coredump: fix race condition between collapse_huge_page() and core dumping

When fixing the race conditions between the coredump and the mmap_sem
holders outside the context of the process, we focused on
mmget_not_zero()/get_task_mm() callers in 04f5866e41fb70 ("coredump: fix
race condition between mmget_not_zero()/get_task_mm() and core
dumping"), but those aren't the only cases where the mmap_sem can be
taken outside of the context of the process as Michal Hocko noticed
while backporting that commit to older -stable kernels.

If mmgrab() is called in the context of the process, but then the
mm_count reference is transferred outside the context of the process,
that can also be a problem if the mmap_sem has to be taken for writing
through that mm_count reference.

khugepaged registration calls mmgrab() in the context of the process,
but the mmap_sem for writing is taken later in the context of the
khugepaged kernel thread.

collapse_huge_page() after taking the mmap_sem for writing doesn't
modify any vma, so it's not obvious that it could cause a problem to the
coredump, but it happens to modify the pmd in a way that breaks an
invariant that pmd_trans_huge_lock() relies upon.  collapse_huge_page()
needs the mmap_sem for writing just to block concurrent page faults that
call pmd_trans_huge_lock().

Specifically the invariant that "!pmd_trans_huge()" cannot become a
"pmd_trans_huge()" doesn't hold while collapse_huge_page() runs.

The coredump will call __get_user_pages() without mmap_sem for reading,
which eventually can invoke a lockless page fault which will need a
functional pmd_trans_huge_lock().

So collapse_huge_page() needs to use mmget_still_valid() to check it's
not running concurrently with the coredump...  as long as the coredump
can invoke page faults without holding the mmap_sem for reading.

This has "Fixes: khugepaged" to facilitate backporting, but in my view
it's more a bug in the coredump code that will eventually have to be
rewritten to stop invoking page faults without the mmap_sem for reading.
So the long term plan is still to drop all mmget_still_valid().

Link: http://lkml.kernel.org/r/20190607161558.32104-1-aarcange@redhat.com
Fixes: ba76149f47d8 ("thp: khugepaged")
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: Michal Hocko <mhocko@suse.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mike Rapoport <rppt@linux.vnet.ibm.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Jason Gunthorpe <jgg@mellanox.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

2 files changed, 7 insertions, 0 deletions

diff --git a/include/linux/sched/mm.h b/include/linux/sched/mm.h
index a3fda9f024c3..4a7944078cc3 100644
--- a/include/linux/sched/mm.h
+++ b/include/linux/sched/mm.h
@@ -54,6 +54,10 @@ static inline void mmdrop(struct mm_struct *mm)
  * followed by taking the mmap_sem for writing before modifying the
  * vmas or anything the coredump pretends not to change from under it.
  *
+ * It also has to be called when mmgrab() is used in the context of
+ * the process, but then the mm_count refcount is transferred outside
+ * the context of the process to run down_write() on that pinned mm.
+ *
  * NOTE: find_extend_vma() called from GUP context is the only place
  * that can modify the "mm" (notably the vm_start/end) under mmap_sem
  * for reading and outside the context of the process, so it is also
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index a335f7c1fac4..0f7419938008 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1004,6 +1004,9 @@ static void collapse_huge_page(struct mm_struct *mm,
 	 * handled by the anon_vma lock + PG_lock.
 	 */
 	down_write(&mm->mmap_sem);
+	result = SCAN_ANY_PROCESS;
+	if (!mmget_still_valid(mm))
+		goto out;
 	result = hugepage_vma_revalidate(mm, address, &vma);
 	if (result)
 		goto out;