6933954 nwamd fails to retry upgrades when boot timing thwarts early attempts

6944167 network is unavailable when using LiveCD/Text installer/bootable AI in osol_137

4 files changed, 81 insertions, 51 deletions

diff --git a/usr/src/cmd/cmd-inet/lib/nwamd/llp.c b/usr/src/cmd/cmd-inet/lib/nwamd/llp.c
index c963216731..b371493149 100644
--- a/usr/src/cmd/cmd-inet/lib/nwamd/llp.c
+++ b/usr/src/cmd/cmd-inet/lib/nwamd/llp.c
@@ -20,8 +20,7 @@
  */
 
 /*
- * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  */
 
 /*
@@ -306,10 +305,20 @@ upgrade_llp_config(void)
 	default:
 		break;
 	}
-	if ((err = nwam_ncp_create(NWAM_NCP_NAME_USER, 0, &user_ncp))
-	    != NWAM_SUCCESS) {
+
+	err = nwam_ncp_create(NWAM_NCP_NAME_USER, 0, &user_ncp);
+	switch (err) {
+	case NWAM_SUCCESS:
+		break;
+	case NWAM_ERROR_BIND:
+	case NWAM_ERROR_INTERNAL:
 		nlog(LOG_ERR, "upgrade_llp_config: "
-		    "could not create user NCP: %s", nwam_strerror(err));
+		    "could not create User NCP: %s", nwam_strerror(err));
+		llp_list_free();
+		return (EAGAIN);
+	default:
+		nlog(LOG_ERR, "upgrade_llp_config: error creating User NCP: %s",
+		    nwam_strerror(err));
 		llp_list_free();
 		return (0);
 	}
@@ -402,21 +411,22 @@ upgrade_llp_config(void)
 			    nwam_strerror(err));
 			/* Schedule a retry - root filesystem may be readonly */
 			llp_list_free();
+			nwam_ncu_free(ip_ncu);
+			nwam_ncu_free(phys_ncu);
+			(void) nwam_ncp_destroy(user_ncp, 0);
 			return (EAGAIN);
 		}
-		nwam_ncu_free(ip_ncu);
-		nwam_ncu_free(phys_ncu);
 	}
 
 	if (err != NWAM_SUCCESS) {
 		nlog(LOG_ERR, "upgrade_llp_config: llp %s: "
 		    "could not set value for property %s: %s", wp->llp_lname,
 		    prop, nwam_strerror(err));
-		nwam_ncu_free(ip_ncu);
-		nwam_ncu_free(phys_ncu);
-		exit(EXIT_FAILURE);
 	}
 	llp_list_free();
+	nwam_ncu_free(ip_ncu);
+	nwam_ncu_free(phys_ncu);
+	nwam_ncp_free(user_ncp);
 	return (0);
 }
 
@@ -452,8 +462,12 @@ nwamd_handle_upgrade(nwamd_event_t event)
 		 */
 		upgrade_event = nwamd_event_init(NWAM_EVENT_TYPE_UPGRADE,
 		    NWAM_OBJECT_TYPE_NCP, 0, NULL);
-		if (upgrade_event == NULL)
+		if (upgrade_event == NULL) {
+			nlog(LOG_ERR, "nwamd_handle_upgrade: "
+			    "could not create retry event to upgrade "
+			    "%s configuration", LLPFILE);
 			return;
+		}
 		nwamd_event_enqueue_timed(upgrade_event,
 		    NWAMD_READONLY_RETRY_INTERVAL);
 		return;
diff --git a/usr/src/cmd/cmd-inet/lib/nwamd/main.c b/usr/src/cmd/cmd-inet/lib/nwamd/main.c
index 332499c80d..108463d603 100644
--- a/usr/src/cmd/cmd-inet/lib/nwamd/main.c
+++ b/usr/src/cmd/cmd-inet/lib/nwamd/main.c
@@ -20,8 +20,7 @@
  */
 
