Re: [rfc patch] mm: protect set_page_dirty() from ongoing truncation

[Johannes Weiner]

On Wed, Nov 26, 2014 at 02:00:06PM -0800, Andrew Morton wrote:
> On Tue, 25 Nov 2014 14:48:41 -0500 Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
> >  The
> > same btw applies for the page_mkwrite case: how is mapping safe to
> > pass to balance_dirty_pages() after unlocking page table and page?
> 
> I'm not sure which code you're referring to here, but it's likely that
> the switch-balancing-to-bdi approach will address that as well?

This code in do_wp_page():

		pte_unmap_unlock(page_table, ptl);
[...]
		put_page(dirty_page);
		if (page_mkwrite) {
			struct address_space *mapping = dirty_page->mapping;

			set_page_dirty(dirty_page);
			unlock_page(dirty_page);
			page_cache_release(dirty_page);
			if (mapping)	{
				/*
				 * Some device drivers do not set page.mapping
				 * but still dirty their pages
				 */
				balance_dirty_pages_ratelimited(mapping);
			}
		}

And there is also this code in do_shared_fault():

	pte_unmap_unlock(pte, ptl);

	if (set_page_dirty(fault_page))
		dirtied = 1;
	mapping = fault_page->mapping;
	unlock_page(fault_page);
	if ((dirtied || vma->vm_ops->page_mkwrite) && mapping) {
		/*
		 * Some device drivers do not set page.mapping but still
		 * dirty their pages
		 */
		balance_dirty_pages_ratelimited(mapping);
	}

I don't see anything that ensures mapping stays alive by the time it's
passed to balance_dirty_pages() in either case.

Argh, but of course there is.  The mmap_sem.  That pins the vma, which
pins the file, which pins the inode.  In all cases.  So I think we can
just stick with passing mapping to balance_dirty_pages() for now.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/