Re: [PATCH 1/2] mm/huge_memory: skip invalid debugfs file entry for folio split

[Andrew Morton]

On Sun, 28 Apr 2024 21:04:50 -0700 Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:

> On Thu, Apr 25, 2024 at 02:01:26PM -0700, Andrew Morton wrote:
> > On Wed, 24 Apr 2024 15:54:48 -0700 Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
> > 
> > > If the file entry is too long we may easily end up going out of bounds
> > > and crash after strsep() on sscanf(). 
> > > 
> > 
> > Can you explain why?  I'm not seeing it.
> 
> I couldn't see it either but I just looked at the crash below and
> its the only thing I could think of. So I think its when userspace
> somehow abuses MAX_INPUT_BUF_SZ a lot somehow.

This isn't a good basis for making kernel changes :(

Can you investigate a little further please?  What actually is present
at *buf when your new checks succeed?  Could we be seeing 0xNNN,0xNNN
and leaving new_order unaltered?  Or something.