kernel,ssh: Add synchronous user_drv protocol

Added a put_chars_sync to the protocol that can be used to
talk to user_drv and made group use it. This is needed in order
to guarantee that bytes has been pushed to the tty port when
doing something like this:

io:format("halting\n"),erlang:halt(0).

Before this change the halting message could be lost in the message
queue of the user_drv process, this is no longer possible.

This commit also fixes ssh_cli as that plugs itself in as a user_drv
process.

OTP-12240

5 files changed, 267 insertions, 120 deletions

diff --git a/erts/emulator/drivers/unix/ttsl_drv.c b/erts/emulator/drivers/unix/ttsl_drv.c
index 73ce27c6ec..be2fee1f25 100644
--- a/erts/emulator/drivers/unix/ttsl_drv.c
+++ b/erts/emulator/drivers/unix/ttsl_drv.c
@@ -93,6 +93,7 @@ static volatile int cols_needs_update = FALSE;
 #define OP_INSC 2
 #define OP_DELC 3
 #define OP_BEEP 4
+#define OP_PUTC_SYNC 5
 /* Control op */
 #define CTRL_OP_GET_WINSIZE 100
 #define CTRL_OP_GET_UNICODE_STATE 101
@@ -137,6 +138,7 @@ static Sint16 get_sint16(char*);
 static ErlDrvPort ttysl_port;
 static int ttysl_fd;
 static int ttysl_terminate = 0;
+static int ttysl_send_ok = 0;
 static ErlDrvBinary *putcbuf;
 static int putcpos;
 static int putclen;
@@ -683,6 +685,14 @@ static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT coun
 	put_chars((byte*)"\n", 1);
 
     switch (buf[0]) {
+    case OP_PUTC_SYNC:
+        /* Using sync means that we have to send an ok to the
+           controlling process for each command call. We delay
+           sending ok if the driver queue exceeds a certain size.
+           We do not set ourselves as a busy port, as this
+           could be very bad for user_drv, if it gets blocked on
+           the port_command. */
+        /* fall through */
     case OP_PUTC:
 	DEBUGLOG(("OP: Putc(%lu)",(unsigned long) count-1));
 	if (check_buf_size((byte*)buf+1, count-1) == 0)
@@ -736,6 +746,22 @@ static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT coun
         }
     }
 
+    if (buf[0] == OP_PUTC_SYNC) {
+        if (driver_sizeq(ttysl_port) > TTY_BUFFSIZE && !ttysl_terminate) {
+            /* We delay sending the ack until the buffer has been consumed */
+            ttysl_send_ok = 1;
+        } else {
+            ErlDrvTermData spec[] = {
+                ERL_DRV_PORT, driver_mk_port(ttysl_port),
+                ERL_DRV_ATOM, driver_mk_atom("ok"),
+                ERL_DRV_TUPLE, 2
+            };
+            ASSERT(ttysl_send_ok == 0);
+            erl_drv_output_term(driver_mk_port(ttysl_port), spec,
+                                sizeof(spec) / sizeof(spec[0]));
+        }
+    }
+
     return; /* TRUE; */
 }
 
@@ -758,6 +784,16 @@ static void ttysl_to_tty(ErlDrvData ttysl_data, ErlDrvEvent fd) {
             }
         } else {
             sz = driver_deq(ttysl_port, written);
+            if (sz < TTY_BUFFSIZE && ttysl_send_ok) {
+                ErlDrvTermData spec[] = {
+                    ERL_DRV_PORT, driver_mk_port(ttysl_port),
+                    ERL_DRV_ATOM, driver_mk_atom("ok"),
+                    ERL_DRV_TUPLE, 2
+                };
+                ttysl_send_ok = 0;
+                erl_drv_output_term(driver_mk_port(ttysl_port), spec,
+                                    sizeof(spec) / sizeof(spec[0]));
+            }
             if (sz == 0) {
                 driver_select(ttysl_port,(ErlDrvEvent)(long)ttysl_fd,
                               ERL_DRV_WRITE,0);
diff --git a/erts/emulator/drivers/win32/ttsl_drv.c b/erts/emulator/drivers/win32/ttsl_drv.c
index 502cb58dfa..851c336a11 100644
--- a/erts/emulator/drivers/win32/ttsl_drv.c
+++ b/erts/emulator/drivers/win32/ttsl_drv.c
@@ -46,6 +46,7 @@ static int rows;		/* Number of rows available. */
 #define OP_INSC 2
 #define OP_DELC 3
 #define OP_BEEP 4
+#define OP_PUTC_SYNC 5
 
