vm_page: Tighten the object lock assertion in vm_page_invalid()

[markj]

Summary

A page must not become invalid while vm_fault_soft_fast() is attempting
to map unbusied pages for reading.

Note that all callers hold the object write lock already, and
vm_page_set_invalid() asserts the object write lock.

I am not really sure it is wise to permit clearing of valid bits without
xbusy, either. Currently all vm_page_invalid() callers have the page
xbusied, but there is one user of vm_page_set_invalid() that only
sbusies the page. It can be trivially converted to xbusy the page, I
think. Is there any reason to permit clearing of valid bits from a
shared-busy page?

[kib]

Are you referencing vfs_bio' vfs_vmio_invalidate() as the place where page is sbusied for invalidation? Generally buffer cache only uses sbusy and not xbusy, because it might recurse.

[markj]

In D33250#751733, @kib wrote:

Are you referencing vfs_bio' vfs_vmio_invalidate() as the place where page is sbusied for invalidation?

Yes.

Generally buffer cache only uses sbusy and not xbusy, because it might recurse.

Right, ok. In this one case I believe it may not be a problem, not sure yet.