mm/gup: Mark lock taken only after a successful retake

It's definitely incorrect to mark the lock as taken even if
down_read_killable() failed.

This wass overlooked when we switched from down_read() to
down_read_killable() because down_read() won't fail while
down_read_killable() could.

Fixes: cc2c8bd3e88f ("mm/gup: allow to react to fatal signals")
Reported-by: syzbot+a8c70b7f3579fc0587dc@syzkaller.appspotmail.com
Signed-off-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/mm/gup.c b/mm/gup.c
index a21230569520913dedbae1ceec94f2df504eb9f4..afce0bc47e70124d20947b569e8877f1396f9c1a 100644 (file)
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -1329,7 +1329,6 @@ retry:
                if (fatal_signal_pending(current))
                        break;
 
-               *locked = 1;
                ret = down_read_killable(&mm->mmap_sem);
                if (ret) {
                        BUG_ON(ret > 0);
@@ -1338,6 +1337,7 @@ retry:
                        break;
                }
 
+               *locked = 1;
                ret = __get_user_pages(tsk, mm, start, 1, flags | FOLL_TRIED,
                                       pages, NULL, locked);
                if (!*locked) {