 /* Control op */
 #define CTRL_OP_GET_WINSIZE 100
@@ -458,6 +459,7 @@ static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT coun
 
     switch (buf[0]) {
     case OP_PUTC:
+    case OP_PUTC_SYNC:
 	DEBUGLOG(("OP: Putc(%I64u)",(unsigned long long)count-1));
 	if (check_buf_size((byte*)buf+1, count-1) == 0)
 	    return; 
@@ -481,6 +483,20 @@ static void ttysl_from_erlang(ErlDrvData ttysl_data, char* buf, ErlDrvSizeT coun
 	/* Unknown op, just ignore. */
 	break;
     }
+
+    if (buf[0] == OP_PUTC_SYNC) {
+        /* On windows we do a blocking write to the tty so we just
+           send the ack immidiately. If at some point in the future
+           someone has a problem with tty output being blocking
+           this has to be changed. */
+        ErlDrvTermData spec[] = {
+            ERL_DRV_PORT, driver_mk_port(ttysl_port),
+            ERL_DRV_ATOM, driver_mk_atom("ok"),
+            ERL_DRV_TUPLE, 2
+        };
+        erl_drv_output_term(driver_mk_port(ttysl_port), spec,
+                            sizeof(spec) / sizeof(spec[0]));
+    }
     return;
 }
 
diff --git a/lib/kernel/src/group.erl b/lib/kernel/src/group.erl
index b36dbf33dd..046885f885 100644
--- a/lib/kernel/src/group.erl
+++ b/lib/kernel/src/group.erl
@@ -111,8 +111,13 @@ start_shell1(Fun) ->
 server_loop(Drv, Shell, Buf0) ->
     receive
 	{io_request,From,ReplyAs,Req} when is_pid(From) ->
-	    Buf = io_request(Req, From, ReplyAs, Drv, Buf0),
-	    server_loop(Drv, Shell, Buf);
+            %% This io_request may cause a transition to a couple of
+            %% selective receive loops elsewhere in this module.
+            Buf = io_request(Req, From, ReplyAs, Drv, Buf0),
+            server_loop(Drv, Shell, Buf);
+        {reply,{{From,ReplyAs},Reply}} ->
+            io_reply(From, ReplyAs, Reply),
+	    server_loop(Drv, Shell, Buf0);
 	{driver_id,ReplyTo} ->
 	    ReplyTo ! {self(),driver_id,Drv},
 	    server_loop(Drv, Shell, Buf0);
@@ -172,10 +177,13 @@ set_unicode_state(Drv,Bool) ->
 			   
 
 io_request(Req, From, ReplyAs, Drv, Buf0) ->
-    case io_request(Req, Drv, Buf0) of
+    case io_request(Req, Drv, {From,ReplyAs}, Buf0) of
 	{ok,Reply,Buf} ->
 	    io_reply(From, ReplyAs, Reply),
 	    Buf;
+        {noreply,Buf} ->
+            %% We expect a {reply,_} message from the Drv when request is done
+            Buf;
 	{error,Reply,Buf} ->
 	    io_reply(From, ReplyAs, Reply),
 	    Buf;
@@ -196,78 +204,85 @@ io_request(Req, From, ReplyAs, Drv, Buf0) ->
 %% io_request({put_chars,unicode,Binary}, Drv, Buf) when is_binary(Binary) -> 
 %%     send_drv(Drv, {put_chars,Binary}),
 %%     {ok,ok,Buf};
-io_request({put_chars,unicode,Chars}, Drv, Buf) ->
+%%
+%% These put requests have to be synchronous to the driver as otherwise
+%% there is no guarantee that the data has actually been printed.
+io_request({put_chars,unicode,Chars}, Drv, From, Buf) ->
     case catch unicode:characters_to_binary(Chars,utf8) of
 	Binary when is_binary(Binary) ->
-	    send_drv(Drv, {put_chars, unicode, Binary}),
-	    {ok,ok,Buf};
+	    send_drv(Drv, {put_chars_sync, unicode, Binary, {From,ok}}),
+	    {noreply,Buf};
 	_ ->
 	    {error,{error,{put_chars, unicode,Chars}},Buf}
     end;
-io_request({put_chars,unicode,M,F,As}, Drv, Buf) ->
+io_request({put_chars,unicode,M,F,As}, Drv, From, Buf) ->
     case catch apply(M, F, As) of
 	Binary when is_binary(Binary) ->
-	    send_drv(Drv, {put_chars, unicode,Binary}),
-	    {ok,ok,Buf};
+	    send_drv(Drv, {put_chars_sync, unicode, Binary, {From,ok}}),
+	    {noreply,Buf};
 	Chars ->
 	    case catch unicode:characters_to_binary(Chars,utf8) of
 		B when is_binary(B) ->
-		    send_drv(Drv, {put_chars, unicode,B}),
-		    {ok,ok,Buf};
+		    send_drv(Drv, {put_chars_sync, unicode, B, {From,ok}}),
+		    {noreply,Buf};
 		_ ->
 		    {error,{error,F},Buf}
 	    end
     end;
-io_request({put_chars,latin1,Binary}, Drv, Buf) when is_binary(Binary) -> 
-    send_drv(Drv, {put_chars, unicode,unicode:characters_to_binary(Binary,latin1)}),
-    {ok,ok,Buf};
-io_request({put_chars,latin1,Chars}, Drv, Buf) ->
+io_request({put_chars,latin1,Binary}, Drv, From, Buf) when is_binary(Binary) -> 
+    send_drv(Drv, {put_chars_sync, unicode,
+                   unicode:characters_to_binary(Binary,latin1),
+                   {From,ok}}),
+    {noreply,Buf};
+io_request({put_chars,latin1,Chars}, Drv, From, Buf) ->
     case catch unicode:characters_to_binary(Chars,latin1) of
 	Binary when is_binary(Binary) ->
-	    send_drv(Drv, {put_chars, unicode,Binary}),
-	    {ok,ok,Buf};
+	    send_drv(Drv, {put_chars_sync, unicode, Binary, {From,ok}}),
+	    {noreply,Buf};
 	_ ->
 	    {error,{error,{put_chars,latin1,Chars}},Buf}
     end;
-io_request({put_chars,latin1,M,F,As}, Drv, Buf) ->
+io_request({put_chars,latin1,M,F,As}, Drv, From, Buf) ->
     case catch apply(M, F, As) of
 	Binary when is_binary(Binary) ->
-	    send_drv(Drv, {put_chars, unicode,unicode:characters_to_binary(Binary,latin1)}),
-	    {ok,ok,Buf};
+	    send_drv(Drv, {put_chars_sync, unicode,
+                           unicode:characters_to_binary(Binary,latin1),
+                           {From,ok}}),
+	    {noreply,Buf};
 	Chars ->
 	    case catch unicode:characters_to_binary(Chars,latin1) of
 		B when is_binary(B) ->
-		    send_drv(Drv, {put_chars, unicode,B}),
-		    {ok,ok,Buf};
+		    send_drv(Drv, {put_chars_sync, unicode, B, {From,ok}}),
+		    {noreply,Buf};
 		_ ->
 		    {error,{error,F},Buf}
 	    end
     end;
 
