don't start the transport connecting to the other nodes until after the startup event script has run

(This used to be ctdb commit afca3cc74211aa2e18b1f74d36b2add8dffcfdc7)

6 files changed, 168 insertions, 26 deletions

diff --git a/ctdb/common/ctdb_daemon.c b/ctdb/common/ctdb_daemon.c
index 015b8a1868..1ffd70f61f 100644
--- a/ctdb/common/ctdb_daemon.c
+++ b/ctdb/common/ctdb_daemon.c
@@ -31,6 +31,28 @@
 
 static void daemon_incoming_packet(void *, struct ctdb_req_header *);
 
+/* called when the "startup" event script has finished */
+static void ctdb_start_transport(struct ctdb_context *ctdb, int status)
+{
+	if (status != 0) {
+		DEBUG(0,("startup event failed!\n"));
+		ctdb_fatal(ctdb, "startup event script failed");		
+	}
+
+	/* start the transport running */
+	if (ctdb->methods->start(ctdb) != 0) {
+		DEBUG(0,("transport failed to start!\n"));
+		ctdb_fatal(ctdb, "transport failed to start");
+	}
+
+	/* start the recovery daemon process */
+	if (ctdb_start_recoverd(ctdb) != 0) {
+		DEBUG(0,("Failed to start recovery daemon\n"));
+		exit(11);
+	}
+}
+
+/* go into main ctdb loop */
 static void ctdb_main_loop(struct ctdb_context *ctdb)
 {
 	int ret = -1;
@@ -50,16 +72,10 @@ static void ctdb_main_loop(struct ctdb_context *ctdb)
 		return;
 	}
 
-	ret = ctdb_event_script(ctdb, "startup");
-	if (ret != 0) {
-		DEBUG(0,("Failed startup event script\n"));
-		return;
-	}
-
-	/* start the transport running */
-	if (ctdb->methods->start(ctdb) != 0) {
-		DEBUG(0,("transport failed to start!\n"));
-		ctdb_fatal(ctdb, "transport failed to start");
+	/* initialise the transport  */
+	if (ctdb->methods->initialise(ctdb) != 0) {
+		DEBUG(0,("transport failed to initialise!\n"));
+		ctdb_fatal(ctdb, "transport failed to initialise");
 	}
 
 	/* tell all other nodes we've just started up */
@@ -68,6 +84,12 @@ static void ctdb_main_loop(struct ctdb_context *ctdb)
 				 CTDB_CTRL_FLAG_NOREPLY,
 				 tdb_null, NULL, NULL);
 
+	ret = ctdb_event_script_callback(ctdb, ctdb_start_transport, "startup");
+	if (ret != 0) {
+		DEBUG(0,("Failed startup event script\n"));
+		return;
+	}
+
 	/* go into a wait loop to allow other nodes to complete */
 	event_loop_wait(ctdb->ev);
 
@@ -744,12 +766,6 @@ int ctdb_start_daemon(struct ctdb_context *ctdb, bool do_fork)
 	/* try to set us up as realtime */
 	ctdb_set_realtime();
 
-	/* start the recovery daemon process */
-	if (ctdb_start_recoverd(ctdb) != 0) {
-		DEBUG(0,("Failed to start recovery daemon\n"));
-		exit(11);
-	}
-
 	/* ensure the socket is deleted on exit of the daemon */
 	domain_socket_name = talloc_strdup(talloc_autofree_context(), ctdb->daemon.name);
 	talloc_set_destructor(domain_socket_name, unlink_destructor);	
diff --git a/ctdb/common/ctdb_recoverd.c b/ctdb/common/ctdb_recoverd.c
index dca340bff2..b05b02050f 100644
--- a/ctdb/common/ctdb_recoverd.c
+++ b/ctdb/common/ctdb_recoverd.c
@@ -972,6 +972,10 @@ int ctdb_start_recoverd(struct ctdb_context *ctdb)
 
 	close(fd[1]);
 
+	/* we need a new event context */
+	talloc_free(ctdb->ev);
+	ctdb->ev = event_context_init(ctdb);
+
 	event_add_fd(ctdb->ev, ctdb, fd[0], EVENT_FD_READ|EVENT_FD_AUTOCLOSE, 
 		     ctdb_recoverd_parent, &fd[0]);	
 
diff --git a/ctdb/ib/ibw_ctdb_init.c b/ctdb/ib/ibw_ctdb_init.c
index 66919568df..b41c61687b 100644
--- a/ctdb/ib/ibw_ctdb_init.c
+++ b/ctdb/ib/ibw_ctdb_init.c
@@ -70,9 +70,9 @@ static int ctdb_ibw_add_node(struct ctdb_node *node)
 }
 
