platform: resynchronize with kernel when we're out of buffer space

Kernel can return ENOBUFS in variety of reasons. If that happens, we know we've
lost events and should pick up kernel state.

Simple reproducer that triggers an ENOBUFS condition no matter how big our
netlink socket buffer is:

  ip link add bridge0 type bridge
  for i in seq $(0 1023); do ip link add dummy$i type dummy; \
    ip link set dummy$i master bridge0; done
  ip link del bridge0

1 files changed, 84 insertions, 10 deletions

diff --git a/src/platform/nm-linux-platform.c b/src/platform/nm-linux-platform.c
index 3cf4cc8306..1fb0bd014e 100644
--- a/src/platform/nm-linux-platform.c
+++ b/src/platform/nm-linux-platform.c
@@ -4028,6 +4028,70 @@ init_link_cache (NMPlatform *platform)
 	} while (object);
 }
 
+/* Calls announce_object with appropriate arguments for all objects
+ * which are not coherent between old and new caches and deallocates
+ * the old cache. */
+static void
+cache_announce_changes (NMPlatform *platform, struct nl_cache *new, struct nl_cache *old)
+{
+	struct nl_object *object;
+
+	if (!old)
+		return;
+
+	for (object = nl_cache_get_first (new); object; object = nl_cache_get_next (object)) {
+		struct nl_object *cached_object = nm_nl_cache_search (old, object);
+
+		if (cached_object) {
+			ObjectType type = object_type_from_nl_object (object);
+			if (nm_nl_object_diff (type, object, cached_object))
+				announce_object (platform, object, NM_PLATFORM_SIGNAL_CHANGED, NM_PLATFORM_REASON_EXTERNAL);
+			nl_object_put (cached_object);
+		} else
+			announce_object (platform, object, NM_PLATFORM_SIGNAL_ADDED, NM_PLATFORM_REASON_EXTERNAL);
+	}
+	for (object = nl_cache_get_first (old); object; object = nl_cache_get_next (object)) {
+		struct nl_object *cached_object = nm_nl_cache_search (new, object);
+		if (cached_object)
+			nl_object_put (cached_object);
+		else
+			announce_object (platform, object, NM_PLATFORM_SIGNAL_REMOVED, NM_PLATFORM_REASON_EXTERNAL);
+	}
+
+	nl_cache_free (old);
+}
+
+/* Creates and populates the netlink object caches. Called upon platform init and
+ * when we run out of sync (out of buffer space, netlink congestion control). In case
+ * the caches already exist, it finds changed, added and removed objects, announces
+ * them and destroys the old caches. */
+static void
+cache_repopulate_all (NMPlatform *platform)
+{
+	NMLinuxPlatformPrivate *priv = NM_LINUX_PLATFORM_GET_PRIVATE (platform);
+	struct nl_cache *old_link_cache = priv->link_cache;
+	struct nl_cache *old_address_cache = priv->address_cache;
+	struct nl_cache *old_route_cache = priv->route_cache;
+	struct nl_object *object;
+
+	debug ("platform: %spopulate platform cache", old_link_cache ? "re" : "");
+
+	/* Allocate new netlink caches */
+	init_link_cache (platform);
+	rtnl_addr_alloc_cache (priv->nlh, &priv->address_cache);
+	rtnl_route_alloc_cache (priv->nlh, AF_UNSPEC, 0, &priv->route_cache);
+	g_assert (priv->link_cache && priv->address_cache && priv->route_cache);
+
+	for (object = nl_cache_get_first (priv->address_cache); object; object = nl_cache_get_next (object)) {
+		_rtnl_addr_hack_lifetimes_rel_to_abs ((struct rtnl_addr *) object);
+	}
+
+	/* Make sure all changes we've missed are announced. */
+	cache_announce_changes (platform, priv->link_cache, old_link_cache);
+	cache_announce_changes (platform, priv->address_cache, old_address_cache);
+	cache_announce_changes (platform, priv->route_cache, old_route_cache);
+}
+
 /******************************************************************/
 
 #define EVENT_CONDITIONS      ((GIOCondition) (G_IO_IN | G_IO_PRI))
@@ -4056,7 +4120,8 @@ event_handler (GIOChannel *channel,
                GIOCondition io_condition,
                gpointer user_data)
 {
-	NMLinuxPlatformPrivate *priv = NM_LINUX_PLATFORM_GET_PRIVATE (user_data);
+	NMPlatform *platform = NM_PLATFORM (user_data);
+	NMLinuxPlatformPrivate *priv = NM_LINUX_PLATFORM_GET_PRIVATE (platform);
 	int nle;
 
 	nle = nl_recvmsgs_default (priv->nlh_event);
@@ -4068,6 +4133,17 @@ event_handler (GIOChannel *channel,
 			 * and can happen easily. */
 			debug ("Uncritical failure to retrieve incoming events: %s (%d)", nl_geterror (nle), nle);
 			break;
+		case -NLE_NOMEM:
+			warning ("Too many netlink events. Need to resynchronize platform cache");
+			/* Drain the event queue, we've lost events and are out of sync anyway and we'd
+			 * like to free up some space. We'll read in the status synchronously. */
+			nl_socket_modify_cb (priv->nlh_event, NL_CB_VALID, NL_CB_DEFAULT, NULL, NULL);
+			do {
+				nle = nl_recvmsgs_default (priv->nlh_event);
+			} while (nle != -NLE_AGAIN);
+			nl_socket_modify_cb (priv->nlh_event, NL_CB_VALID, NL_CB_CUSTOM, event_notification, user_data);
+			cache_repopulate_all (platform);
+			break;
 		default:
 			error ("Failed to retrieve incoming events: %s (%d)", nl_geterror (nle), nle);
 			break;
@@ -4099,6 +4175,12 @@ setup_socket (gboolean event, gpointer user_data)
 	nle = nl_socket_set_passcred (sock, 1);
 	g_assert (!nle);
 
+	/* No blocking for event socket, so that we can drain it safely. */
+	if (event) {
+		nle = nl_socket_set_nonblocking (sock);
+		g_assert (!nle);
+	}
+
 	return sock;
 }
 
@@ -4244,7 +4326,6 @@ setup (NMPlatform *platform)
 	int channel_flags;
 	gboolean status;
 	int nle;
-	struct nl_object *object;
 
 	/* Initialize netlink socket for requests */
 	priv->nlh = setup_socket (FALSE, platform);
@@ -4280,14 +4361,7 @@ setup (NMPlatform *platform)
 		(EVENT_CONDITIONS | ERROR_CONDITIONS | DISCONNECT_CONDITIONS),
 		event_handler, platform);
 
-	/* Allocate netlink caches */
-	init_link_cache (platform);
-	rtnl_addr_alloc_cache (priv->nlh, &priv->address_cache);
-	rtnl_route_alloc_cache (priv->nlh, AF_UNSPEC, 0, &priv->route_cache);
-	g_assert (priv->link_cache && priv->address_cache && priv->route_cache);
-
-	for (object = nl_cache_get_first (priv->address_cache); object; object = nl_cache_get_next (object))
-		_rtnl_addr_hack_lifetimes_rel_to_abs ((struct rtnl_addr *) object);
+	cache_repopulate_all (platform);
 
 #if HAVE_LIBNL_INET6_ADDR_GEN_MODE
 	/* Initial check for user IPv6LL support once the link cache is allocated