Merge branch 'lukas/kernel/fix_os_cmd_close_race/OTP-14232' into maint

* lukas/kernel/fix_os_cmd_close_race/OTP-14232:
  kernel: Fix hanging os:cmd race condition

1 files changed, 15 insertions, 9 deletions

diff --git a/lib/kernel/src/os.erl b/lib/kernel/src/os.erl
index f8519d3a5e..03d4324992 100644
--- a/lib/kernel/src/os.erl
+++ b/lib/kernel/src/os.erl
@@ -289,12 +289,11 @@ get_data(Port, MonRef, Eot, Sofar) ->
                 more ->
                     get_data(Port, MonRef, Eot, [Sofar,Bytes]);
                 Last ->
-                    Port ! {self(), close},
-                    flush_until_closed(Port),
-                    flush_exit(Port),
+                    catch port_close(Port),
+                    flush_until_down(Port, MonRef),
                     iolist_to_binary([Sofar, Last])
             end;
-        {'DOWN', MonRef, _, _ , _} ->
+        {'DOWN', MonRef, _, _, _} ->
 	    flush_exit(Port),
 	    iolist_to_binary(Sofar)
     end.
@@ -308,18 +307,25 @@ eot(Bs, Eot) ->
             binary:part(Bs,{0, Pos})
     end.
 
-flush_until_closed(Port) ->
+%% When port_close returns we know that all the
+%% messages sent have been sent and that the
+%% DOWN message is after them all.
+flush_until_down(Port, MonRef) ->
     receive
         {Port, {data, _Bytes}} ->
-            flush_until_closed(Port);
-        {Port, closed} ->
-            true
+            flush_until_down(Port, MonRef);
+        {'DOWN', MonRef, _, _, _} ->
+            flush_exit(Port)
     end.
 
+%% The exit signal is always delivered before
+%% the down signal, so we can be sure that if there
+%% was an exit message sent, it will be in the
+%% mailbox now.
 flush_exit(Port) ->
     receive
         {'EXIT',  Port,  _} ->
             ok
-    after 1 ->				% force context switch
+    after 0 ->
             ok
     end.