userfaultfd: shmem: handle coredumping in handle_userfault()

Anon and hugetlbfs handle FOLL_DUMP set by get_dump_page() internally to
__get_user_pages().

shmem as opposed has no special FOLL_DUMP handling there so
handle_mm_fault() is invoked without mmap_sem and ends up calling
handle_userfault() that isn't expecting to be invoked without mmap_sem
held.

This makes handle_userfault() fail immediately if invoked through
shmem_vm_ops->fault during coredumping and solves the problem.

The side effect is a BUG_ON with no lock held triggered by the
coredumping process which exits.  Only 4.11 is affected, pre-4.11 anon
memory holes are skipped in __get_user_pages by checking FOLL_DUMP
explicitly against empty pagetables (mm/gup.c:no_page_table()).

It's zero cost as we already had a check for current->flags to prevent
futex to trigger userfaults during exit (PF_EXITING).

Link: http://lkml.kernel.org/r/20170615214838.27429-1-aarcange@redhat.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reported-by: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: <stable@vger.kernel.org>	[4.11+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 21 insertions, 8 deletions

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index f7555fc25877..1d622f276e3a 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -340,9 +340,28 @@ int handle_userfault(struct vm_fault *vmf, unsigned long reason)
         bool must_wait, return_to_userland;
         long blocking_state;
 
-        BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
-
         ret = VM_FAULT_SIGBUS;
+
+        /*
+         * We don't do userfault handling for the final child pid update.
+         *
+         * We also don't do userfault handling during
+         * coredumping. hugetlbfs has the special
+         * follow_hugetlb_page() to skip missing pages in the
+         * FOLL_DUMP case, anon memory also checks for FOLL_DUMP with
+         * the no_page_table() helper in follow_page_mask(), but the
+         * shmem_vm_ops->fault method is invoked even during
+         * coredumping without mmap_sem and it ends up here.
+         */
+        if (current->flags & (PF_EXITING|PF_DUMPCORE))
+                goto out;
+
+        /*
+         * Coredumping runs without mmap_sem so we can only check that
+         * the mmap_sem is held, if PF_DUMPCORE was not set.
+         */
+        WARN_ON_ONCE(!rwsem_is_locked(&mm->mmap_sem));
+
         ctx = vmf->vma->vm_userfaultfd_ctx.ctx;
         if (!ctx)
                 goto out;
@@ -361,12 +380,6 @@ int handle_userfault(struct vm_fault *vmf, unsigned long reason)
                 goto out;
 
         /*
-         * We don't do userfault handling for the final child pid update.
-         */
-        if (current->flags & PF_EXITING)
-                goto out;
-
-        /*
          * Check that we can return VM_FAULT_RETRY.
          *
          * NOTE: it should become possible to return VM_FAULT_RETRY