mm/munlock: mlock_vma_page() check against VM_SPECIAL

Although mmap_region() and mlock_fixup() take care that VM_LOCKED
is never left set on a VM_SPECIAL vma, there is an interval while
file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
still be set while VM_SPECIAL bits are added: so mlock_vma_page()
should ignore VM_LOCKED while any VM_SPECIAL bits are set.

This showed up as a "Bad page" still mlocked, when vfree()ing pages
which had been vm_inserted by remap_vmalloc_range_partial(): while
release_pages() and __page_cache_release(), and so put_page(), catch
pages still mlocked when freeing (and clear_page_mlock() caught them
when unmapping), the vfree() path is unprepared for them: fix it?
but these pages should not have been mlocked in the first place.

I assume that an mlockall(MCL_FUTURE) had been done in the past; or
maybe the user got to specify MAP_LOCKED on a vmalloc'ing driver mmap.
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>

mm/internal.h

@@ -413,8 +413,15 @@ void mlock_page(struct page *page);
 static inline void mlock_vma_page(struct page *page,
 			struct vm_area_struct *vma, bool compound)
 {
-	/* VM_IO check prevents migration from double-counting during mlock */
-	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_IO)) == VM_LOCKED) &&
+	/*
+	 * The VM_SPECIAL check here serves two purposes.
+	 * 1) VM_IO check prevents migration from double-counting during mlock.
+	 * 2) Although mmap_region() and mlock_fixup() take care that VM_LOCKED
+	 *    is never left set on a VM_SPECIAL vma, there is an interval while
+	 *    file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may
+	 *    still be set while VM_SPECIAL bits are added: so ignore it then.
+	 */
+	if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) &&
 	    (compound || !PageTransCompound(page)))
 		mlock_page(page);
 }