No OneTemporary
Actions

Size

66 KB

Referenced Files

None

Subscribers

None

View Options

	diff --git a/libknet/handle.c b/libknet/handle.c
	index c8f87e3d..9f90b7f5 100644
	--- a/libknet/handle.c
	+++ b/libknet/handle.c
	@@ -1,1418 +1,1419 @@
	/*
	* Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved.
	*
	* Authors: Fabio M. Di Nitto <fabbione@kronosnet.org>
	* Federico Simoncelli <fsimon@kronosnet.org>
	*
	* This software licensed under GPL-2.0+, LGPL-2.0+
	*/

	#include "config.h"

	#include <stdlib.h>
	#include <string.h>
	#include <unistd.h>
	#include <errno.h>
	#include <pthread.h>
	#include <sys/uio.h>
	#include <math.h>
	#include <sys/time.h>
	#include <sys/resource.h>

	#include "internals.h"
	#include "crypto.h"
	#include "compat.h"
	#include "common.h"
	#include "threads_common.h"
	#include "threads_heartbeat.h"
	#include "threads_pmtud.h"
	#include "threads_dsthandler.h"
	#include "threads_rx.h"
	#include "threads_tx.h"
	#include "transports.h"
	#include "logging.h"

	static pthread_mutex_t handle_config_mutex = PTHREAD_MUTEX_INITIALIZER;

	static int _init_locks(knet_handle_t knet_h)
	{
	int savederrno = 0;

	savederrno = pthread_rwlock_init(&knet_h->global_rwlock, NULL);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize list rwlock: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	knet_h->lock_init_done = 1;

	savederrno = pthread_mutex_init(&knet_h->pmtud_mutex, NULL);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize pmtud mutex: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_cond_init(&knet_h->pmtud_cond, NULL);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize pmtud conditional mutex: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_mutex_init(&knet_h->hb_mutex, NULL);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize hb_thread mutex: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_mutex_init(&knet_h->tx_mutex, NULL);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize tx_thread mutex: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_mutex_init(&knet_h->tx_seq_num_mutex, NULL);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize tx_seq_num_mutex mutex: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	return 0;

	exit_fail:
	errno = savederrno;
	return -1;
	}

	static void _destroy_locks(knet_handle_t knet_h)
	{
	knet_h->lock_init_done = 0;
	pthread_rwlock_destroy(&knet_h->global_rwlock);
	pthread_mutex_destroy(&knet_h->pmtud_mutex);
	pthread_cond_destroy(&knet_h->pmtud_cond);
	pthread_mutex_destroy(&knet_h->hb_mutex);
	pthread_mutex_destroy(&knet_h->tx_mutex);
	pthread_mutex_destroy(&knet_h->tx_seq_num_mutex);
	}

	static int _init_socks(knet_handle_t knet_h)
	{
	int savederrno = 0;

	if (_init_socketpair(knet_h, knet_h->hostsockfd)) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize internal hostsockpair: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	if (_init_socketpair(knet_h, knet_h->dstsockfd)) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to initialize internal dstsockpair: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	return 0;

	exit_fail:
	errno = savederrno;
	return -1;
	}

	static void _close_socks(knet_handle_t knet_h)
	{
	_close_socketpair(knet_h, knet_h->dstsockfd);
	_close_socketpair(knet_h, knet_h->hostsockfd);
	}

	static int _init_buffers(knet_handle_t knet_h)
	{
	int savederrno = 0;
	int i;
	size_t bufsize;

	for (i = 0; i < PCKT_FRAG_MAX; i++) {
	knet_h->send_to_links_buf[i] = malloc(KNET_HEADER_DATA_SIZE);
	if (!knet_h->send_to_links_buf[i]) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory datafd to link buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->send_to_links_buf[i], 0, KNET_HEADER_DATA_SIZE);

