to 2 - setting protected_regular=1 is
> already a good win and so far has seen little fallout that I know of
> (Apport has already worked around this), but given how close we are to
> FF I am a bit hesitant to mandate for this to be =2 (given we
> already have one known failure as a result). From a security point of
> view, =2 is a better default but this is not the only view that
> matters.
> So my preference would be to keep this at =2 for now but be confident
> that we can back it out to =1 in the event that we suddenly discover a
> bunch more issues in the near term.
Thanks, it's easy enough to back out later (as long as someone actually
raises a flag when things break!), so I'm ok with that.
Since postgresql-common's tests have now been fixed for compatibility with
=2, the procps with this new behavior will be unblocked and should reach the
focal release today or tomorrow.
--
Steve Langasek Give me a lever long enough and a Free OS
Debian Developer to set it on, and I can move the world.
Ubuntu Developer https://www.debian.org/
slangasek@ubuntu.com vorlon@debian.org
