tools/xenstore: reduce number of watch events

When removing a watched node outside of a transaction, two watch events
are being produced instead of just a single one.

When finalizing a transaction watch events can be generated for each
node which is being modified, even if outside a transaction such
modifications might not have resulted in a watch event.

This happens e.g.:

- for nodes which are only modified due to added/removed child entries
- for nodes being removed or created implicitly (e.g. creation of a/b/c
  is implicitly creating a/b, resulting in watch events for a, a/b and
  a/b/c instead of a/b/c only)

Avoid these additional watch events, in order to reduce the needed
memory inside Xenstore for queueing them.

This is being achieved by adding event flags to struct accessed_node
specifying whether an event should be triggered, and whether it should
be an exact match of the modified path. Both flags can be set from
fire_watches() instead of implying them only.

This is part of XSA-326.

Signed-off-by: Juergen Gross <jgross@suse.com>
Reviewed-by: Julien Grall <jgrall@amazon.com>
(cherry picked from commit 3a96013a3e)

tools/xenstore/xenstored_core.c

@@ -1182,7 +1182,7 @@ static void delete_child(struct connection *conn,
 }
 
 static int delete_node(struct connection *conn, const void *ctx,
-		       struct node *parent, struct node *node)
+		       struct node *parent, struct node *node, bool watch_exact)
 {
 	char *name;
 
@@ -1194,7 +1194,7 @@ static int delete_node(struct connection *conn, const void *ctx,
 				       node->children);
 		child = name ? read_node(conn, node, name) : NULL;
 		if (child) {
-			if (delete_node(conn, ctx, node, child))
+			if (delete_node(conn, ctx, node, child, true))
 				return errno;
 		} else {
 			trace("delete_node: Error deleting child '%s/%s'!\n",
@@ -1206,7 +1206,12 @@ static int delete_node(struct connection *conn, const void *ctx,
 		talloc_free(name);
 	}
 
-	fire_watches(conn, ctx, node->name, node, true, NULL);
+	/*
+	 * Fire the watches now, when we can still see the node permissions.
+	 * This fine as we are single threaded and the next possible read will
+	 * be handled only after the node has been really removed.
+	 */
+	fire_watches(conn, ctx, node->name, node, watch_exact, NULL);
 	delete_node_single(conn, node);
 	delete_child(conn, parent, basename(node->name));
 	talloc_free(node);
@@ -1232,13 +1237,7 @@ static int _rm(struct connection *conn, const void *ctx, struct node *node,
 		return (errno == ENOMEM) ? ENOMEM : EINVAL;
 	node->parent = parent;
 
-	/*
-	 * Fire the watches now, when we can still see the node permissions.
-	 * This fine as we are single threaded and the next possible read will
-	 * be handled only after the node has been really removed.
-	 */
-	fire_watches(conn, ctx, name, node, false, NULL);
-	return delete_node(conn, ctx, parent, node);
+	return delete_node(conn, ctx, parent, node, false);
 }
 
 

tools/xenstore/xenstored_transaction.c

@@ -130,6 +130,10 @@ struct accessed_node
 
 	/* Transaction node in data base? */
 	bool ta_node;
+
+	/* Watch event flags. */
+	bool fire_watch;
+	bool watch_exact;
 };
 
 struct changed_domain
@@ -329,6 +333,29 @@ err:
 	return ret;
 }
 
+/*
+ * A watch event should be fired for a node modified inside a transaction.
+ * Set the corresponding information. A non-exact event is replacing an exact
+ * one, but not the other way round.
+ */
+void queue_watches(struct connection *conn, const char *name, bool watch_exact)
+{
+	struct accessed_node *i;
+
+	i = find_accessed_node(conn->transaction, name);
+	if (!i) {
+		conn->transaction->fail = true;
+		return;
+	}
+
+	if (!i->fire_watch) {
+		i->fire_watch = true;
+		i->watch_exact = watch_exact;
+	} else if (!watch_exact) {
+		i->watch_exact = false;
+	}
+}
+
 /*
  * Finalize transaction:
  * Walk through accessed nodes and check generation against global data.
@@ -383,15 +410,15 @@ static int finalize_transaction(struct connection *conn,
 				ret = tdb_store(tdb_ctx, key, data,
 						TDB_REPLACE);
 				talloc_free(data.dptr);
-				if (ret)
-					goto err;
-				fire_watches(conn, trans, i->node, NULL, false,
-					     i->perms.p ? &i->perms : NULL);
 			} else {
-				fire_watches(conn, trans, i->node, NULL, false,
+				ret = tdb_delete(tdb_ctx, key);
+			}
+			if (ret)
+				goto err;
+			if (i->fire_watch) {
+				fire_watches(conn, trans, i->node, NULL,
+					     i->watch_exact,
 					     i->perms.p ? &i->perms : NULL);
-				if (tdb_delete(tdb_ctx, key))
-					goto err;
 			}
 		}
 

tools/xenstore/xenstored_transaction.h

@@ -42,6 +42,9 @@ void transaction_entry_dec(struct transaction *trans, unsigned int domid);
 int access_node(struct connection *conn, struct node *node,
                 enum node_access_type type, TDB_DATA *key);
 
+/* Queue watches for a modified node. */
+void queue_watches(struct connection *conn, const char *name, bool watch_exact);
+
 /* Prepend the transaction to name if appropriate. */
 int transaction_prepend(struct connection *conn, const char *name,
                         TDB_DATA *key);

tools/xenstore/xenstored_watch.c

@@ -29,6 +29,7 @@
 #include "xenstore_lib.h"
 #include "utils.h"
 #include "xenstored_domain.h"
+#include "xenstored_transaction.h"
 
 extern int quota_nb_watch_per_domain;
 
@@ -143,9 +144,11 @@ void fire_watches(struct connection *conn, const void *ctx, const char *name,
 	struct connection *i;
 	struct watch *watch;
 
-	/* During transactions, don't fire watches. */
-	if (conn && conn->transaction)
+	/* During transactions, don't fire watches, but queue them. */
+	if (conn && conn->transaction) {
+		queue_watches(conn, name, exact);
 		return;
+	}
 
 	/* Create an event for each watch. */
 	list_for_each_entry(i, &connections, list) {