	- knet_h->recv_from_sock_buf[i] = malloc(KNET_DATABUFSIZE);
	- if (!knet_h->recv_from_sock_buf[i]) {
	- savederrno = errno;
	- log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for app to datafd buffer: %s",
	- strerror(savederrno));
	- goto exit_fail;
	- }
	- memset(knet_h->recv_from_sock_buf[i], 0, KNET_DATABUFSIZE);
	-
	knet_h->recv_from_links_buf[i] = malloc(KNET_DATABUFSIZE);
	if (!knet_h->recv_from_links_buf[i]) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for link to datafd buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->recv_from_links_buf[i], 0, KNET_DATABUFSIZE);
	}

	+ knet_h->recv_from_sock_buf = malloc(KNET_DATABUFSIZE);
	+ if (!knet_h->recv_from_sock_buf) {
	+ savederrno = errno;
	+ log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for app to datafd buffer: %s",
	+ strerror(savederrno));
	+ goto exit_fail;
	+ }
	+ memset(knet_h->recv_from_sock_buf, 0, KNET_DATABUFSIZE);
	+
	knet_h->pingbuf = malloc(KNET_HEADER_PING_SIZE);
	if (!knet_h->pingbuf) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for hearbeat buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->pingbuf, 0, KNET_HEADER_PING_SIZE);

	knet_h->pmtudbuf = malloc(KNET_PMTUD_SIZE_V6);
	if (!knet_h->pmtudbuf) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for pmtud buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->pmtudbuf, 0, KNET_PMTUD_SIZE_V6);

	for (i = 0; i < PCKT_FRAG_MAX; i++) {
	bufsize = ceil((float)KNET_MAX_PACKET_SIZE / (i + 1)) + KNET_HEADER_ALL_SIZE + KNET_DATABUFSIZE_CRYPT_PAD;
	knet_h->send_to_links_buf_crypt[i] = malloc(bufsize);
	if (!knet_h->send_to_links_buf_crypt[i]) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for crypto datafd to link buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->send_to_links_buf_crypt[i], 0, bufsize);
	}

	knet_h->recv_from_links_buf_decrypt = malloc(KNET_DATABUFSIZE_CRYPT);
	if (!knet_h->recv_from_links_buf_decrypt) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto link to datafd buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->recv_from_links_buf_decrypt, 0, KNET_DATABUFSIZE_CRYPT);

	knet_h->recv_from_links_buf_crypt = malloc(KNET_DATABUFSIZE_CRYPT);
	if (!knet_h->recv_from_links_buf_crypt) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto link to datafd buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->recv_from_links_buf_crypt, 0, KNET_DATABUFSIZE_CRYPT);

	knet_h->pingbuf_crypt = malloc(KNET_DATABUFSIZE_CRYPT);
	if (!knet_h->pingbuf_crypt) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_CRYPTO, "Unable to allocate memory for crypto hearbeat buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->pingbuf_crypt, 0, KNET_DATABUFSIZE_CRYPT);

	knet_h->pmtudbuf_crypt = malloc(KNET_DATABUFSIZE_CRYPT);
	if (!knet_h->pmtudbuf_crypt) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to allocate memory for crypto pmtud buffer: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	memset(knet_h->pmtudbuf_crypt, 0, KNET_DATABUFSIZE_CRYPT);

	memset(knet_h->knet_transport_fd_tracker, KNET_MAX_TRANSPORTS, sizeof(knet_h->knet_transport_fd_tracker));

	return 0;

	exit_fail:
	errno = savederrno;
	return -1;
	}

	static void _destroy_buffers(knet_handle_t knet_h)
	{
	int i;

	for (i = 0; i < PCKT_FRAG_MAX; i++) {
	free(knet_h->send_to_links_buf[i]);
	- free(knet_h->recv_from_sock_buf[i]);
	free(knet_h->send_to_links_buf_crypt[i]);
	free(knet_h->recv_from_links_buf[i]);
	}
	+
	+ free(knet_h->recv_from_sock_buf);
	free(knet_h->recv_from_links_buf_decrypt);
	free(knet_h->recv_from_links_buf_crypt);
	free(knet_h->pingbuf);
	free(knet_h->pingbuf_crypt);
	free(knet_h->pmtudbuf);
	free(knet_h->pmtudbuf_crypt);
	}

	static int _init_epolls(knet_handle_t knet_h)
	{
	struct epoll_event ev;
	int savederrno = 0;

