Remove ill-conceived concept of watches blocking reply on

connection which did write/mkdir/rm/setperm etc.
This causes deadlocks in real life, and I can't see a sane way
of avoiding them: it is reasonable for someone to ignore watch
notifications while doing other actions, and that means that
we can do other writes.  These writes can block pending other
watchers; if one of these is the process blocked awaiting our
ack, we deadlock.
Signed-off-by: Rusty Russel <rusty@rustcorp.com.au>
Signed-off-by: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>

1 files changed, 2 insertions, 7 deletions

diff --git a/tools/xenstore/xenstored_transaction.c b/tools/xenstore/xenstored_transaction.c
index afaef1bef2..eef4d29038 100644
--- a/tools/xenstore/xenstored_transaction.c
+++ b/tools/xenstore/xenstored_transaction.c
@@ -307,7 +307,6 @@ void do_transaction_end(struct connection *conn, const char *arg)
 {
 	struct changed_node *i;
 	struct transaction *trans;
-	bool fired = false;
 
 	if (!arg || (!streq(arg, "T") && !streq(arg, "F"))) {
 		send_error(conn, EINVAL);
@@ -337,12 +336,8 @@ void do_transaction_end(struct connection *conn, const char *arg)
 
 		/* Fire off the watches for everything that changed. */
 		list_for_each_entry(i, &trans->changes, list)
-			fired |= fire_watches(conn, i->node, i->recurse);
+			fire_watches(conn, i->node, i->recurse);
 	}
-
-	if (fired)
-		conn->watch_ack = XS_TRANSACTION_END;
-	else
-		send_ack(conn, XS_TRANSACTION_END);
+	send_ack(conn, XS_TRANSACTION_END);
 }