source4 smbd prefork: restart on non zero exit code

Restart any pre-fork master or worker process that exits with a non
zero exit code.

Signed-off-by: Gary Lockyer <gary@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

7 files changed, 119 insertions, 68 deletions

diff --git a/source4/kdc/kdc-heimdal.c b/source4/kdc/kdc-heimdal.c
index 67c07c3d81a..b5de5a790d4 100644
--- a/source4/kdc/kdc-heimdal.c
+++ b/source4/kdc/kdc-heimdal.c
@@ -276,7 +276,8 @@ static NTSTATUS kdc_task_init(struct task_server *task)
 		return NT_STATUS_INVALID_DOMAIN_ROLE;
 	case ROLE_DOMAIN_PDC:
 	case ROLE_DOMAIN_BDC:
-		task_server_terminate(task, "Cannot start KDC as a 'classic Samba' DC", true);
+		task_server_terminate(
+		    task, "Cannot start KDC as a 'classic Samba' DC", false);
 		return NT_STATUS_INVALID_DOMAIN_ROLE;
 	case ROLE_ACTIVE_DIRECTORY_DC:
 		/* Yes, we want a KDC */
diff --git a/source4/smbd/process_model.h b/source4/smbd/process_model.h
index 2a226cef8cb..2892dd43c04 100644
--- a/source4/smbd/process_model.h
+++ b/source4/smbd/process_model.h
@@ -65,12 +65,17 @@ struct model_ops {
 			 const struct service_details*,
 			 const int);
 
-	/* function to terminate a connection or task */
-	void (*terminate)(struct tevent_context *,
-			  struct loadparm_context *lp_ctx,
-			  const char *reason,
-			  bool fatal,
-			  void *process_context);
+	/* function to terminate a task */
+	void (*terminate_task)(struct tevent_context *,
+			       struct loadparm_context *lp_ctx,
+			       const char *reason,
+			       bool fatal,
+			       void *process_context);
+	/* function to terminate a connection */
+	void (*terminate_connection)(struct tevent_context *,
+				     struct loadparm_context *lp_ctx,
+				     const char *reason,
+				     void *process_context);
 
 	/* function to set a title for the connection or task */
 	void (*set_title)(struct tevent_context *, const char *title);
diff --git a/source4/smbd/process_prefork.c b/source4/smbd/process_prefork.c
index c470820efb8..d2627937b44 100644
--- a/source4/smbd/process_prefork.c
+++ b/source4/smbd/process_prefork.c
@@ -147,6 +147,33 @@ static void prefork_pipe_handler(struct tevent_context *event_ctx,
 	exit(0);
 }
 
+static void prefork_restart(struct tevent_context *ev,
+			    struct restart_context *rc)
+{
+	if (rc->master != NULL) {
+		DBG_ERR("Restarting [%s] pre-fork master\n", rc->service_name);
+		prefork_new_task(ev,
+				 rc->master->lp_ctx,
+				 rc->service_name,
+				 rc->master->new_task_fn,
+				 rc->master->private_data,
+				 rc->service_details,
+				 rc->from_parent_fd);
+	} else if (rc->worker != NULL) {
+		struct process_details pd = initial_process_details;
+		DBG_ERR("Restarting [%s] pre-fork worker(%d)\n",
+			rc->service_name,
+			rc->worker->instance);
+		pd.instances = rc->worker->instance;
+		prefork_fork_worker(rc->worker->task,
+				    ev,
+				    rc->worker->ev2,
+				    rc->service_details,
+				    rc->service_name,
+				    rc->worker->control_pipe,
+				    &pd);
+	}
+}
 /*
   handle EOF on the child pipe in the parent, so we know when a
   process terminates without using SIGCHLD or waiting on all possible pids.
@@ -176,49 +203,26 @@ static void prefork_child_pipe_handler(struct tevent_context *ev,
 		DBG_ERR("Parent %d, Child %d terminated, "
 			"unable to get status code from tfork\n",
 			getpid(), pid);
+		prefork_restart(ev, rc);
 	} else if (WIFEXITED(status)) {
 		status = WEXITSTATUS(status);
 		DBG_ERR("Parent %d, Child %d exited with status %d\n",
 			 getpid(), pid,  status);
+		if (status != 0) {
+			prefork_restart(ev, rc);
+		}
 	} else if (WIFSIGNALED(status)) {
 		status = WTERMSIG(status);
 		DBG_ERR("Parent %d, Child %d terminated with signal %d\n",
 			getpid(), pid, status);
 		if (status == SIGABRT || status == SIGBUS || status == SIGFPE ||
-		    status == SIGILL || status == SIGSYS) {
-			/*
-			 * Lets restart the process.
-			 *
-			 */
-			tfork_destroy(&rc->t);
-			if (rc->master != NULL) {
-				DBG_ERR("Restarting [%s] pre-fork master\n",
-					rc->service_name);
-				prefork_new_task(ev,
-						 rc->master->lp_ctx,
-						 rc->service_name,
-						 rc->master->new_task_fn,
-						 rc->master->private_data,
-						 rc->service_details,
-						 rc->from_parent_fd);
-			} else if (rc->worker != NULL) {
-				struct process_details pd =
-				    initial_process_details;
-				DBG_ERR("Restarting [%s] pre-fork worker(%d)\n",
-					rc->service_name,
-					rc->worker->instance);
-				pd.instances = rc->worker->instance;
-				prefork_fork_worker(rc->worker->task,
-						    ev,
-						    rc->worker->ev2,
-						    rc->service_details,
-						    rc->service_name,
-						    rc->worker->control_pipe,
-						    &pd);
-			}
+		    status == SIGILL || status == SIGSYS || status == SIGSEGV) {
+
+			prefork_restart(ev, rc);
 		}
 	}
 	/* tfork allocates tfork structures with malloc */
+	tfork_destroy(&rc->t);
 	free(rc->t);
 	TALLOC_FREE(rc);
 	return;
@@ -529,15 +533,31 @@ static void prefork_new_task(
 
