On Thu, Sep 22, 2011 at 11:07:43AM -0700, Jesse Gross wrote:
> Yeah, I understand the rationale for the FD-swapping and I think it
> makes sense in the context of transactions and dumps being intermixed.
> The specific part that made me nervous is usage with multicast or
> unicast upcalls where it's important to have the right socket. And
> since existing users already segregate asynchronous usage it's not
> really worth the extra complexity to fully handle. Your patch below
> looks good to me.
Thanks. I pushed it, after fixing up a comment that I noticed was
wrong:
diff --git a/lib/netlink-socket.c b/lib/netlink-socket.c
index 085918d..8f9c3d5 100644
--- a/lib/netlink-socket.c
+++ b/lib/netlink-socket.c
@@ -538,9 +538,8 @@ nl_dump_start(struct nl_dump *dump,
dump->seq = nlmsghdr->nlmsg_seq;
dump->buffer = NULL;
if (sock->dump) {
- /* 'sock' already has an ongoing dump. Clone the socket to avoid
- * possibly intermixing multicast messages or previous dump results
- * with our results. */
+ /* 'sock' already has an ongoing dump. Clone the socket because
+ * Netlink only allows one dump at a time. */
dump->status = nl_sock_clone(sock, &dump->sock);
if (dump->status) {
return;
_______________________________________________
dev mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/dev
