futex: Fix uninterruptible loop due to gate_area

It was found (by Sasha) that if you use a futex located in the gate
area we get stuck in an uninterruptible infinite loop, much like the
ZERO_PAGE issue.

While looking at this problem, PeterZ realized you'll get into similar
trouble when hitting any install_special_pages() mapping.  And are there
still drivers setting up their own special mmaps without page->mapping,
and without special VM or pte flags to make get_user_pages fail?

In most cases, if page->mapping is NULL, we do not need to retry at all:
Linus points out that even /proc/sys/vm/drop_caches poses no problem,
because it ends up using remove_mapping(), which takes care not to
interfere when the page reference count is raised.

But there is still one case which does need a retry: if memory pressure
called shmem_writepage in between get_user_pages_fast dropping page
table lock and our acquiring page lock, then the page gets switched from
filecache to swapcache (and ->mapping set to NULL) whatever the refcount.
Fault it back in to get the page->mapping needed for key->shared.inode.

Reported-by: Sasha Levin <levinsasha928@gmail.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 20 insertions, 8 deletions

diff --git a/kernel/futex.c b/kernel/futex.c
index ea87f4d2f455..1614be20173d 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -314,17 +314,29 @@ again:
 #endif
 
         lock_page(page_head);
+
+        /*
+         * If page_head->mapping is NULL, then it cannot be a PageAnon
+         * page; but it might be the ZERO_PAGE or in the gate area or
+         * in a special mapping (all cases which we are happy to fail);
+         * or it may have been a good file page when get_user_pages_fast
+         * found it, but truncated or holepunched or subjected to
+         * invalidate_complete_page2 before we got the page lock (also
+         * cases which we are happy to fail).  And we hold a reference,
+         * so refcount care in invalidate_complete_page's remove_mapping
+         * prevents drop_caches from setting mapping to NULL beneath us.
+         *
+         * The case we do have to guard against is when memory pressure made
+         * shmem_writepage move it from filecache to swapcache beneath us:
+         * an unlikely race, but we do need to retry for page_head->mapping.
+         */
         if (!page_head->mapping) {
+                int shmem_swizzled = PageSwapCache(page_head);
                 unlock_page(page_head);
                 put_page(page_head);
-                /*
-                * 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;
+                if (shmem_swizzled)
+                        goto again;
+                return -EFAULT;
         }
 
         /*