Re: [PATCH 0/4] vfs: if RESOLVE_NO_XDEV passed to openat2, don't *trigger* automounts

[Christian Brauner <brauner@kernel.org>]

On Mon, Aug 25, 2025 at 04:46:34PM +0400, Askar Safin wrote:
> 
>  ---- On Tue, 19 Aug 2025 12:21:33 +0400  Christian Brauner <brauner@kernel.org> wrote --- 
>  > On Mon, Aug 18, 2025 at 03:31:27PM +1000, Aleksa Sarai wrote:
>  > > I would merge the first three patches -- adding and removing code like
>  > Agreed.
> 
> May I still not merge these patches?
> 
> All they may (hypothetically) fail on their own.
> 
> If they do, then it will be valuable to know from bisection which of them failed.
> 
> Let's discuss them one-by-one.
> 
> The first patch moves checks from handle_mounts to traverse_mounts.
> But handle_mounts is not the only caller of traverse_mounts.
> traverse_mounts is also called by follow_down.
> I. e. theoretically follow_down-related code paths can lead to problems.
> I just checked all them, none of them set LOOKUP_NO_XDEV.
> So, they should not lead to problems. But in kernel we, of course, never
> can be sure. They should not lead to problems, but still can.
> 
> The second patch removes LOOKUP_NO_XDEV check.
> This is okay, because if "jumped" is set and "LOOKUP_NO_XDEV" is set, then
> this means that we already set error, and thus ND_JUMPED should
> not be read, because it is not read in error path. But this is not obvious, and
> so Al asked me add comment (
> https://lore.kernel.org/linux-fsdevel/20250817180057.GA222315@ZenIV/
> ), and, of course, I will add it in the second version in any case.
> So, ND_JUMPED should not be checked in error path, and thus this should
> not lead to problems. But still can.
> 
> The third patch makes traverse_mounts fail
> immidiately after first mount-crossing
> (if LOOKUP_NO_XDEV is set). As opposed to very end.
> This should not lead to problems. But can.
> 
> So, again, any of these 3 patches can (hypothetically)
> lead to its own problems.

You can send them separately if you like but I'll still reserve the
right to squash them when applying. I don't see the value in these
minimal changes yet and the regression potential is completely
theoretical so far.