Re: [PATCH] mm/filemap: fix NULL pointer dereference in do_read_cache_folio()

[Darrick J. Wong]

On Tue, Nov 18, 2025 at 03:37:09PM +0000, Matthew Wilcox wrote:
> On Tue, Nov 18, 2025 at 05:03:24AM -0800, Christoph Hellwig wrote:
> > On Mon, Nov 17, 2025 at 10:45:31AM -0800, Andrii Nakryiko wrote:
> > > As I replied on another email, ideally we'd have some low-level file
> > > reading interface where we wouldn't have to know about secretmem, or
> > > XFS+DAX, or whatever other unusual combination of conditions where
> > > exposed internal APIs like filemap_get_folio() + read_cache_folio()
> > > can crash.
> > 
> > The problem is that you did something totally insane and it kinda works
> > most of the time.
> 
> ... on 64-bit systems.  The HIGHMEM handling is screwed up too.
> 
> > But bpf or any other file system consumer has
> > absolutely not business poking into the page cache to start with.
> 
> Agreed.
> 
> > And I'm really pissed off that you wrote and merged this code without
> > ever bothering to talk to a FS or MM person who have immediately told
> > you so.  Let's just rip out this buildid junk for now and restart
> > because the problem isn't actually that easy.
> 
> Oh, they did talk to fs & mm people originally and were told NO, so they
> sneaked it in through the BPF tree.
> 
> https://lore.kernel.org/all/20230316170149.4106586-1-jolsa@xxxxxxxxxx/
> 
> > > The only real limitation is that we'd like to be able to control
> > > whether we are ok sleeping or not, as this code can be called from
> > > pretty much anywhere BPF might run, which includes NMI context.
> > > 
> > > Would this kiocb_read() approach work under those circumstances?
> > 
> > No.  IOCB_NOWAIT is just a hint to avoid blocking function calls.
> > It is not guarantee and a guarantee is basically impossible.
> 
> I'm not sure I'd go that far -- I think we're pretty good about not
> sleeping when IOCB_NOWAIT is specified and any remaining places can
> be fixed up.
> 
> But I am inclined to rip out the buildid code, just because the
> authors have been so rude.

Which fstest actually checks the functionality of the buildid code?
I don't find any, which means none of the fs people have a good signal
for breakage in this, um, novel file I/O path.

--D