-io_request({get_chars,Encoding,Prompt,N}, Drv, Buf) ->
+io_request({get_chars,Encoding,Prompt,N}, Drv, _From, Buf) ->
     get_chars(Prompt, io_lib, collect_chars, N, Drv, Buf, Encoding);
-io_request({get_line,Encoding,Prompt}, Drv, Buf) ->
+io_request({get_line,Encoding,Prompt}, Drv, _From, Buf) ->
     get_chars(Prompt, io_lib, collect_line, [], Drv, Buf, Encoding);
-io_request({get_until,Encoding, Prompt,M,F,As}, Drv, Buf) ->
+io_request({get_until,Encoding, Prompt,M,F,As}, Drv, _From, Buf) ->
     get_chars(Prompt, io_lib, get_until, {M,F,As}, Drv, Buf, Encoding);
-io_request({get_password,_Encoding},Drv,Buf) ->
+io_request({get_password,_Encoding},Drv,_From,Buf) ->
     get_password_chars(Drv, Buf);
-io_request({setopts,Opts}, Drv, Buf) when is_list(Opts) ->
+io_request({setopts,Opts}, Drv, _From, Buf) when is_list(Opts) ->
     setopts(Opts, Drv, Buf);
-io_request(getopts, Drv, Buf) ->
+io_request(getopts, Drv, _From, Buf) ->
     getopts(Drv, Buf);
-io_request({requests,Reqs}, Drv, Buf) ->
-    io_requests(Reqs, {ok,ok,Buf}, Drv);
+io_request({requests,Reqs}, Drv, From, Buf) ->
+    io_requests(Reqs, {ok,ok,Buf}, From, Drv);
 
 %% New in R12
-io_request({get_geometry,columns},Drv,Buf) ->
+io_request({get_geometry,columns},Drv,_From,Buf) ->
     case get_tty_geometry(Drv) of
 	{W,_H} ->
 	    {ok,W,Buf};
 	_ ->
 	    {error,{error,enotsup},Buf}
     end;
-io_request({get_geometry,rows},Drv,Buf) ->
+io_request({get_geometry,rows},Drv,_From,Buf) ->
     case get_tty_geometry(Drv) of
 	{_W,H} ->
 	    {ok,H,Buf};
@@ -276,38 +291,49 @@ io_request({get_geometry,rows},Drv,Buf) ->
     end;
 