 /*
- * Start infiniband
+ * initialise infiniband
  */
-static int ctdb_ibw_start(struct ctdb_context *ctdb)
+static int ctdb_ibw_initialise(struct ctdb_context *ctdb)
 {
 	int i, ret;
 
@@ -88,6 +88,17 @@ static int ctdb_ibw_start(struct ctdb_context *ctdb)
 		}
 	}
 
+	return 0;
+}
+
+
+/*
+ * Start infiniband
+ */
+static int ctdb_ibw_start(struct ctdb_context *ctdb)
+{
+	int i, ret;
+
 	/* listen on our own address */
 	if (ctdb_ibw_listen(ctdb, 10)) /* TODO: backlog as param */
 		return -1;
@@ -190,6 +201,7 @@ static int ctdb_ibw_stop(struct ctdb_context *cctx)
 #endif /* __NOTDEF__ */
 
 static const struct ctdb_methods ctdb_ibw_methods = {
+	.initialise= ctdb_ibw_initialise,
 	.start     = ctdb_ibw_start,
 	.queue_pkt = ctdb_ibw_queue_pkt,
 	.add_node = ctdb_ibw_add_node,
diff --git a/ctdb/include/ctdb_private.h b/ctdb/include/ctdb_private.h
index 37176748e9..92cc1a4263 100644
--- a/ctdb/include/ctdb_private.h
+++ b/ctdb/include/ctdb_private.h
@@ -129,6 +129,7 @@ struct ctdb_node {
   transport specific methods
 */
 struct ctdb_methods {
+	int (*initialise)(struct ctdb_context *); /* initialise transport structures */	
 	int (*start)(struct ctdb_context *); /* start protocol processing */	
 	int (*add_node)(struct ctdb_node *); /* setup a new node */	
 	int (*queue_pkt)(struct ctdb_node *, uint8_t *data, uint32_t length);
@@ -949,6 +950,9 @@ int32_t ctdb_control_startup(struct ctdb_context *ctdb, uint32_t vnn);
 
 void ctdb_takeover_client_destructor_hook(struct ctdb_client *client);
 int ctdb_event_script(struct ctdb_context *ctdb, const char *fmt, ...) PRINTF_ATTRIBUTE(2,3);
+int ctdb_event_script_callback(struct ctdb_context *ctdb, 
+			       void (*callback)(struct ctdb_context *, int),
+			       const char *fmt, ...) PRINTF_ATTRIBUTE(3,4);
 void ctdb_release_all_ips(struct ctdb_context *ctdb);
 
 
diff --git a/ctdb/takeover/system.c b/ctdb/takeover/system.c
index c6624f2485..a1b6da6cf5 100644
--- a/ctdb/takeover/system.c
+++ b/ctdb/takeover/system.c
@@ -24,6 +24,7 @@
 #include "system/filesys.h"
 #include "system/wait.h"
 #include "../include/ctdb_private.h"
+#include "lib/events/events.h"
 #include <net/ethernet.h>
 #include <net/if_arp.h>
 
@@ -258,17 +259,24 @@ bool ctdb_sys_have_ip(const char *ip)
 }
 
 /*
-  run the event script
+  run the event script - varargs version
  */
-int ctdb_event_script(struct ctdb_context *ctdb, const char *fmt, ...)
+static int ctdb_event_script_v(struct ctdb_context *ctdb, const char *fmt, va_list ap)
 {
-	va_list ap;
 	char *options, *cmdstr;
 	int ret;
+	va_list ap2;
+	struct stat st;
 
-	va_start(ap, fmt);
-	options  = talloc_vasprintf(ctdb, fmt, ap);
-	va_end(ap);
+	if (stat(ctdb->takeover.event_script, &st) != 0 && 
+	    errno == ENOENT) {
+		DEBUG(0,("No event script found at '%s'\n", ctdb->takeover.event_script));
+		return 0;
+	}
+
+	va_copy(ap2, ap);
+	options  = talloc_vasprintf(ctdb, fmt, ap2);
+	va_end(ap2);
 	CTDB_NO_MEMORY(ctdb, options);
 
 	cmdstr = talloc_asprintf(ctdb, "%s %s", ctdb->takeover.event_script, options);
@@ -285,3 +293,90 @@ int ctdb_event_script(struct ctdb_context *ctdb, const char *fmt, ...)
 	return ret;
 }
 
