Re: [GIT PULL] futex fixlet

From: Linus Torvalds
Date: Fri Dec 30 2011 - 15:06:50 EST

On Fri, Dec 30, 2011 at 9:07 AM, Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> wrote:
> So I _think_ you're completely right and we can simply kill the whole
> thing, but I've been trying very hard to forget everything kernel
> related for a week, and I really shouldn't kick-start my brain until
> somewhere next week.


Ingo, I'd suggest you put a patch like that into -next, and mark it
for stable after it has gotten lots more testing.

Something like the appended *totally* untested code. It compiles, but
maybe there really is some "!page->mapping" case that could possibly

I can't see it, though - but we should definitely get some testing for
this before I'd put it in a release. Since we do that
"get_user_pages_fast()" and ask for a writable page, the result should
be as stable as it will ever get. There is still the race with
hugepage splitting, but at least that one has a comment.

kernel/futex.c | 20 ++++++++++++--------
1 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/kernel/futex.c b/kernel/futex.c
index ea87f4d2f455..8d2ee2f66418 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -310,21 +310,25 @@ again:
if (page != page_head) {
+ /* Avoid "unused label" for non-THP config */
+ if (0)
+ goto again;

+ /*
+ * If the page doesn't have a mapping associated with it,
+ * this is either an anonymous page that was accessed RO
+ * (so no COW etc) and we would just return EFAULT anyway,
+ * or some special case page (ZERO_PAGE, gate area, special
+ * mapping..).
+ */
if (!page_head->mapping) {
- /*
- * ZERO_PAGE pages don't have a mapping. Avoid a busy loop
- * trying to find one. RW mapping would have COW'd (and thus
- * have a mapping) so this page is RO and won't ever change.
- */
- if ((page_head == ZERO_PAGE(address)))
- return -EFAULT;
- goto again;
+ return -EFAULT;