On Mon, Oct 24, 2022 at 5:21 PM Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
>
> It's all about updating read only pages that are executable with a shadow mm.
Right. And it doesn't actually need the mm at all, all it wants is the
kernel page tables. Which is why all the "dup_mmap()" stuff seems so
wrong.
I suspect mm_alloc() does everything that VM actually needs.
IOW, it shouldn't have used the fork() helper, it should have used the
execve() helper that actually starts out from a clean slate. Because a
clean slate is exactly what that code wants.
No?
Linus