 /*
- * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  */
 
 #include <errno.h>
@@ -376,6 +375,15 @@ main(int argc, char *argv[])
 	}
 
 	/*
+	 * Create the event queue before starting event sources, including
+	 * signal handling, so we are ready to handle incoming events.  Also
+	 * start before attempting to upgrade, in case there's a problem
+	 * upgrading and we need to retry (in which case we schedule an event
+	 * to do so).
+	 */
+	nwamd_event_queue_init();
+
+	/*
 	 * Handle upgrade of legacy config.  Absence of version property
 	 * (which did not exist in phase 0 or 0.5) is the indication that
 	 * we need to upgrade to phase 1 (version 1).
@@ -389,13 +397,6 @@ main(int argc, char *argv[])
 	 */
 	nwamd_object_lists_init();
 
-	/*
-	 * Start the event handling thread before starting event sources,
-	 * including signal handling, so we are ready to handle incoming
-	 * events.
-	 */
-	nwamd_event_queue_init();
-
 	init_signalhandling();
 
 	/* Enqueue init event */
@@ -403,6 +404,7 @@ main(int argc, char *argv[])
 	if (event == NULL)
 		pfail("nwamd could not create init event, exiting");
 	nwamd_event_enqueue(event);
+
 	/*
 	 * Collect initial user configuration.
 	 */
diff --git a/usr/src/cmd/cmd-inet/lib/nwamd/ncp.c b/usr/src/cmd/cmd-inet/lib/nwamd/ncp.c
index 55bbbf707c..7292b3f2df 100644
--- a/usr/src/cmd/cmd-inet/lib/nwamd/ncp.c
+++ b/usr/src/cmd/cmd-inet/lib/nwamd/ncp.c
@@ -20,8 +20,7 @@
  */
 
 /*
- * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  */
 
 #include <arpa/inet.h>
@@ -116,23 +115,34 @@ nwamd_ncp_handle_enable_event(nwamd_event_t event)
 
 	nwamd_fini_ncus();
 
-	err = nwam_ncp_read(new_ncp, 0, &new_ncph);
-	switch (err) {
-	case NWAM_ENTITY_NOT_FOUND:
+	if ((err = nwam_ncp_read(new_ncp, 0, &new_ncph))
+	    == NWAM_ENTITY_NOT_FOUND) {
 		err = nwam_ncp_create(new_ncp, 0, &new_ncph);
-		break;
-	case NWAM_SUCCESS:
-		break;
-	default:
-		nlog(LOG_ERR, "nwamd_ncp_handle_enable_event: error %s",
-		    nwam_strerror(err));
-		return;
 	}
-	nwam_ncp_free(new_ncph);
 
 	if (err == NWAM_SUCCESS) {
+		nwam_ncp_free(new_ncph);
 		nwamd_object_set_state(NWAM_OBJECT_TYPE_NCP, new_ncp,
 		    NWAM_STATE_ONLINE, NWAM_AUX_STATE_ACTIVE);
+	} else if (initial_ncp_enable) {
+		/*
+		 * We weren't able to enable the NCP when nwamd starts up,
+		 * retry in a few seconds.
+		 */
+		nwamd_event_t retry_event = nwamd_event_init_object_action
+		    (NWAM_OBJECT_TYPE_NCP, new_ncp, NULL, NWAM_ACTION_ENABLE);
+		if (retry_event == NULL) {
+			nlog(LOG_ERR, "nwamd_ncp_handle_enable_event: "
+			    "could not create retry event to enable %s NCP",
+			    new_ncp);
+			return;
+		}
+
+		nlog(LOG_ERR, "nwamd_ncp_handle_enable_event: "
+		    "failed to enable %s NCP, retrying in %d seconds",
+		    new_ncp, NWAMD_READONLY_RETRY_INTERVAL);
+		nwamd_event_enqueue_timed(retry_event,
+		    NWAMD_READONLY_RETRY_INTERVAL);
 	} else {
 		nlog(LOG_ERR, "nwamd_ncp_handle_enable_event: error %s",
 		    nwam_strerror(err));
@@ -240,7 +250,6 @@ find_next_priority_group_cb(nwamd_object_t object, void *data)
 	return (0);
 }
 
