On Wed, Aug 11, 2021 at 4:45 AM David Hildenbrand [off-list ref] wrote:

I proposed a while ago to get rid of VM_DENYWRITE completely:

https://lkml.kernel.org/r/20210423131640.20080-1-david@redhat.com

I haven't looked how much it still applies to current upstream, but
maybe that might help cleaning up that code.

I like it.

I agree that we could - and probably should - just do it this way.

We don't expose MAP_DENYWRITE to user space any more - and the old
legacy library loading code certainly isn't worth it - and so
effectively the only way to set it is with execve().

And yes, it gets rid of all the silly games with the per-mapping flags.

               Linus