The race is as follows.  Suppose a multi-threaded task forks a new
process, thus bumping up the ref count on all the pages.  While the fork
is occurring (and thus we have marked all the PTEs as read-only), another
thread in the original process tries to write to a huge page, taking an
access violation from the write-protect and calling hugetlb_cow().  Now,
suppose the fork() fails.  It will undo the COW and decrement the ref
count on the pages, so the ref count on the huge page drops back to 1.
Meanwhile hugetlb_cow() also decrements the ref count by one on the
original page, since the original address space doesn't need it any more,
having copied a new page to replace the original page.  This leaves the
ref count at zero, and when we call unlock_page(), we panic.

The solution is to take an extra reference to the page while we are
holding the lock on it.

Cc: stable@kernel.org
Signed-off-by: Chris Metcalf <redacted>
---
 mm/hugetlb.c |    8 ++++++--
 1 files changed, 6 insertions(+), 2 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 4531be2..ab674fc 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c

@@ -2703,8 +2703,10 @@ int hugetlb_fault(struct mm_struct *mm, struct vm_area_struct *vma,

        * so no worry about deadlock.
        */
       page = pte_page(entry);
-       if (page != pagecache_page)
+       if (page != pagecache_page) {
+               get_page(page);
               lock_page(page);
+       }