On Mon, Apr 15, 2024 at 3:15 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
>
> On Mon, Apr 15, 2024 at 03:05:44PM -0700, Vishal Moola wrote:
> > Commit 9acad7ba3e25 ("hugetlb: use vmf_anon_prepare() instead of
> > anon_vma_prepare()") may bailout after allocating a folio if we do not
> > hold the mmap lock. When this occurs, vmf_anon_prepare() will release the
> > vma lock. Hugetlb then attempts to call restore_reserve_on_error(),
> > which depends on the vma lock being held.
> >
> > We can move vmf_anon_prepare() prior to the folio allocation in order to
> > avoid calling restore_reserve_on_error() without the vma lock.
>
> But now you're calling vmf_anon_prepare() in the wrong place -- before
> we've determined that we need an anon folio. So we'll create an
> anon_vma even when we don't need one for this vma.
That's true. Though that can be addressed through something like:
if (!(vma->vm_flags & VM_MAYSHARE)) {
ret = vmf_anon_prepare(vmf);
if (unlikely(ret))
goto out;
}
> This is definitely a pre-existing bug which you've exposed by making it
> happen more easily. Needs a different fix though.
I interpreted the bug report to showcase how restore_reserve_on_error()
depends on the vma lock being held - and vmf_anon_prepare() drops that
lock by the time we get to restore_reserve_on_error(). In this case, this
would address it without reworking restore_reserve_on_error().
There very well could be something completely different going on, however
I have no ideas as to what that may be.
