binder: handle syscall restart in BINDER_WRITE_READ ioctl

The blocking wait in the "read" half of the ioctl could wake up on
signal delivery, and the default ERESTARTSYS handling would cause the
whole syscall to restart resulting in a doubled write.

Issue: AXIA-1880
Change-Id: Id49e4a088d08feca63bff43d28be6530820d9c3d
Signed-off-by: Andy Ross <andy.ross@windriver.com>

diff --git a/drivers/staging/android/binder.c b/drivers/staging/android/binder.c
index c69c40d..0782fc9 100644 (file)
--- a/drivers/staging/android/binder.c
+++ b/drivers/staging/android/binder.c
@@ -2190,8 +2190,13 @@ retry:
                proc->ready_threads--;
        thread->looper &= ~BINDER_LOOPER_STATE_WAITING;
 
+       /* We cannot return -ERESTARTSYS here.  This code is called
+        * after binder_thread_write() has already interpreted the
+        * input buffer.  A restart will result in a doubled set of
+        * commands.  Just return success, having consumed zero
+        * bytes. */
        if (ret)
-               return ret;
+               return ret == -ERESTARTSYS ? 0 : ret;
 
        while (1) {
                uint32_t cmd;