	/*
	* even if the kernel does dynamic allocation with epoll_ctl
	* we need to reserve one extra for host to host communication
	*/
	knet_h->send_to_links_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS + 1);
	if (knet_h->send_to_links_epollfd < 0) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll datafd to link fd: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	knet_h->recv_from_links_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS);
	if (knet_h->recv_from_links_epollfd < 0) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll link to datafd fd: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	knet_h->dst_link_handler_epollfd = epoll_create(KNET_EPOLL_MAX_EVENTS);
	if (knet_h->dst_link_handler_epollfd < 0) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to create epoll dst cache fd: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	if (_fdset_cloexec(knet_h->send_to_links_epollfd)) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on datafd to link epoll fd: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	if (_fdset_cloexec(knet_h->recv_from_links_epollfd)) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on link to datafd epoll fd: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	if (_fdset_cloexec(knet_h->dst_link_handler_epollfd)) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on dst cache epoll fd: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	memset(&ev, 0, sizeof(struct epoll_event));
	ev.events = EPOLLIN;
	ev.data.fd = knet_h->hostsockfd[0];

	if (epoll_ctl(knet_h->send_to_links_epollfd,
	EPOLL_CTL_ADD, knet_h->hostsockfd[0], &ev)) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to add hostsockfd[0] to epoll pool: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	memset(&ev, 0, sizeof(struct epoll_event));
	ev.events = EPOLLIN;
	ev.data.fd = knet_h->dstsockfd[0];

	if (epoll_ctl(knet_h->dst_link_handler_epollfd,
	EPOLL_CTL_ADD, knet_h->dstsockfd[0], &ev)) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to add dstsockfd[0] to epoll pool: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	return 0;

	exit_fail:
	errno = savederrno;
	return -1;
	}

	static void _close_epolls(knet_handle_t knet_h)
	{
	struct epoll_event ev;
	int i;

	memset(&ev, 0, sizeof(struct epoll_event));

	for (i = 0; i < KNET_DATAFD_MAX; i++) {
	if (knet_h->sockfd[i].in_use) {
	epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->sockfd[i].sockfd[knet_h->sockfd[i].is_created], &ev);
	if (knet_h->sockfd[i].sockfd[knet_h->sockfd[i].is_created]) {
	_close_socketpair(knet_h, knet_h->sockfd[i].sockfd);
	}
	}
	}

	epoll_ctl(knet_h->send_to_links_epollfd, EPOLL_CTL_DEL, knet_h->hostsockfd[0], &ev);
	epoll_ctl(knet_h->dst_link_handler_epollfd, EPOLL_CTL_DEL, knet_h->dstsockfd[0], &ev);
	close(knet_h->send_to_links_epollfd);
	close(knet_h->recv_from_links_epollfd);
	close(knet_h->dst_link_handler_epollfd);
	}

	static int _start_transports(knet_handle_t knet_h)
	{
	int i, savederrno = 0, err = 0;

	for (i=0; i<KNET_MAX_TRANSPORTS; i++) {
	switch (i) {
	case KNET_TRANSPORT_UDP:
	knet_h->transport_ops[i] = get_udp_transport();
	break;
	case KNET_TRANSPORT_SCTP:
	knet_h->transport_ops[i] = get_sctp_transport();
	break;
	}