-
 /* Set current_pg to next pg in NCP that is >= start_pg */
 boolean_t
 nwamd_ncp_find_next_priority_group(int64_t minpriority,
diff --git a/usr/src/cmd/cmd-inet/lib/nwamd/ncu.c b/usr/src/cmd/cmd-inet/lib/nwamd/ncu.c
index 9a49ce37b4..b44565c1d7 100644
--- a/usr/src/cmd/cmd-inet/lib/nwamd/ncu.c
+++ b/usr/src/cmd/cmd-inet/lib/nwamd/ncu.c
@@ -20,8 +20,7 @@
  */
 
 /*
- * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
- * Use is subject to license terms.
+ * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  */
 
 #include <arpa/inet.h>
@@ -1275,21 +1274,10 @@ nwamd_ncu_handle_link_action_event(nwamd_event_t event)
 	if ((err = nwam_ncp_read(NWAM_NCP_NAME_AUTOMATIC, 0, &ncph))
 	    == NWAM_ENTITY_NOT_FOUND) {
 		/* Automatic NCP doesn't exist, create it */
-		if ((err = nwam_ncp_create(NWAM_NCP_NAME_AUTOMATIC, 0,
-		    &ncph)) != NWAM_SUCCESS) {
-			nlog(LOG_ERR,
-			    "nwamd_ncu_handle_link_action_event: "
-			    "could not create %s NCP: %s",
-			    NWAM_NCP_NAME_AUTOMATIC,
-			    nwam_strerror(err));
-			goto cleanup_exit;
-		}
-	} else if (err != NWAM_SUCCESS) {
-		nlog(LOG_ERR, "nwamd_ncu_handle_link_action_event: "
-		    "failed to read Automatic NCP: %s",
-		    nwam_strerror(err));
-		goto cleanup_exit;
+		err = nwam_ncp_create(NWAM_NCP_NAME_AUTOMATIC, 0, &ncph);
 	}
+	if (err != NWAM_SUCCESS)
+		goto fail;
 
 	/* add or remove NCUs from Automatic NCP */
 	if (action == NWAM_ACTION_ADD) {
@@ -1323,8 +1311,24 @@ nwamd_ncu_handle_link_action_event(nwamd_event_t event)
 		nwamd_create_ncu_check_event(NEXT_FEW_SECONDS);
 	}
 
-cleanup_exit:
+fail:
 	free(name);
+	if (err != NWAM_SUCCESS) {
+		nwamd_event_t retry_event = nwamd_event_init_link_action(name,
+		    action);
+		if (retry_event == NULL) {
+			nlog(LOG_ERR, "nwamd_ncu_handle_link_action_event: "
+			    "could not create retry event to read/create "
+			    "%s NCP", NWAM_NCP_NAME_AUTOMATIC);
+			return;
+		}
+
+		nlog(LOG_ERR, "nwamd_ncu_handle_link_action_event: "
+		    "could not read/create %s NCP, retrying in %d seconds",
+		    NWAM_NCP_NAME_AUTOMATIC, NWAMD_READONLY_RETRY_INTERVAL);
+		nwamd_event_enqueue_timed(retry_event,
+		    NWAMD_READONLY_RETRY_INTERVAL);
+	}
 }
 
 /*
@@ -1529,6 +1533,7 @@ nwamd_ncu_handle_init_event(nwamd_event_t event)
 		nlog(LOG_DEBUG,
 		    "nwamd_ncu_handle_init_event: active NCP handle NULL");
 		nwamd_event_do_not_send(event);
+		free(name);
 		(void) pthread_mutex_unlock(&active_ncp_mutex);
 		return;
 	}