+/*
+  run the event script
+ */
+int ctdb_event_script(struct ctdb_context *ctdb, const char *fmt, ...)
+{
+	va_list ap;
+	int ret;
+
+	va_start(ap, fmt);
+	ret = ctdb_event_script_v(ctdb, fmt, ap);
+	va_end(ap);
+
+	return ret;
+}
+
+
+struct ctdb_event_script_state {
+	struct ctdb_context *ctdb;
+	pid_t child;
+	void (*callback)(struct ctdb_context *, int);
+	int fd[2];
+};
+
+/* called when child is finished */
+static void ctdb_event_script_handler(struct event_context *ev, struct fd_event *fde, 
+				      uint16_t flags, void *p)
+{
+	struct ctdb_event_script_state *state = 
+		talloc_get_type(p, struct ctdb_event_script_state);
+	int status = -1;
+	waitpid(state->child, &status, 0);
+	if (status != -1) {
+		status = WEXITSTATUS(status);
+	}
+	state->callback(state->ctdb, status);
+	talloc_free(state);
+}
+
+
+/*
+  run the event script in the background, calling the callback when 
+  finished
+ */
+int ctdb_event_script_callback(struct ctdb_context *ctdb, 
+			       void (*callback)(struct ctdb_context *, int),
+			       const char *fmt, ...)
+{
+	struct ctdb_event_script_state *state;
+	va_list ap;
+	int ret;
+
+	state = talloc(ctdb, struct ctdb_event_script_state);
+	CTDB_NO_MEMORY(ctdb, state);
+
+	state->ctdb = ctdb;
+	state->callback = callback;
+	
+	ret = pipe(state->fd);
+	if (ret != 0) {
+		talloc_free(state);
+		return -1;
+	}
+
+	state->child = fork();
+
+	if (state->child == (pid_t)-1) {
+		close(state->fd[0]);
+		close(state->fd[1]);
+		talloc_free(state);
+		return -1;
+	}
+
+	if (state->child == 0) {
+		close(state->fd[0]);
+		va_start(ap, fmt);
+		ret = ctdb_event_script_v(ctdb, fmt, ap);
+		va_end(ap);
+		_exit(ret);
+	}
+
+	close(state->fd[1]);
+
+	event_add_fd(ctdb->ev, state, state->fd[0], EVENT_FD_READ|EVENT_FD_AUTOCLOSE,
+		     ctdb_event_script_handler, state);
+
+	return 0;
+}
diff --git a/ctdb/tcp/tcp_init.c b/ctdb/tcp/tcp_init.c
index 35b43bd5e2..665336e947 100644
--- a/ctdb/tcp/tcp_init.c
+++ b/ctdb/tcp/tcp_init.c
@@ -46,9 +46,9 @@ static int ctdb_tcp_add_node(struct ctdb_node *node)
 }
 
 /*
-  start the protocol going
+  initialise transport structures
 */
-static int ctdb_tcp_start(struct ctdb_context *ctdb)
+static int ctdb_tcp_initialise(struct ctdb_context *ctdb)
 {
 	int i;
 
@@ -59,6 +59,16 @@ static int ctdb_tcp_start(struct ctdb_context *ctdb)
 		}
 	}
 	
+	return 0;
+}
+
+/*
+  start the protocol going
+*/
+static int ctdb_tcp_start(struct ctdb_context *ctdb)
+{
+	int i;
+
 	/* listen on our own address */
 	if (ctdb_tcp_listen(ctdb) != 0) return -1;
 
@@ -90,6 +100,7 @@ static void *ctdb_tcp_allocate_pkt(TALLOC_CTX *mem_ctx, size_t size)
 
 
 static const struct ctdb_methods ctdb_tcp_methods = {
+	.initialise   = ctdb_tcp_initialise,
 	.start        = ctdb_tcp_start,
 	.queue_pkt    = ctdb_tcp_queue_pkt,
 	.add_node     = ctdb_tcp_add_node,