	if (knet_h->transport_ops[i]) {
	if (knet_h->transport_ops[i]->transport_init(knet_h) < 0) {
	savederrno = errno;
	log_err(knet_h, KNET_SUB_HANDLE, "Failed to allocate transport handle for %s: %s",
	knet_h->transport_ops[i]->transport_name,
	strerror(savederrno));
	err = -1;
	goto out;
	}
	}
	}

	out:
	errno = savederrno;
	return err;
	}

	static void _stop_transports(knet_handle_t knet_h)
	{
	int i;

	for (i=0; i<KNET_MAX_TRANSPORTS; i++) {
	if (knet_h->transport_ops[i]) {
	knet_h->transport_ops[i]->transport_free(knet_h);
	}
	}
	}

	static int _start_threads(knet_handle_t knet_h)
	{
	int savederrno = 0;

	savederrno = pthread_create(&knet_h->pmtud_link_handler_thread, 0,
	_handle_pmtud_link_thread, (void *) knet_h);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to start pmtud link thread: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_create(&knet_h->dst_link_handler_thread, 0,
	_handle_dst_link_handler_thread, (void *) knet_h);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to start dst cache thread: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_create(&knet_h->send_to_links_thread, 0,
	_handle_send_to_links_thread, (void *) knet_h);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to start datafd to link thread: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_create(&knet_h->recv_from_links_thread, 0,
	_handle_recv_from_links_thread, (void *) knet_h);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to start link to datafd thread: %s",
	strerror(savederrno));
	goto exit_fail;
	}

	savederrno = pthread_create(&knet_h->heartbt_thread, 0,
	_handle_heartbt_thread, (void *) knet_h);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to start heartbeat thread: %s",
	strerror(savederrno));
	goto exit_fail;
	}
	return 0;

	exit_fail:
	errno = savederrno;
	return -1;
	}

	static void _stop_threads(knet_handle_t knet_h)
	{
	void *retval;

	/*
	* allow threads to catch on shutdown request
	* and release locks before we stop them.
	* this isn't the most efficent way to handle it
	* but it works good enough for now
	*/

	sleep(1);

	if (knet_h->heartbt_thread) {
	pthread_cancel(knet_h->heartbt_thread);
	pthread_join(knet_h->heartbt_thread, &retval);
	}

	if (knet_h->send_to_links_thread) {
	pthread_cancel(knet_h->send_to_links_thread);
	pthread_join(knet_h->send_to_links_thread, &retval);
	}

	if (knet_h->recv_from_links_thread) {
	pthread_cancel(knet_h->recv_from_links_thread);
	pthread_join(knet_h->recv_from_links_thread, &retval);
	}

	if (knet_h->dst_link_handler_thread) {
	pthread_cancel(knet_h->dst_link_handler_thread);
	pthread_join(knet_h->dst_link_handler_thread, &retval);
	}

	pthread_mutex_lock(&knet_h->pmtud_mutex);
	pthread_cond_signal(&knet_h->pmtud_cond);
	pthread_mutex_unlock(&knet_h->pmtud_mutex);

	sleep(1);

	if (knet_h->pmtud_link_handler_thread) {
	pthread_cancel(knet_h->pmtud_link_handler_thread);
	pthread_join(knet_h->pmtud_link_handler_thread, &retval);
	}
	}

	knet_handle_t knet_handle_new(knet_node_id_t host_id,
	int log_fd,
	uint8_t default_log_level)
	{
	knet_handle_t knet_h;
	int savederrno = 0;
	struct rlimit cur;

	if (getrlimit(RLIMIT_NOFILE, &cur) < 0) {
	return NULL;
	}

	if ((log_fd < 0) \|\| ((unsigned int)log_fd >= cur.rlim_max)) {
	errno = EINVAL;
	return NULL;
	}

	/*
	* validate incoming request
	*/

	if ((log_fd) && (default_log_level > KNET_LOG_DEBUG)) {
	errno = EINVAL;
	return NULL;
	}

