UPSTREAM: ANDROID: binder: remove waitqueue when thread exits.
authorMartijn Coenen <maco@android.com>
Fri, 5 Jan 2018 10:27:07 +0000 (11:27 +0100)
committerDanny Wood <danwood76@gmail.com>
Fri, 8 Nov 2019 12:03:13 +0000 (12:03 +0000)
binder_poll() passes the thread->wait waitqueue that
can be slept on for work. When a thread that uses
epoll explicitly exits using BINDER_THREAD_EXIT,
the waitqueue is freed, but it is never removed
from the corresponding epoll data structure. When
the process subsequently exits, the epoll cleanup
code tries to access the waitlist, which results in
a use-after-free.

Prevent this by using POLLFREE when the thread exits.

(cherry picked from commit f5cb779ba16334b45ba8946d6bfa6d9834d1527f)

Change-Id: Ib34b1cbb8ab2192d78c3d9956b2f963a66ecad2e
Signed-off-by: Martijn Coenen <maco@android.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: stable <stable@vger.kernel.org> # 4.14
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/android/binder.c

index f1e64f00d45c92b203efd58e8df59db65d30fcaa..f4aaae464c4b74763aaf63f315cb1293eb9c2d84 100644 (file)
@@ -4529,6 +4529,18 @@ static int binder_thread_release(struct binder_proc *proc,
                if (t)
                        spin_lock(&t->lock);
        }
+
+       /*
+        * If this thread used poll, make sure we remove the waitqueue
+        * from any epoll data structures holding it with POLLFREE.
+        * waitqueue_active() is safe to use here because we're holding
+        * the inner lock.
+        */
+       if ((thread->looper & BINDER_LOOPER_STATE_POLL) &&
+           waitqueue_active(&thread->wait)) {
+               wake_up_poll(&thread->wait, POLLHUP | POLLFREE);
+       }
+
        binder_inner_proc_unlock(thread->proc);
 
        if (send_reply)