erts, heart: Ensure erl_crash.dump is written

When a crash dump is about to be written and we have
heartbeat enabled on a system. We need time to write it
before heart explicitly kills the beam.

6 files changed, 138 insertions, 73 deletions

diff --git a/erts/emulator/beam/atom.names b/erts/emulator/beam/atom.names
index 106fad030b..afcbd732df 100644
--- a/erts/emulator/beam/atom.names
+++ b/erts/emulator/beam/atom.names
@@ -252,6 +252,7 @@ atom heap_block_size
 atom heap_size
 atom heap_sizes
 atom heap_type
+atom heart_port
 atom heir
 atom hidden
 atom hide
diff --git a/erts/emulator/beam/global.h b/erts/emulator/beam/global.h
index dbf95f5bd7..2c20e3da3b 100755
--- a/erts/emulator/beam/global.h
+++ b/erts/emulator/beam/global.h
@@ -983,6 +983,9 @@ Uint erts_port_ioq_size(Port *pp);
 void erts_stale_drv_select(Eterm, ErlDrvEvent, int, int);
 void erts_port_cleanup(Port *);
 void erts_fire_port_monitor(Port *prt, Eterm ref);
+
+Port *erts_get_heart_port(void);
+
 #ifdef ERTS_SMP
 void erts_smp_xports_unlock(Port *);
 #endif
diff --git a/erts/emulator/beam/io.c b/erts/emulator/beam/io.c
index 35b194f927..502ded4eca 100644
--- a/erts/emulator/beam/io.c
+++ b/erts/emulator/beam/io.c
@@ -5235,3 +5235,24 @@ erl_drv_getenv(char *key, char *value, size_t *value_size)
 {
     return erts_sys_getenv_raw(key, value, value_size);
 }
+
+/* get heart_port
+ * used by erl_crash_dump
+ * - uses the fact that heart_port is registered when starting heart
+ */
+
+Port *erts_get_heart_port() {
+
+    Port* port;
+    Uint  ix;
+
+    for(ix = 0; ix < erts_max_ports; ix++) {
+	port = &erts_port[ix];
+	/* immediate compare */
+	if (port->reg && port->reg->name == am_heart_port) {
+	    return port;
+	}
+    }
+
+    return NULL;
+}
diff --git a/erts/emulator/sys/unix/sys.c b/erts/emulator/sys/unix/sys.c
index c1fa00b4ea..37dfcb1dd4 100644
--- a/erts/emulator/sys/unix/sys.c
+++ b/erts/emulator/sys/unix/sys.c
@@ -58,7 +58,6 @@
 #define __DARWIN__ 1
 #endif
 
-
 #ifdef USE_THREADS
 #include "erl_threads.h"
 #endif
@@ -71,7 +70,6 @@ static erts_smp_rwmtx_t environ_rwmtx;
 #define MAX_VSIZE 16		/* Max number of entries allowed in an I/O
 				 * vector sock_sendv().
 				 */
-
 /*
  * Don't need global.h, but bif_table.h (included by bif.h),
  * won't compile otherwise
@@ -123,6 +121,15 @@ struct ErtsSysReportExit_ {
 #endif
 };
 
+/* This data is shared by these drivers - initialized by spawn_init() */
+static struct driver_data {
+    int port_num, ofd, packet_bytes;
+    ErtsSysReportExit *report_exit;
+    int pid;
+    int alive;
+    int status;
+} *driver_data;			/* indexed by fd */
+
 static ErtsSysReportExit *report_exit_list;
 #if CHLDWTHR && !defined(ERTS_SMP)
 static ErtsSysReportExit *report_exit_transit_list;
@@ -685,12 +692,33 @@ prepare_crash_dump(void)
     int i, max;
     char env[21]; /* enough to hold any 64-bit integer */
     size_t envsz;
+    Port *heart_port;
+    Eterm heap[3];
+    Eterm *hp = heap;
+    Eterm list = NIL;
+
+    int heart_fd[2] = {-1,-1};
 
     if (ERTS_PREPARED_CRASH_DUMP)
 	return; /* We have already been called */
 
+    heart_port = erts_get_heart_port();
+    if (heart_port) {
+	/* hearts input fd
+	 * We "know" drv_data is the in_fd since the port is started with read|write
+	 */
+	heart_fd[0] = (int)heart_port->drv_data;
+	heart_fd[1] = (int)driver_data[heart_fd[0]].ofd;
+
+	list = CONS(hp, make_small(8), list); hp += 2;
+
+	/* send to heart port, CMD = 8, i.e. prepare crash dump =o */
+	erts_write_to_port(NIL, heart_port, list);
+    }
+
     /* Make sure we unregister at epmd (unknown fd) and get at least
        one free filedescriptor (for erl_crash.dump) */
+
     max = max_files;
     if (max < 1024)
 	max = 1024;
@@ -704,6 +732,10 @@ prepare_crash_dump(void)
 	if (i == async_fd[0] || i == async_fd[1])
 	    continue;
 #endif
+	/* We don't want to close our heart yet ... */
+	if (i == heart_fd[0] || i == heart_fd[1])
+	    continue;
+
 	close(i);
     }
 
