[sctp] remove sockets from epoll earlier

on some slow machines, where scheduling of some threads can be blocked
for some time, keeping a FD on error on the epoll can cause even more
load and reduce chances of the error handling thread to be scheduled.

Signed-off-by: Fabio M. Di Nitto <fdinitto@redhat.com>

libknet/transport_sctp.c

@@ -380,6 +380,7 @@ int sctp_transport_tx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err,
 int sctp_transport_rx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno)
 {
 	struct epoll_event ev;
+	sctp_connect_link_info_t *info = knet_h->knet_transport_fd_tracker[sockfd].data;
 	sctp_accepted_link_info_t *accepted_info = knet_h->knet_transport_fd_tracker[sockfd].data;
 	sctp_listen_link_info_t *listen_info;
 	sctp_handle_info_t *handle_info = knet_h->transports[KNET_TRANSPORT_SCTP];
@@ -395,6 +396,17 @@ int sctp_transport_rx_sock_error(knet_handle_t knet_h, int sockfd, int recv_err,
 			 * they follow a notification (double notification)
 			 */
 			if (recv_err != 1) {
+				if (info->on_rx_epoll) {
+					memset(&ev, 0, sizeof(struct epoll_event));
+					ev.events = EPOLLIN;
+					ev.data.fd = info->connect_sock;
+					if (epoll_ctl(knet_h->recv_from_links_epollfd, EPOLL_CTL_DEL, info->connect_sock, &ev)) {
+						log_err(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to remove connected socket from epoll pool: %s",
+						strerror(errno));
+						return -1;
+					}
+					info->on_rx_epoll = 0;
+				}
 				log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Notifying connect thread that sockfd %d received an error", sockfd);
 				if (sendto(handle_info->connectsockfd[1], &sockfd, sizeof(int), MSG_DONTWAIT | MSG_NOSIGNAL, NULL, 0) != sizeof(int)) {
 					log_debug(knet_h, KNET_SUB_TRANSP_SCTP, "Unable to notify connect thread: %s", strerror(errno));