 }
 
-
-/* called when a task goes down */
-static void prefork_terminate(struct tevent_context *ev,
-			      struct loadparm_context *lp_ctx,
-			      const char *reason,
-			      bool fatal,
-			      void *process_context)
+/*
+ * called when a task terminates
+ */
+static void prefork_terminate_task(struct tevent_context *ev,
+				   struct loadparm_context *lp_ctx,
+				   const char *reason,
+				   bool fatal,
+				   void *process_context)
 {
 	DBG_DEBUG("called with reason[%s]\n", reason);
+	if (fatal == true) {
+		exit(127);
+	} else {
+		exit(0);
+	}
+}
+
+/*
+ * called when a connection completes
+ */
+static void prefork_terminate_connection(struct tevent_context *ev,
+					 struct loadparm_context *lp_ctx,
+					 const char *reason,
+					 void *process_context)
+{
 }
 
 /* called to set a title of a task or connection */
@@ -550,7 +570,8 @@ static const struct model_ops prefork_ops = {
 	.model_init		= prefork_model_init,
 	.accept_connection	= prefork_accept_connection,
 	.new_task		= prefork_new_task,
-	.terminate		= prefork_terminate,
+	.terminate_task		= prefork_terminate_task,
+	.terminate_connection	= prefork_terminate_connection,
 	.set_title		= prefork_set_title,
 };
 
diff --git a/source4/smbd/process_single.c b/source4/smbd/process_single.c
index 061f09673dd..12ed2e30a74 100644
--- a/source4/smbd/process_single.c
+++ b/source4/smbd/process_single.c
@@ -119,17 +119,28 @@ static void single_new_task(struct tevent_context *ev,
 	}
 }
 
-
-/* called when a task goes down */
-static void single_terminate(struct tevent_context *ev,
-			     struct loadparm_context *lp_ctx,
-			     const char *reason,
-			     bool fatal,
-			     void *process_context)
+/*
+ * Called when a task goes down
+ */
+static void single_terminate_task(struct tevent_context *ev,
+				  struct loadparm_context *lp_ctx,
+				  const char *reason,
+				  bool fatal,
+				  void *process_context)
 {
 	DBG_NOTICE("single_terminate: reason[%s]\n",reason);
 }
 
