Re: [RFC PATCH 1/3] mm: introduce new .mmap_proto() f_op callback

[Lorenzo Stoakes]

On Wed, Apr 30, 2025 at 11:58:14PM +0200, David Hildenbrand wrote:
> On 30.04.25 21:54, Lorenzo Stoakes wrote:
> > Provide a means by which drivers can specify which fields of those
> > permitted to be changed should be altered to prior to mmap()'ing a
> > range (which may either result from a merge or from mapping an entirely new
> > VMA).
> >
> > Doing so is substantially safer than the existing .mmap() calback which
> > provides unrestricted access to the part-constructed VMA and permits
> > drivers and file systems to do 'creative' things which makes it hard to
> > reason about the state of the VMA after the function returns.
> >
> > The existing .mmap() callback's freedom has caused a great deal of issues,
> > especially in error handling, as unwinding the mmap() state has proven to
> > be non-trivial and caused significant issues in the past, for instance
> > those addressed in commit 5de195060b2e ("mm: resolve faulty mmap_region()
> > error path behaviour").
> >
> > It also necessitates a second attempt at merge once the .mmap() callback
> > has completed, which has caused issues in the past, is awkward, adds
> > overhead and is difficult to reason about.
> >
> > The .mmap_proto() callback eliminates this requirement, as we can update
> > fields prior to even attempting the first merge. It is safer, as we heavily
> > restrict what can actually be modified, and being invoked very early in the
> > mmap() process, error handling can be performed safely with very little
> > unwinding of state required.
> >
> > Update vma userland test stubs to account for changes.
> >
> > Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@xxxxxxxxxx>
>
>
> I really don't like the "proto" terminology. :)
>
> [yes, David and his naming :P ]
>
> No, the problem is that it is fairly unintuitive what is happening here.
>
> Coming from a different direction, the callback is trigger after
> __mmap_prepare() ... could we call it "->mmap_prepare" or something like
> that? (mmap_setup, whatever)
>
> Maybe mmap_setup and vma_setup_param? Just a thought ...

Haha that's fine, I'm not sure I love 'proto' either to be honest, naming is
hard...

I would rather not refer to VMA's at all to be honest, if I had my way, no
driver would ever have access to a VMA at all...

But mmap_setup() or mmap_prepare() sound good!

>
>
> In general (although it's late in Germany), it does sound like an
> interesting approach.

Thanks! Appreciate it :) I really want to attack this, as I _hate_ how we
effectively allow drivers to do _anything_ with VMAs like this.

Yes, hate-driven development...

Locking this down is just a generally good idea I think!

Was late in UK too when I sent :P hence why I managed to not send it properly
the first time... (sorry again...)

>
> How feasiable is it to remove ->mmap in the long run, and would we maybe
> need other callbacks to make that possible?

I do think it is, because we can do this super-incrementally, and I'm willing to
put in the legwork to gradually move drivers over.

I think it might be folio-like in taking some time, but we'll get there
(obviously _nowhere near_ the impact of that work, a mere humble effort, but
comparable somewhat in this regard).

I actually took the time to look through ~350 or so .mmap() callbacks so it's
not so crazy to delve in either.

Re: other callbacks, yes I suspect we will need. But I think we are fine to
start with this and add as needed.

I suspect esp. given Jann's comments we might want to make .mmap_prepare() and
.mmap() mutualy exclusive actually. Idea of allowing them both wass flexibility
but I think is more downside than up.

We can then add additional callbacks as needed. Also good for the transition
away from .mmap() which I really want to absolutely deprecate.

>
>
> --
> Cheers,
>
> David / dhildenb
>