@@ -725,7 +757,6 @@ prepare_crash_dump(void)
 	sec = (unsigned) i != 0 ? 0 : atoi(env);
 	alarm(sec);
     }
-
 }
 
 void
@@ -1021,15 +1052,6 @@ void fini_getenv_state(GETENV_STATE *state)
 
 #define ERTS_SYS_READ_BUF_SZ (64*1024)
 
-/* This data is shared by these drivers - initialized by spawn_init() */
-static struct driver_data {
-    int port_num, ofd, packet_bytes;
-    ErtsSysReportExit *report_exit;
-    int pid;
-    int alive;
-    int status;
-} *driver_data;			/* indexed by fd */
-
 /* Driver interfaces */
 static ErlDrvData spawn_start(ErlDrvPort, char*, SysDriverOpts*);
 static ErlDrvData fd_start(ErlDrvPort, char*, SysDriverOpts*);
diff --git a/erts/emulator/sys/win32/sys.c b/erts/emulator/sys/win32/sys.c
index 6894d682e7..c5664d8e8a 100755
--- a/erts/emulator/sys/win32/sys.c
+++ b/erts/emulator/sys/win32/sys.c
@@ -258,8 +258,22 @@ void erl_sys_args(int* argc, char** argv)
 void
 erts_sys_prepare_crash_dump(void)
 {
+    Port *heart_port;
+    Eterm heap[3];
+    Eterm *hp = heap;
+    Eterm list = NIL;
+
+    heart_port = erts_get_heart_port();
+
+    if (heart_port) {
+
+	list = CONS(hp, make_small(8), list); hp += 2;
+
+	/* send to heart port, CMD = 8, i.e. prepare crash dump =o */
+	erts_write_to_port(NIL, heart_port, list);
+    }
+
     /* Windows - free file descriptors are hopefully available */
-    return;
 }
 
 static void
diff --git a/erts/etc/common/heart.c b/erts/etc/common/heart.c
index 70c2b3bb23..c6ce017d8c 100644
--- a/erts/etc/common/heart.c
+++ b/erts/etc/common/heart.c
@@ -153,13 +153,14 @@ struct msg {
 };
 
 /* operations */
-#define  HEART_ACK    1
-#define  HEART_BEAT   2
-#define  SHUT_DOWN    3
-#define  SET_CMD      4
-#define  CLEAR_CMD    5
-#define  GET_CMD      6
-#define  HEART_CMD    7
+#define  HEART_ACK       (1)
+#define  HEART_BEAT      (2)
+#define  SHUT_DOWN       (3)
+#define  SET_CMD         (4)
+#define  CLEAR_CMD       (5)
+#define  GET_CMD         (6)
+#define  HEART_CMD       (7)
+#define  PREPARING_CRASH (8)
 
 
 /*  Maybe interesting to change */
@@ -187,10 +188,11 @@ unsigned long heart_beat_kill_pid = 0;
 #define SOL_WD_TIMEOUT (heart_beat_timeout+heart_beat_boot_delay)
 
 /* reasons for reboot */
-#define  R_TIMEOUT          1
-#define  R_CLOSED           2
-#define  R_ERROR            3
-#define  R_SHUT_DOWN        4
+#define  R_TIMEOUT          (1)
+#define  R_CLOSED           (2)
+#define  R_ERROR            (3)
+#define  R_SHUT_DOWN        (4)
+#define  R_CRASHING         (5) /* Doing a crash dump and we will wait for it */
 
 
 /*  macros */
@@ -200,8 +202,8 @@ unsigned long heart_beat_kill_pid = 0;
 
 /*  prototypes */
 
-static int message_loop(int,int);
-static void do_terminate(int);
+static int message_loop(int, int);
+static void do_terminate(int, int);
 static int notify_ack(int);
 static int heart_cmd_reply(int, char *);
 static int write_message(int, struct msg *);
@@ -376,7 +378,7 @@ main(int argc, char **argv)
     program_name[sizeof(program_name)-1] = '\0';
     notify_ack(erlout_fd);
     cmd[0] = '\0';
-    do_terminate(message_loop(erlin_fd,erlout_fd));
+    do_terminate(erlin_fd,message_loop(erlin_fd,erlout_fd));
     return 0;
 }
 
