Re: [PATCH] xfs: detect cycles in recovered unlinked inode lists

[Michael Bommarito]

On Wed, May 20, 2026 at 8:19 AM Carlos Maiolino <cem@xxxxxxxxxx> wrote:
>
> Please leave the message you're replying to quoted in your email...

(Sorry about that. Christoph's message was never actually delivered to
my @gmail.com so I had to respond like it was 1999.  @Christoph, not
sure if on my end or yours...)

> The patch ain't completely invalid, but by the time you've somebody
> untrusted with physical access to the system to put a USB stick on,
> your security is already gone.
> Automounting filesystems (a thing we've been fighting against for
> years) just adds on top of it.

I get it.  Better a known and accepted risk than hidden.

> I don't think such tests belongs neither in-tree nor in xfstests, but
> this seems a good idea to start a project related to test filesystems
> resiliency against those issues. syscaller already does a lot of fuzzing
> but mostly random and not containing a bunch of carefully crafted
> filesystems to triggers specific issues.

Yes, and this is where LLMs have been very helpful in particular.
Given an AST or parser / format specification, you can delegate the
mechanics of implementing hundreds of different tests / perturbations.
Many of those strategies are analogous across file systems.  For
example, this XFS finding was found by expanding from my NTFS finding
here:
https://lore.kernel.org/all/20260517234140.1261718-1-michael.bommarito@xxxxxxxxx/

I found a little prior work but nothing that seems active and comprehensive:
https://github.com/sslab-gatech/janus
https://github.com/stevegrubb/fsfuzzer

I'm close to finishing triage and disclosure for all of the stuff I've
found so far, so as soon as that's done, I'll share what I have
publicly and RFC to continue this thread separately.

In the meantime, it sounds like we just park this until we know it
won't break things.