releases/4.9.45/mm-fix-double-mmap_sem-unlock-on-mmf_unstable-enforced-sigbus.patch

   1 From 5b53a6ea886700a128b697a6fe8375340dea2c30 Mon Sep 17 00:00:00 2001
   2 From: Michal Hocko <mhocko@suse.com>
   3 Date: Fri, 18 Aug 2017 15:16:12 -0700
   4 Subject: mm: fix double mmap_sem unlock on MMF_UNSTABLE enforced SIGBUS
   5
   6 From: Michal Hocko <mhocko@suse.com>
   7
   8 commit 5b53a6ea886700a128b697a6fe8375340dea2c30 upstream.
   9
  10 Tetsuo Handa has noticed that MMF_UNSTABLE SIGBUS path in
  11 handle_mm_fault causes a lockdep splat
  12
  13   Out of memory: Kill process 1056 (a.out) score 603 or sacrifice child
  14   Killed process 1056 (a.out) total-vm:4268108kB, anon-rss:2246048kB, file-rss:0kB, shmem-rss:0kB
  15   a.out (1169) used greatest stack depth: 11664 bytes left
  16   DEBUG_LOCKS_WARN_ON(depth <= 0)
  17   ------------[ cut here ]------------
  18   WARNING: CPU: 6 PID: 1339 at kernel/locking/lockdep.c:3617 lock_release+0x172/0x1e0
  19   CPU: 6 PID: 1339 Comm: a.out Not tainted 4.13.0-rc3-next-20170803+ #142
  20   Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 07/02/2015
  21   RIP: 0010:lock_release+0x172/0x1e0
  22   Call Trace:
  23      up_read+0x1a/0x40
  24      __do_page_fault+0x28e/0x4c0
  25      do_page_fault+0x30/0x80
  26      page_fault+0x28/0x30
  27
  28 The reason is that the page fault path might have dropped the mmap_sem
  29 and returned with VM_FAULT_RETRY.  MMF_UNSTABLE check however rewrites
  30 the error path to VM_FAULT_SIGBUS and we always expect mmap_sem taken in
  31 that path.  Fix this by taking mmap_sem when VM_FAULT_RETRY is held in
  32 the MMF_UNSTABLE path.
  33
  34 We cannot simply add VM_FAULT_SIGBUS to the existing error code because
  35 all arch specific page fault handlers and g-u-p would have to learn a
  36 new error code combination.
  37
  38 Link: http://lkml.kernel.org/r/20170807113839.16695-2-mhocko@kernel.org
  39 Fixes: 3f70dc38cec2 ("mm: make sure that kthreads will not refault oom reaped memory")
  40 Reported-by: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
  41 Signed-off-by: Michal Hocko <mhocko@suse.com>
  42 Acked-by: David Rientjes <rientjes@google.com>
  43 Cc: Andrea Argangeli <andrea@kernel.org>
  44 Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
  45 Cc: Oleg Nesterov <oleg@redhat.com>
  46 Cc: Wenwei Tao <wenwei.tww@alibaba-inc.com>
  47 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  48 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  49 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  50
  51 ---
  52  mm/memory.c |   12 +++++++++++-
  53  1 file changed, 11 insertions(+), 1 deletion(-)
  54
  55 --- a/mm/memory.c
  56 +++ b/mm/memory.c
  57 @@ -3635,8 +3635,18 @@ int handle_mm_fault(struct vm_area_struc
  58          * further.
  59          */
  60         if (unlikely((current->flags & PF_KTHREAD) && !(ret & VM_FAULT_ERROR)
  61 -                               && test_bit(MMF_UNSTABLE, &vma->vm_mm->flags)))
  62 +                               && test_bit(MMF_UNSTABLE, &vma->vm_mm->flags))) {
  63 +
  64 +               /*
  65 +                * We are going to enforce SIGBUS but the PF path might have
  66 +                * dropped the mmap_sem already so take it again so that
  67 +                * we do not break expectations of all arch specific PF paths
  68 +                * and g-u-p
  69 +                */
  70 +               if (ret & VM_FAULT_RETRY)
  71 +                       down_read(&vma->vm_mm->mmap_sem);
  72                 ret = VM_FAULT_SIGBUS;
  73 +       }
  74
  75         return ret;
  76  }