+/*
+ * Called when a connection has ended
+ */
+static void single_terminate_connection(struct tevent_context *ev,
+					struct loadparm_context *lp_ctx,
+					const char *reason,
+					void *process_context)
+{
+}
+
 /* called to set a title of a task or connection */
 static void single_set_title(struct tevent_context *ev, const char *title) 
 {
@@ -138,9 +149,10 @@ static void single_set_title(struct tevent_context *ev, const char *title)
 const struct model_ops single_ops = {
 	.name			= "single",
 	.model_init		= single_model_init,
-	.new_task               = single_new_task,
+	.new_task		= single_new_task,
 	.accept_connection	= single_accept_connection,
-	.terminate              = single_terminate,
+	.terminate_task		= single_terminate_task,
+	.terminate_connection	= single_terminate_connection,
 	.set_title		= single_set_title,
 };
 
diff --git a/source4/smbd/process_standard.c b/source4/smbd/process_standard.c
index 536a6be3007..d25538716ba 100644
--- a/source4/smbd/process_standard.c
+++ b/source4/smbd/process_standard.c
@@ -510,15 +510,27 @@ static void standard_new_task(struct tevent_context *ev,
 
 
 /* called when a task goes down */
-static void standard_terminate(struct tevent_context *ev,
-			       struct loadparm_context *lp_ctx,
-			       const char *reason,
-			       bool fatal,
-			       void *process_context)
+static void standard_terminate_task(struct tevent_context *ev,
+				    struct loadparm_context *lp_ctx,
+				    const char *reason,
+				    bool fatal,
+				    void *process_context)
+{
+	if (fatal == true) {
+		exit(127);
+	}
+	exit(0);
+}
+
+/* called when a connection terminates*/
+static void standard_terminate_connection(struct tevent_context *ev,
+					  struct loadparm_context *lp_ctx,
+					  const char *reason,
+					  void *process_context)
 {
 	struct process_context *proc_ctx = NULL;
 
-	DBG_DEBUG("process terminating reason[%s]\n", reason);
+	DBG_DEBUG("connection terminating reason[%s]\n", reason);
 	if (process_context == NULL) {
 		smb_panic("Panicking process_context is NULL");
 	}
@@ -547,7 +559,6 @@ static void standard_terminate(struct tevent_context *ev,
 	/* terminate this process */
 	exit(0);
 }
-
 /* called to set a title of a task or connection */
 static void standard_set_title(struct tevent_context *ev, const char *title) 
 {
@@ -562,9 +573,10 @@ static const struct model_ops standard_ops = {
 	.name			= "standard",
 	.model_init		= standard_model_init,
 	.accept_connection	= standard_accept_connection,
-	.new_task               = standard_new_task,
-	.terminate              = standard_terminate,
-	.set_title              = standard_set_title,
+	.new_task		= standard_new_task,
+	.terminate_task		= standard_terminate_task,
+	.terminate_connection	= standard_terminate_connection,
+	.set_title		= standard_set_title,
 };
 
 /*
diff --git a/source4/smbd/service_stream.c b/source4/smbd/service_stream.c
index 336a0cb32aa..d0ad45aca3a 100644
--- a/source4/smbd/service_stream.c
+++ b/source4/smbd/service_stream.c
@@ -57,7 +57,6 @@ void stream_terminate_connection(struct stream_connection *srv_conn, const char
 	const struct model_ops *model_ops = srv_conn->model_ops;
 	struct loadparm_context *lp_ctx = srv_conn->lp_ctx;
 	void *process_context = srv_conn->process_context;
-	bool fatal = true;
 	TALLOC_CTX *frame = NULL;
 
 	if (!reason) reason = "unknown reason";
@@ -92,7 +91,8 @@ void stream_terminate_connection(struct stream_connection *srv_conn, const char
 	srv_conn->event.fde = NULL;
 	imessaging_cleanup(srv_conn->msg_ctx);
 	TALLOC_FREE(srv_conn);
-	model_ops->terminate(event_ctx, lp_ctx, reason, fatal, process_context);
+	model_ops->terminate_connection(
+	    event_ctx, lp_ctx, reason, process_context);
 	TALLOC_FREE(frame);
 }
 
diff --git a/source4/smbd/service_task.c b/source4/smbd/service_task.c
index cb1f4d5ad1f..d911027db0a 100644
--- a/source4/smbd/service_task.c
+++ b/source4/smbd/service_task.c
@@ -54,7 +54,7 @@ void task_server_terminate(struct task_server *task, const char *reason, bool fa
 
 	imessaging_cleanup(task->msg_ctx);
 
-	model_ops->terminate(
+	model_ops->terminate_task(
 	    event_ctx, task->lp_ctx, reason, fatal, task->process_context);
 	/* don't free this above, it might contain the 'reason' being printed */
 	talloc_free(task);