 %% BC with pre-R13
-io_request({put_chars,Chars}, Drv, Buf) ->
-    io_request({put_chars,latin1,Chars}, Drv, Buf);
-io_request({put_chars,M,F,As}, Drv, Buf) ->
-    io_request({put_chars,latin1,M,F,As}, Drv, Buf);
-io_request({get_chars,Prompt,N}, Drv, Buf) ->
-    io_request({get_chars,latin1,Prompt,N}, Drv, Buf);
-io_request({get_line,Prompt}, Drv, Buf) ->
-    io_request({get_line,latin1,Prompt}, Drv, Buf);
-io_request({get_until, Prompt,M,F,As}, Drv, Buf) ->
-    io_request({get_until,latin1, Prompt,M,F,As}, Drv, Buf);
-io_request(get_password,Drv,Buf) ->
-    io_request({get_password,latin1},Drv,Buf);
-
-
-
-io_request(_, _Drv, Buf) ->
+io_request({put_chars,Chars}, Drv, From, Buf) ->
+    io_request({put_chars,latin1,Chars}, Drv, From, Buf);
+io_request({put_chars,M,F,As}, Drv, From, Buf) ->
+    io_request({put_chars,latin1,M,F,As}, Drv, From, Buf);
+io_request({get_chars,Prompt,N}, Drv, From, Buf) ->
+    io_request({get_chars,latin1,Prompt,N}, Drv, From, Buf);
+io_request({get_line,Prompt}, Drv, From, Buf) ->
+    io_request({get_line,latin1,Prompt}, Drv, From, Buf);
+io_request({get_until, Prompt,M,F,As}, Drv, From, Buf) ->
+    io_request({get_until,latin1, Prompt,M,F,As}, Drv, From, Buf);
+io_request(get_password,Drv,From,Buf) ->
+    io_request({get_password,latin1},Drv,From,Buf);
+
+
+
+io_request(_, _Drv, _From, Buf) ->
     {error,{error,request},Buf}.
 
-%% Status = io_requests(RequestList, PrevStat, Drv)
-%%  Process a list of output requests as long as the previous status is 'ok'.
-
-io_requests([R|Rs], {ok,ok,Buf}, Drv) ->
-    io_requests(Rs, io_request(R, Drv, Buf), Drv);
-io_requests([_|_], Error, _Drv) ->
+%% Status = io_requests(RequestList, PrevStat, From, Drv)
+%%  Process a list of output requests as long as
+%%  the previous status is 'ok' or noreply.
+%%
+%%  We use undefined as the From for all but the last request
+%%  in order to discards acknowledgements from those requests.
+%%
+io_requests([R|Rs], {noreply,Buf}, From, Drv) ->
+    ReqFrom = if Rs =:= [] -> From; true -> undefined end,
+    io_requests(Rs, io_request(R, Drv, ReqFrom, Buf), From, Drv);
+io_requests([R|Rs], {ok,ok,Buf}, From, Drv) ->
+    ReqFrom = if Rs =:= [] -> From; true -> undefined end,
+    io_requests(Rs, io_request(R, Drv, ReqFrom, Buf), From, Drv);
+io_requests([_|_], Error, _From, _Drv) ->
     Error;
-io_requests([], Stat, _) ->
+io_requests([], Stat, _From, _) ->
     Stat.
 
 %% io_reply(From, ReplyAs, Reply)
 %%  The function for sending i/o command acknowledgement.
 %%  The ACK contains the return value.
 
+io_reply(undefined, _ReplyAs, _Reply) ->
+    %% Ignore these replies as they are generated from io_requests/4.
+    ok;
 io_reply(From, ReplyAs, Reply) ->
     From ! {io_reply,ReplyAs,Reply},
     ok.
@@ -619,6 +645,10 @@ more_data(What, Cont0, Drv, Ls, Encoding) ->
 	    io_request(Req, From, ReplyAs, Drv, []), %WRONG!!!
 	    send_drv_reqs(Drv, edlin:redraw_line(Cont)),
 	    get_line1({more_chars,Cont,[]}, Drv, Ls, Encoding);
+        {reply,{{From,ReplyAs},Reply}} ->
+            %% We take care of replies from puts here as well
+            io_reply(From, ReplyAs, Reply),
+            more_data(What, Cont0, Drv, Ls, Encoding);
 	{'EXIT',Drv,interrupt} ->
 	    interrupted;
 	{'EXIT',Drv,_} ->
@@ -641,6 +671,10 @@ get_line_echo_off1({Chars,[]}, Drv) ->
 	{io_request,From,ReplyAs,Req} when is_pid(From) ->
 	    io_request(Req, From, ReplyAs, Drv, []),
 	    get_line_echo_off1({Chars,[]}, Drv);
+        {reply,{{From,ReplyAs},Reply}} when From =/= undefined ->
+            %% We take care of replies from puts here as well
+            io_reply(From, ReplyAs, Reply),
+            get_line_echo_off1({Chars,[]},Drv);
 	{'EXIT',Drv,interrupt} ->
 	    interrupted;
 	{'EXIT',Drv,_} ->
@@ -790,6 +824,10 @@ get_password1({Chars,[]}, Drv) ->
 	    %% set to []. But do we expect anything but plain output?
 
