[PATCH v1 16/16] mm/memory: support VM_MIXEDMAP in zap_special_vma_range()

[Alice Ryhl]

On Wed, Mar 11, 2026 at 09:04:18AM -0300, Jason Gunthorpe wrote:
> On Wed, Mar 11, 2026 at 09:38:45AM +0000, Alice Ryhl wrote:
> > It doesn't really make sense to have multiple binder VMAs. What happens
> > with Rust Binder is that process A is receiving transactions and has the
> > VMA mapped once.
> 
> IIRC the problem is the kernel doesn't guarentee singleton VMAs,
> userspace can always clone them with fork or something. Did binder
> solve this somehow?

The Binder VMA is DONTCOPY, so it will not be present after fork.

> Since you can't assume there is only one VMA the locking becomes a
> mess to cover all the cases where userspace can trigger a VMA clone.
> 
> address space deals with this internally.
> 
> Thus, zap_special_vma_range() is extremely hard to use.

I mean, the hard part about the locking is keeping them in sync. Binder
just doesn't do that. Only the original VMA gets pages inserted or
zapped. If you create a second VMA, you just get a useless read-only VMA
that you can't do anything with.

Alice