Re: Please consider backporting coredump %F patch to stable kernels

[Christian Brauner <brauner@kernel.org>]

On Mon, Jun 02, 2025 at 11:32:44AM +0200, Greg Kroah-Hartman wrote:
> On Mon, Jun 02, 2025 at 11:09:05AM +0200, Christian Brauner wrote:
> > On Fri, May 30, 2025 at 10:44:16AM +0100, Luca Boccassi wrote:
> > > Dear stable maintainer(s),
> > > 
> > > The following series was merged for 6.16:
> > > 
> > > https://lore.kernel.org/all/20250414-work-coredump-v2-0-685bf231f828@kernel.org/
> > > 
> > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=c57f07b235871c9e5bffaccd458dca2d9a62b164
> > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=95c5f43181fe9c1b5e5a4bd3281c857a5259991f
> > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=b5325b2a270fcaf7b2a9a0f23d422ca8a5a8bdea
> > > 
> > > This allows the userspace coredump handler to get a PIDFD referencing
> > > the crashed process.
> > > 
> > > We have discovered that there are real world exploits that can be used
> > > to trick coredump handling userspace software to act on foreign
> > > processes due to PID reuse attacks:
> > > 
> > > https://security-tracker.debian.org/tracker/CVE-2025-4598
> > > 
> > > We have fixed the worst case scenario, but to really and
> > > comprehensively fix the whole problem we need this new %F option. We
> > > have backported the userspace side to the systemd stable branch. Would
> > > it be possible to backport the above 3 patches to at least the 6.12
> > > series, so that the next Debian stable can be fully covered? The first
> > > two are small bug fixes so it would be good to have them, and the
> > > third one is quite small and unless explicitly configured in the
> > > core_pattern, it will be inert, so risk should be low.
> > 
> > I agree that we should try and backport this if Greg agrees we can do
> > this. v6.15 will be easy to do. Further back might need some custom work
> > though. Let's see what Greg thinks.
> 
> Yes, seems like a good thing to backport to at least 6.12.y if possible.
> 
> Is it just the above 3 commits?

Yes, just those three:

b5325b2a270f ("coredump: hand a pidfd to the usermode coredump helper")
95c5f43181fe ("coredump: fix error handling for replace_fd()")
c57f07b23587 ("pidfs: move O_RDWR into pidfs_alloc_file()")

That should apply cleanly to v6.15 but for the others it requires custom
backports. So here are a couple of trees all based on linux-*.*.y from
the stable repo. You might need to adjust to your stable commit message
format though:

v6.12:
https://github.com/brauner/linux-stable/tree/vfs-6.12.coredump.pidfd

v6.6:
https://github.com/brauner/linux-stable/tree/vfs-6.6.coredump.pidfd

v6.1:
https://github.com/brauner/linux-stable/tree/vfs-6.1.coredump.pidfd

v5.14
https://github.com/brauner/linux-stable/tree/vfs-5.14.coredump.pidfd

v5.10
https://github.com/brauner/linux-stable/tree/vfs-5.10.coredump.pidfd

v5.4
https://github.com/brauner/linux-stable/tree/vfs-5.4.coredump.pidfd