Re: Ok, explained.. (was Re: [PATCH] mm: fix page_mkclean_one)

From: Nick Piggin
Date: Fri Dec 29 2006 - 07:10:22 EST

Hey nice work Linus!

Linus Torvalds wrote:

On Fri, 29 Dec 2006, Linus Torvalds wrote:

Hmm? I'd love it if somebody else wrote the patch and tested it, because I'm getting sick and tired of this bug ;)

Who the hell am I kidding? I haven't been able to sleep right for the last few days over this bug. It was really getting to me.

And putting on the thinking cap, there's actually a fairly simple an nonintrusive patch.

Yeah *this* makes more sense. And in retrospect it was simple, we
can't just throw out pte dirtiness information if the page doesn't
have all buffers dirtied.

It still has a tiny tiny race (see the comment), but I bet nobody can really hit it in real life anyway, and I know several ways to fix it, so I'm not really _that_ worried about it.

Well the race isn't a data loss one, is it? Just a case where the
pte may be dirty but the page dirty state not accounted for.

Can we fix it by just putting the page_mkclean back inside the
TestClearPageDirty check, and re-clearing PG_dirty after redoing
the set_page_dirty?

The patch is mostly a comment. The "real" meat of it is actually just a few lines.

Can anybody get corruption with this thing applied? It goes on top of plain v2.6.20-rc2.


diff --git a/mm/page-writeback.c b/mm/page-writeback.c
index b3a198c..ec01da1 100644
--- a/mm/page-writeback.c
+++ b/mm/page-writeback.c
@@ -862,17 +862,46 @@ int clear_page_dirty_for_io(struct page *page)
struct address_space *mapping = page_mapping(page);
- if (!mapping)
- return TestClearPageDirty(page);
- if (TestClearPageDirty(page)) {
- if (mapping_cap_account_dirty(mapping)) {
- page_mkclean(page);
+ if (mapping && mapping_cap_account_dirty(mapping)) {
+ /*
+ * Yes, Virginia, this is indeed insane.
+ *
+ * We use this sequence to make sure that
+ * (a) we account for dirty stats properly
+ * (b) we tell the low-level filesystem to
+ * mark the whole page dirty if it was
+ * dirty in a pagetable. Only to then
+ * (c) clean the page again and return 1 to
+ * cause the writeback.
+ *
+ * This way we avoid all nasty races with the
+ * dirty bit in multiple places and clearing
+ * them concurrently from different threads.
+ *
+ * Note! Normally the "set_page_dirty(page)"
+ * has no effect on the actual dirty bit - since
+ * that will already usually be set. But we
+ * need the side effects, and it can help us
+ * avoid races.
+ *
+ * We basically use the page "master dirty bit"
+ * as a serialization point for all the different
+ * threds doing their things.
+ *
+ * FIXME! We still have a race here: if somebody
+ * adds the page back to the page tables in
+ * between the "page_mkclean()" and the "TestClearPageDirty()",
+ * we might have it mapped without the dirty bit set.
+ */
+ if (page_mkclean(page))
+ set_page_dirty(page);
+ if (TestClearPageDirty(page)) {
dec_zone_page_state(page, NR_FILE_DIRTY);
+ return 1;
- return 1;
+ return 0;
- return 0;
+ return TestClearPageDirty(page);

SUSE Labs, Novell Inc.
Send instant messages to your online friends -
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at