Re: [PATCH v3 1/3] kernfs: remove page_mkwrite() from vm_operations_struct

[Christoph Hellwig]

On Thu, Jul 04, 2024 at 02:43:04PM -0600, Martin Oliveira wrote:
> On 2024-07-04 11:02, Matthew Wilcox wrote:> Seems to me we should actually _handle_ that, not do something wrong.
> > eg:
> > 
> >         if (vma->vm_ops) {
> >                 if (vma->vm_ops->close)
> >                         goto out_put;
> >                 if (WARN_ON(vma->vm_ops->page_mkwrite))
> >                         goto out_put;
> >         }
> 
> Good point.

Btw, sorry if I mislead you with my WARN_ON_ONCE suggestion.  That
was always intended in addition to the error handling, not instead.
(In fact there are very few reasons to use WARN_ON* without actually
handling the error as well).

> 
> > or maybe this doesn't need to be a WARN at all?  After all, there
> > isn't one for having a ->close method, so why is page_mkwrite special?
> 
> Hmm yeah, they should probably be treated the same.
> 
> Maybe ->close should be converted to WARN as well? It would be easier to
> catch an error this way than chasing the EINVAL, but I'm OK either way.

Yes, doing the same for ->close or anything unimplemented would be
nice.  But it's not really in scope for this series.

kernfs really should be using it's own ops instead of abusing
file_operations, but that's even more out of scope..