On Fri, Apr 19, 2024 at 02:58:43PM +0200, David Hildenbrand wrote:
> On 18.04.24 21:41, Michael Roth wrote:
> > kvm_gmem_get_folio() may return a PTR_ERR() rather than just NULL. In
> > particular, for cases where EEXISTS is returned when FGP_CREAT_ONLY
> > flag is used. Handle this properly in __kvm_gmem_get_pfn().
> >
> > Signed-off-by: Michael Roth <michael.roth@xxxxxxx>
> > ---
> > virt/kvm/guest_memfd.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/virt/kvm/guest_memfd.c b/virt/kvm/guest_memfd.c
> > index ccf22e44f387..9d7c6a70c547 100644
> > --- a/virt/kvm/guest_memfd.c
> > +++ b/virt/kvm/guest_memfd.c
> > @@ -580,8 +580,8 @@ static int __kvm_gmem_get_pfn(struct file *file, struct kvm_memory_slot *slot,
> > }
> > folio = kvm_gmem_get_folio(file_inode(file), index, prepare);
> > - if (!folio)
> > - return -ENOMEM;
> > + if (IS_ERR_OR_NULL(folio))
> > + return folio ? PTR_ERR(folio) : -ENOMEM;
>
> Will it even return NULL? Staring at other filemap_grab_folio() users, they
> all check for IS_ERR().
Looks like the NULL case is handled with PTR_ERR(-ENOENT), so IS_ERR()
would be sufficient. I think in the past kvm_gmem_get_folio() itself
would return NULL in some cases, but as of commit 2b01b7e994e95 that's
no longer the case.
I'll fix this up to expect only PTR_ERR() when I re-spin v14, and also
address the other kvm_gmem_get_folio() / __filemap_get_folio() call
sites.
>
> > if (folio_test_hwpoison(folio)) {
> > r = -EHWPOISON;
>
> Do we have a Fixes: tag?
Fixes: 2b01b7e994e95 ("KVM: guest_memfd: pass error up from filemap_grab_folio")
Will add that in the re-spin as well.
Thanks!
-Mike
>
> --
> Cheers,
>
> David / dhildenb
>