 	    get_password1({Chars, []}, Drv);
+        {reply,{{From,ReplyAs},Reply}} ->
+            %% We take care of replies from puts here as well
+            io_reply(From, ReplyAs, Reply),
+            get_password1({Chars, []},Drv);
 	{'EXIT',Drv,interrupt} ->
 	    interrupted;
 	{'EXIT',Drv,_} ->
diff --git a/lib/kernel/src/user_drv.erl b/lib/kernel/src/user_drv.erl
index a91c23539d..79f80f6604 100644
--- a/lib/kernel/src/user_drv.erl
+++ b/lib/kernel/src/user_drv.erl
@@ -29,6 +29,7 @@
 -define(OP_INSC,2).
 -define(OP_DELC,3).
 -define(OP_BEEP,4).
+-define(OP_PUTC_SYNC,5).
 % Control op
 -define(CTRL_OP_GET_WINSIZE,100).
 -define(CTRL_OP_GET_UNICODE_STATE,101).
@@ -133,7 +134,7 @@ server1(Iport, Oport, Shell) ->
 				      [erlang:system_info(system_version)]))},
 	       Iport, Oport),
     %% Enter the server loop.
-    server_loop(Iport, Oport, Curr, User, Gr).
+    server_loop(Iport, Oport, Curr, User, Gr, queue:new()).
 
 rem_sh_opts(Node) ->
     [{expand_fun,fun(B)-> rpc:call(Node,edlin_expand,expand,[B]) end}].
@@ -158,42 +159,46 @@ start_user() ->
 	    User
     end.
    
-server_loop(Iport, Oport, User, Gr) ->
+server_loop(Iport, Oport, User, Gr, IOQueue) ->
     Curr = gr_cur_pid(Gr),
     put(current_group, Curr),
-    server_loop(Iport, Oport, Curr, User, Gr).
+    server_loop(Iport, Oport, Curr, User, Gr, IOQueue).
 
