On Tue, Oct 13, 2020 at 01:03:30PM -0700, Linus Torvalds wrote: > On Tue, Oct 13, 2020 at 12:59 PM Linus Torvalds > <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > > > Comments? So we remove strict serialization against truncation in filemap_map_pages() if the mapping is private. IIUC, we can end up with a page with ->mapping == NULL set up in a PTE for such mappings. The "page->mapping != mapping" check makes the race window smaller, but doesn't remove it. I'm not sure all codepaths are fine with this. For instance, looks like migration will back off such pages: __unmap_and_move() doesn't know how to deal with mapped pages with ->mapping == NULL. Yes, it is not crash, but still... Do I miss something? -- Kirill A. Shutemov
