Re: [PATCH v3 kvm/queue 03/16] mm/memfd: Introduce MEMFD_OPS

[Chao Peng]

On Tue, Jan 04, 2022 at 05:38:38PM +0000, Sean Christopherson wrote:
> On Fri, Dec 31, 2021, Chao Peng wrote:
> > On Fri, Dec 24, 2021 at 11:53:15AM +0800, Robert Hoo wrote:
> > > On Thu, 2021-12-23 at 20:29 +0800, Chao Peng wrote:
> > > > From: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx>
> > > >  
> > > > +static void notify_fallocate(struct inode *inode, pgoff_t start,
> > > > pgoff_t end)
> > > > +{
> > > > +#ifdef CONFIG_MEMFD_OPS
> > > > +	struct shmem_inode_info *info = SHMEM_I(inode);
> > > > +	const struct memfd_falloc_notifier *notifier;
> > > > +	void *owner;
> > > > +	bool ret;
> > > > +
> > > > +	if (!info->falloc_notifier)
> > > > +		return;
> > > > +
> > > > +	spin_lock(&info->lock);
> > > > +	notifier = info->falloc_notifier;
> > > > +	if (!notifier) {
> > > > +		spin_unlock(&info->lock);
> > > > +		return;
> > > > +	}
> > > > +
> > > > +	owner = info->owner;
> > > > +	ret = notifier->get_owner(owner);
> > > > +	spin_unlock(&info->lock);
> > > > +	if (!ret)
> > > > +		return;
> > > > +
> > > > +	notifier->fallocate(inode, owner, start, end);
> > > 
> > > I see notifier->fallocate(), i.e. memfd_fallocate(), discards
> > > kvm_memfd_fallocate_range()'s return value. Should it be checked?
> > 
> > I think we can ignore it, just like how current mmu_notifier does,
> > the return value of __kvm_handle_hva_range is discarded in
> > kvm_mmu_notifier_invalidate_range_start(). Even when KVM side failed,
> > it's not fatal, it should not block the operation in the primary MMU.
> 
> If the return value is ignored, it'd be better to have no return value at all so
> that it's clear fallocate() will continue on regardless of whether or not the
> secondary MMU callback succeeds.  E.g. if KVM can't handle the fallocate() for
> whatever reason, then knowing that fallocate() will continue on means KVM should
> mark the VM as dead so that the broken setup cannot be abused by userspace.

After a close look, kvm_unmap_gfn_range() actually does not return a
error code, so it's safe to not return in kvm_memfd_handle_range().

Chao