-server_loop(Iport, Oport, Curr, User, Gr) ->
+server_loop(Iport, Oport, Curr, User, Gr, IOQueue) ->
     receive
 	{Iport,{data,Bs}} ->
 	    BsBin = list_to_binary(Bs),
 	    Unicode = unicode:characters_to_list(BsBin,utf8),
-	    port_bytes(Unicode, Iport, Oport, Curr, User, Gr);
+	    port_bytes(Unicode, Iport, Oport, Curr, User, Gr, IOQueue);
 	{Iport,eof} ->
 	    Curr ! {self(),eof},
-	    server_loop(Iport, Oport, Curr, User, Gr);
-	{User,Req} ->	% never block from user!
-	    io_request(Req, Iport, Oport),
-	    server_loop(Iport, Oport, Curr, User, Gr);
-	{Curr,tty_geometry} ->
-	    Curr ! {self(),tty_geometry,get_tty_geometry(Iport)},
-	    server_loop(Iport, Oport, Curr, User, Gr);
-	{Curr,get_unicode_state} ->
-	    Curr ! {self(),get_unicode_state,get_unicode_state(Iport)},
-	    server_loop(Iport, Oport, Curr, User, Gr);
-	{Curr,set_unicode_state, Bool} ->
-	    Curr ! {self(),set_unicode_state,set_unicode_state(Iport,Bool)},
-	    server_loop(Iport, Oport, Curr, User, Gr);
-	{Curr,Req} ->
-	    io_request(Req, Iport, Oport),
-	    server_loop(Iport, Oport, Curr, User, Gr);
+	    server_loop(Iport, Oport, Curr, User, Gr, IOQueue);
+        {User,Req} when element(1,Req) == tty_geometry;
+                        element(1,Req) == get_unicode_state;
+                        element(1,Req) == set_unicode_state ->
+            %% We ignore these requests from User, could be a bug?
+            server_loop(Iport, Oport, Curr, User, Gr, IOQueue);
+        Req when element(1,Req) =:= User orelse element(1,Req) =:= Curr,
+                 tuple_size(Req) =:= 2 orelse tuple_size(Req) =:= 3 ->
+            %% We match {User|Curr,_}|{User|Curr,_,_}
+            NewQ = handle_req(Req, Iport, Oport, IOQueue),
+            server_loop(Iport, Oport, Curr, User, Gr, NewQ);
+        {Oport,ok} ->
+            %% We get this ok from the port, in io_request we store
+            %% info about where to send reply at head of queue
+            {{value,{Origin,Reply}},ReplyQ} = queue:out(IOQueue),
+            Origin ! {reply,Reply},
+            NewQ = handle_req(next, Iport, Oport, ReplyQ),
+            server_loop(Iport, Oport, Curr, User, Gr, NewQ);
 	{'EXIT',Iport,_R} ->
-	    server_loop(Iport, Oport, Curr, User, Gr);
+	    server_loop(Iport, Oport, Curr, User, Gr, IOQueue);
 	{'EXIT',Oport,_R} ->
-	    server_loop(Iport, Oport, Curr, User, Gr);
+	    server_loop(Iport, Oport, Curr, User, Gr, IOQueue);
+        {'EXIT',User,shutdown} ->               % force data to port
+	    server_loop(Iport, Oport, Curr, User, Gr, IOQueue);
 	{'EXIT',User,_R} ->			% keep 'user' alive
 	    NewU = start_user(),
-	    server_loop(Iport, Oport, Curr, NewU, gr_set_num(Gr, 1, NewU, {}));
+	    server_loop(Iport, Oport, Curr, NewU, gr_set_num(Gr, 1, NewU, {}), IOQueue);
 	{'EXIT',Pid,R} ->			% shell and group leader exit
 	    case gr_cur_pid(Gr) of
 		Pid when R =/= die ,
@@ -213,18 +218,51 @@ server_loop(Iport, Oport, Curr, User, Gr) ->
 			    {ok,Gr2} = gr_set_cur(gr_set_num(Gr1, Ix, Pid1, 
 							     {shell,start,Params}), Ix),
 			    put(current_group, Pid1),
-			    server_loop(Iport, Oport, Pid1, User, Gr2);
+			    server_loop(Iport, Oport, Pid1, User, Gr2, IOQueue);
 			_ -> % remote shell
 			    io_requests([{put_chars,unicode,"(^G to start new job) ***\n"}],
 					Iport, Oport),	    
-			    server_loop(Iport, Oport, Curr, User, Gr1)
+			    server_loop(Iport, Oport, Curr, User, Gr1, IOQueue)
 		    end;
 		_ ->				% not current, just remove it
-		    server_loop(Iport, Oport, Curr, User, gr_del_pid(Gr, Pid))
+		    server_loop(Iport, Oport, Curr, User, gr_del_pid(Gr, Pid), IOQueue)
 	    end;
 	_X ->
 	    %% Ignore unknown messages.
-	    server_loop(Iport, Oport, Curr, User, Gr)
+	    server_loop(Iport, Oport, Curr, User, Gr, IOQueue)
+    end.
+
+%% We always handle geometry and unicode requests
+handle_req({Curr,tty_geometry},Iport,_Oport,IOQueue) ->
+    Curr ! {self(),tty_geometry,get_tty_geometry(Iport)},
+    IOQueue;
+handle_req({Curr,get_unicode_state},Iport,_Oport,IOQueue) ->
+    Curr ! {self(),get_unicode_state,get_unicode_state(Iport)},
+    IOQueue;
+handle_req({Curr,set_unicode_state, Bool},Iport,_Oport,IOQueue) ->
+    Curr ! {self(),set_unicode_state,set_unicode_state(Iport,Bool)},
+    IOQueue;
+handle_req(next,Iport,Oport,IOQueue) ->
+    case queue:out(IOQueue) of
+        {{value,Next},ExecQ} ->
+            NewQ = handle_req(Next,Iport,Oport,queue:new()),
+            queue:join(NewQ,ExecQ);
+        {empty,_} ->
+            IOQueue
+    end;
+handle_req(Msg,Iport,Oport,IOQueue) ->
+    case queue:peek(IOQueue) of
+        empty ->
+            {Origin,Req} = Msg,
+            case io_request(Req, Iport, Oport) of
+                ok -> IOQueue;
+                Reply ->
+                    %% Push reply info to front of queue
+                    queue:in_r({Origin,Reply},IOQueue)
+            end;
+        _Else ->
+            %% All requests are queued when we have outstanding sync put_chars
+            queue:in(Msg,IOQueue)
     end.
 
 %% port_bytes(Bytes, InPort, OutPort, CurrentProcess, UserProcess, Group)
@@ -232,34 +270,34 @@ server_loop(Iport, Oport, Curr, User, Gr) ->
 %%  either escape to switch_loop or restart the shell. Otherwise send 
 %%  the bytes to Curr.
 
