Re: [PATCH v2 1/3] userfaultfd: ensure mremap_userfaultfd_fail() releases mmap_changing

[Mike Rapoport]

On Mon, May 11, 2026 at 11:15:34AM +0200, David Hildenbrand (Arm) wrote:
> On 5/1/26 16:54, Mike Rapoport wrote:
> > From: "Mike Rapoport (Microsoft)" <rppt@xxxxxxxxxx>
> > 
> > Sashiko says:
> > 
> >   mremap_userfaultfd_prep() increments ctx->mmap_changing to stall
> >   concurrent operations, but mremap_userfaultfd_fail() does not
> >   decrement it before dropping the context reference.
> > 
> > If an mremap operation fails, ctx->mmap_changing remains elevated. This
> > will causes subsequent userfaultfd operations like a UFFDIO_COPY to fail
> > with -EAGAIN.
> > 
> 
> Sounds like we should CC stable?

Yes.
 
> > Decrement ctx->mmap_changing in mremap_userfaultfd_fail().
> > 
> > Link: https://sashiko.dev/#/patchset/20260430113512.115938-1-rppt@xxxxxxxxxx
> > Fixes: df2cc96e7701 ("userfaultfd: prevent non-cooperative events vs mcopy_atomic races")
> > Signed-off-by: Mike Rapoport (Microsoft) <rppt@xxxxxxxxxx>
> > ---
> >  fs/userfaultfd.c | 1 +
> >  1 file changed, 1 insertion(+)
> > 
> > diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> > index 4b53dc4a3266..ef963a58f1a1 100644
> > --- a/fs/userfaultfd.c
> > +++ b/fs/userfaultfd.c
> > @@ -786,6 +786,7 @@ void mremap_userfaultfd_fail(struct vm_userfaultfd_ctx *vm_ctx)
> >  	if (!ctx)
> >  		return;
> >  
> > +	atomic_dec(&ctx->mmap_changing);
> 
> I'll note that other users have a
> 
> 	VM_WARN_ON_ONCE(atomic_read(&ctx->mmap_changing) < 0);
> 
> In there. Likely we should do the same?

Yeah, we could.
 
> -- 
> Cheers,
> 
> David

-- 
Sincerely yours,
Mike.