@@ -410,6 +412,7 @@ message_loop(erlin_fd, erlout_fd)
 #endif
 
   while (1) {
+      /* REFACTOR: below to select/tmo function */
 #ifdef __WIN32__
 	wresult = WaitForSingleObject(hevent_dataready,SELECT_TIMEOUT*1000+ 2);
 	if (wresult == WAIT_FAILED) {
@@ -504,6 +507,10 @@ message_loop(erlin_fd, erlout_fd)
 				    free_env_val(env);
 				}
 			        break;
+			case PREPARING_CRASH:
+				/* Erlang has reached a crushdump point (is crashing for sure) */
+				print_error("Erlang is crashing .. (waiting for crash dump file)");
+				return R_CRASHING;
 			default:
 				/* ignore all other messages */
 				break;
@@ -635,69 +642,66 @@ void win_system(char *command)
  * do_terminate
  */
 static void 
-do_terminate(reason)
-  int reason;
-{
+do_terminate(int erlin_fd, int reason) {
   /*
     When we get here, we have HEART_BEAT_BOOT_DELAY secs to finish
     (plus heart_beat_report_delay if under VxWorks), so we don't need
     to call wd_reset().
     */
-  
+  struct msg message;
+
   switch (reason) {
   case R_SHUT_DOWN:
     break;
+  case R_CRASHING:
+    print_error("Waiting for dump");
+    read_message(erlin_fd, &message); /* read until we get something */
   case R_TIMEOUT:
-  case R_ERROR:
   case R_CLOSED:
+  case R_ERROR:
   default:
-#if defined(__WIN32__) /* Not VxWorks */
     {
-      if(!cmd[0]) {
-	char *command = get_env(HEART_COMMAND_ENV);
-	if(!command)
-	  print_error("Would reboot. Terminating.");
-	else {
-	  kill_old_erlang();
-	  /* High prio combined with system() works badly indeed... */
-	  SetPriorityClass(GetCurrentProcess(), NORMAL_PRIORITY_CLASS);
-	  win_system(command);
-	  print_error("Executed \"%s\". Terminating.",command);
+#if defined(__WIN32__) /* Not VxWorks */
+	if(!cmd[0]) {
+	    char *command = get_env(HEART_COMMAND_ENV);
+	    if(!command)
+		print_error("Would reboot. Terminating.");
+	    else {
+		kill_old_erlang();
+		/* High prio combined with system() works badly indeed... */
+		SetPriorityClass(GetCurrentProcess(), NORMAL_PRIORITY_CLASS);
+		win_system(command);
+		print_error("Executed \"%s\". Terminating.",command);
+	    }
+	    free_env_val(command);
+	} else {
+	    kill_old_erlang();
+	    /* High prio combined with system() works badly indeed... */
+	    SetPriorityClass(GetCurrentProcess(), NORMAL_PRIORITY_CLASS);
+	    win_system(&cmd[0]);
+	    print_error("Executed \"%s\". Terminating.",cmd);
 	}
-	free_env_val(command);
-      }
-      else {
-	kill_old_erlang();
-	/* High prio combined with system() works badly indeed... */
-	SetPriorityClass(GetCurrentProcess(), NORMAL_PRIORITY_CLASS);
-	win_system(&cmd[0]);
-	print_error("Executed \"%s\". Terminating.",cmd);
-      }
-    }
-
 #else
-    {
-      if(!cmd[0]) {
-	char *command = get_env(HEART_COMMAND_ENV);
-	if(!command)
-	  print_error("Would reboot. Terminating.");
-	else {
-	  kill_old_erlang();
-	  /* suppress gcc warning with 'if' */
-	  if(system(command));
-	  print_error("Executed \"%s\". Terminating.",command);
+	if(!cmd[0]) {
+	    char *command = get_env(HEART_COMMAND_ENV);
+	    if(!command)
+		print_error("Would reboot. Terminating.");
+	    else {
+		kill_old_erlang();
+		/* suppress gcc warning with 'if' */
+		if(system(command));
+		print_error("Executed \"%s\". Terminating.",command);
+	    }
+	    free_env_val(command);
+	} else {
+	    kill_old_erlang();
+	    /* suppress gcc warning with 'if' */
+	    if(system((char*)&cmd[0]));
+	    print_error("Executed \"%s\". Terminating.",cmd);
 	}
-	free_env_val(command);
-      }
-      else {
-	kill_old_erlang();
-	/* suppress gcc warning with 'if' */
-	if(system((char*)&cmd[0]));
-	print_error("Executed \"%s\". Terminating.",cmd);
-      }
+#endif
     }
     break;
-#endif
   } /* switch(reason) */
 }