-port_bytes([$\^G|_Bs], Iport, Oport, _Curr, User, Gr) ->
-    handle_escape(Iport, Oport, User, Gr);
+port_bytes([$\^G|_Bs], Iport, Oport, _Curr, User, Gr, IOQueue) ->
+    handle_escape(Iport, Oport, User, Gr, IOQueue);
 
-port_bytes([$\^C|_Bs], Iport, Oport, Curr, User, Gr) ->
-    interrupt_shell(Iport, Oport, Curr, User, Gr);
+port_bytes([$\^C|_Bs], Iport, Oport, Curr, User, Gr, IOQueue) ->
+    interrupt_shell(Iport, Oport, Curr, User, Gr, IOQueue);
 
-port_bytes([B], Iport, Oport, Curr, User, Gr) ->
+port_bytes([B], Iport, Oport, Curr, User, Gr, IOQueue) ->
     Curr ! {self(),{data,[B]}},
-    server_loop(Iport, Oport, Curr, User, Gr);
-port_bytes(Bs, Iport, Oport, Curr, User, Gr) ->
+    server_loop(Iport, Oport, Curr, User, Gr, IOQueue);
+port_bytes(Bs, Iport, Oport, Curr, User, Gr, IOQueue) ->
     case member($\^G, Bs) of
 	true ->
-	    handle_escape(Iport, Oport, User, Gr);
+	    handle_escape(Iport, Oport, User, Gr, IOQueue);
 	false ->
 	    Curr ! {self(),{data,Bs}},
-	    server_loop(Iport, Oport, Curr, User, Gr)
+	    server_loop(Iport, Oport, Curr, User, Gr, IOQueue)
     end.
 
-interrupt_shell(Iport, Oport, Curr, User, Gr) ->
+interrupt_shell(Iport, Oport, Curr, User, Gr, IOQueue) ->
     case gr_get_info(Gr, Curr) of
 	undefined -> 
 	    ok;					% unknown
 	_ ->
 	    exit(Curr, interrupt)
     end,
-    server_loop(Iport, Oport, Curr, User, Gr).
+    server_loop(Iport, Oport, Curr, User, Gr, IOQueue).
 
-handle_escape(Iport, Oport, User, Gr) ->
+handle_escape(Iport, Oport, User, Gr, IOQueue) ->
     case application:get_env(stdlib, shell_esc) of
 	{ok,abort} ->
 	    Pid = gr_cur_pid(Gr),
@@ -278,11 +316,11 @@ handle_escape(Iport, Oport, User, Gr) ->
 	    Pid1 = group:start(self(), {shell,start,[]}),
 	    io_request({put_chars,unicode,"\n"}, Iport, Oport),
 	    server_loop(Iport, Oport, User, 
-			gr_add_cur(Gr1, Pid1, {shell,start,[]}));
+			gr_add_cur(Gr1, Pid1, {shell,start,[]}), IOQueue);
 
 	_ ->					% {ok,jcl} | undefined
 	    io_request({put_chars,unicode,"\nUser switch command\n"}, Iport, Oport),
-	    server_loop(Iport, Oport, User, switch_loop(Iport, Oport, Gr))
+	    server_loop(Iport, Oport, User, switch_loop(Iport, Oport, Gr), IOQueue)
     end.
 
 switch_loop(Iport, Oport, Gr) ->
@@ -492,9 +530,12 @@ set_unicode_state(Iport, Bool) ->
 
 io_request(Request, Iport, Oport) ->
     try io_command(Request) of
-        Command ->
+        {command,_} = Command ->
             Oport ! {self(),Command},
-            ok
+            ok;
+        {Command,Reply} ->
+            Oport ! {self(),Command},
+            Reply
     catch
         {requests,Rs} ->
             io_requests(Rs, Iport, Oport);
@@ -511,6 +552,13 @@ io_requests([], _Iport, _Oport) ->
 put_int16(N, Tail) ->
     [(N bsr 8)band 255,N band 255|Tail].
 
+%% When a put_chars_sync command is used, user_drv guarantees that
+%% the bytes have been put in the buffer of the port before an acknowledgement
+%% is sent back to the process sending the request. This command was added in
+%% OTP 18 to make sure that data sent from io:format is actually printed
+%% to the console before the vm stops when calling erlang:halt(integer()).
+io_command({put_chars_sync, unicode,Cs,Reply}) ->
+    {{command,[?OP_PUTC_SYNC|unicode:characters_to_binary(Cs,utf8)]},Reply};
 io_command({put_chars, unicode,Cs}) ->
     {command,[?OP_PUTC|unicode:characters_to_binary(Cs,utf8)]};
 io_command({move_rel,N}) ->
diff --git a/lib/ssh/src/ssh_cli.erl b/lib/ssh/src/ssh_cli.erl
index 77453e8fd7..a1abf6c851 100644
--- a/lib/ssh/src/ssh_cli.erl
+++ b/lib/ssh/src/ssh_cli.erl
@@ -98,7 +98,7 @@ handle_ssh_msg({ssh_cm, ConnectionHandler,
     Pty = Pty0#ssh_pty{width = Width, height = Height,
 		       pixel_width = PixWidth,
 		       pixel_height = PixHeight},
-    {Chars, NewBuf} = io_request({window_change, Pty0}, Buf, Pty),
+    {Chars, NewBuf} = io_request({window_change, Pty0}, Buf, Pty, undefined),
     write_chars(ConnectionHandler, ChannelId, Chars),
     {ok, State#state{pty = Pty, buf = NewBuf}};
 
