Re: [PATCH v4 3/3] userfaultfd: use per-vma locks in userfaultfd operations

From: Liam R. Howlett
Date: Mon Feb 12 2024 - 17:54:26 EST

* Lokesh Gidra <lokeshgidra@xxxxxxxxxx> [240212 17:31]:
> I have also introduced a handler for finding dst_vma and preparing its
> anon_vma, which is used in lock_vma() and find_vmas_mm_locked().
> Sounds good?
> > I've also thought of how you can name the abstraction in the functions:
> > use a 'prepare() and complete()' to find/lock and unlock what you need.
> > Might be worth exploring? If we fail to 'prepare()' then we don't need
> > to 'complete()', which means there won't be mismatched locking hanging
> > around. Maybe it's too late to change to this sort of thing, but I
> > thought I'd mention it.
> >
> Nice suggestion! But after (fortunately) finding the function names
> that are self-explanatory, dropping them seems like going in the wrong
> direction. Please let me know if you think this is a missing piece. I
> am open to incorporating this.

This plan sounds good, thanks!