@@ -188,7 +188,7 @@ handle_msg({Group, tty_geometry}, #state{group = Group,
 handle_msg({Group, Req}, #state{group = Group, buf = Buf, pty = Pty,
 				 cm = ConnectionHandler,
 				 channel = ChannelId} = State) ->
-    {Chars, NewBuf} = io_request(Req, Buf, Pty),
+    {Chars, NewBuf} = io_request(Req, Buf, Pty, Group),
     write_chars(ConnectionHandler, ChannelId, Chars),
     {ok, State#state{buf = NewBuf}};
 
@@ -263,40 +263,49 @@ eval(Error) ->
 %%% displaying device...
 %%% We are *not* really unicode aware yet, we just filter away characters 
 %%% beyond the latin1 range. We however handle the unicode binaries...
-io_request({window_change, OldTty}, Buf, Tty) ->
+io_request({window_change, OldTty}, Buf, Tty, _Group) ->
     window_change(Tty, OldTty, Buf);
-io_request({put_chars, Cs}, Buf, Tty) ->
+io_request({put_chars, Cs}, Buf, Tty, _Group) ->
     put_chars(bin_to_list(Cs), Buf, Tty);
-io_request({put_chars, unicode, Cs}, Buf, Tty) ->
+io_request({put_chars, unicode, Cs}, Buf, Tty, _Group) ->
     put_chars(unicode:characters_to_list(Cs,unicode), Buf, Tty);
-io_request({insert_chars, Cs}, Buf, Tty) ->
+io_request({insert_chars, Cs}, Buf, Tty, _Group) ->
     insert_chars(bin_to_list(Cs), Buf, Tty);
-io_request({insert_chars, unicode, Cs}, Buf, Tty) ->
+io_request({insert_chars, unicode, Cs}, Buf, Tty, _Group) ->
     insert_chars(unicode:characters_to_list(Cs,unicode), Buf, Tty);
-io_request({move_rel, N}, Buf, Tty) ->
+io_request({move_rel, N}, Buf, Tty, _Group) ->
     move_rel(N, Buf, Tty);
-io_request({delete_chars,N}, Buf, Tty) ->
+io_request({delete_chars,N}, Buf, Tty, _Group) ->
     delete_chars(N, Buf, Tty);
-io_request(beep, Buf, _Tty) ->
+io_request(beep, Buf, _Tty, _Group) ->
     {[7], Buf};
 
 %% New in R12
-io_request({get_geometry,columns},Buf,Tty) ->
+io_request({get_geometry,columns},Buf,Tty, _Group) ->
     {ok, Tty#ssh_pty.width, Buf};
-io_request({get_geometry,rows},Buf,Tty) ->
+io_request({get_geometry,rows},Buf,Tty, _Group) ->
     {ok, Tty#ssh_pty.height, Buf};
-io_request({requests,Rs}, Buf, Tty) ->
-    io_requests(Rs, Buf, Tty, []);
-io_request(tty_geometry, Buf, Tty) ->
-    io_requests([{move_rel, 0}, {put_chars, unicode, [10]}], Buf, Tty, []);
+io_request({requests,Rs}, Buf, Tty, Group) ->
+    io_requests(Rs, Buf, Tty, [], Group);
+io_request(tty_geometry, Buf, Tty, Group) ->
+    io_requests([{move_rel, 0}, {put_chars, unicode, [10]}],
+                Buf, Tty, [], Group);
      %{[], Buf};
-io_request(_R, Buf, _Tty) ->
+
+%% New in 18
+io_request({put_chars_sync, Class, Cs, Reply}, Buf, Tty, Group) ->
+    %% We handle these asynchronous for now, if we need output guarantees
+    %% we have to handle these synchronously
+    Group ! {reply, Reply},
+    io_request({put_chars, Class, Cs}, Buf, Tty, Group);
+
+io_request(_R, Buf, _Tty, _Group) ->
     {[], Buf}.
 
-io_requests([R|Rs], Buf, Tty, Acc) ->
-    {Chars, NewBuf} = io_request(R, Buf, Tty),
-    io_requests(Rs, NewBuf, Tty, [Acc|Chars]);
-io_requests([], Buf, _Tty, Acc) ->
+io_requests([R|Rs], Buf, Tty, Acc, Group) ->
+    {Chars, NewBuf} = io_request(R, Buf, Tty, Group),
+    io_requests(Rs, NewBuf, Tty, [Acc|Chars], Group);
+io_requests([], Buf, _Tty, Acc, _Group) ->
     {Acc, Buf}.
 
 %%% return commands for cursor navigation, assume everything is ansi