	/*
	* allocate handle
	*/

	knet_h = malloc(sizeof(struct knet_handle));
	if (!knet_h) {
	errno = ENOMEM;
	return NULL;
	}
	memset(knet_h, 0, sizeof(struct knet_handle));

	savederrno = pthread_mutex_lock(&handle_config_mutex);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get handle mutex lock: %s",
	strerror(savederrno));
	errno = savederrno;
	goto exit_fail;
	}

	/*
	* copy config in place
	*/

	knet_h->host_id = host_id;
	knet_h->logfd = log_fd;
	if (knet_h->logfd > 0) {
	memset(&knet_h->log_levels, default_log_level, KNET_MAX_SUBSYSTEMS);
	}

	/*
	* set pmtud default timers
	*/
	knet_h->pmtud_interval = KNET_PMTUD_DEFAULT_INTERVAL;

	/*
	* set transports reconnect default timers
	*/
	knet_h->reconnect_int = KNET_TRANSPORT_DEFAULT_RECONNECT_INTERVAL;

	/*
	* init main locking structures
	*/

	if (_init_locks(knet_h)) {
	savederrno = errno;
	goto exit_fail;
	}

	/*
	* init sockets
	*/

	if (_init_socks(knet_h)) {
	savederrno = errno;
	goto exit_fail;
	}

	/*
	* allocate packet buffers
	*/

	if (_init_buffers(knet_h)) {
	savederrno = errno;
	goto exit_fail;
	}

	/*
	* create epoll fds
	*/

	if (_init_epolls(knet_h)) {
	savederrno = errno;
	goto exit_fail;
	}

	/*
	* start transports
	*/

	if (_start_transports(knet_h)) {
	savederrno = errno;
	goto exit_fail;
	}

	/*
	* start internal threads
	*/

	if (_start_threads(knet_h)) {
	savederrno = errno;
	goto exit_fail;
	}

	pthread_mutex_unlock(&handle_config_mutex);
	return knet_h;

	exit_fail:
	pthread_mutex_unlock(&handle_config_mutex);
	knet_handle_free(knet_h);
	errno = savederrno;
	return NULL;
	}

	int knet_handle_free(knet_handle_t knet_h)
	{
	int savederrno = 0;

	savederrno = pthread_mutex_lock(&handle_config_mutex);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get handle mutex lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	if (!knet_h) {
	pthread_mutex_unlock(&handle_config_mutex);
	errno = EINVAL;
	return -1;
	}

	if (!knet_h->lock_init_done) {
	goto exit_nolock;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	pthread_mutex_unlock(&handle_config_mutex);
	errno = savederrno;
	return -1;
	}

	if (knet_h->host_head != NULL) {
	savederrno = EBUSY;
	log_err(knet_h, KNET_SUB_HANDLE,
	"Unable to free handle: host(s) or listener(s) are still active: %s",
	strerror(savederrno));
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	pthread_mutex_unlock(&handle_config_mutex);
	errno = savederrno;
	return -1;
	}

	knet_h->fini_in_progress = 1;

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	_stop_threads(knet_h);
	_stop_transports(knet_h);
	_close_epolls(knet_h);
	_destroy_buffers(knet_h);
	_close_socks(knet_h);
	crypto_fini(knet_h);

	_destroy_locks(knet_h);

	exit_nolock:
	free(knet_h);
	knet_h = NULL;
	pthread_mutex_unlock(&handle_config_mutex);
	return 0;
	}

	int knet_handle_enable_sock_notify(knet_handle_t knet_h,
	void *sock_notify_fn_private_data,
	void (*sock_notify_fn) (
	void *private_data,
	int datafd,
	int8_t channel,
	uint8_t tx_rx,
	int error,
	int errorno))
	{
	int savederrno = 0, err = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (!sock_notify_fn) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	knet_h->sock_notify_fn_private_data = sock_notify_fn_private_data;
	knet_h->sock_notify_fn = sock_notify_fn;
	log_debug(knet_h, KNET_SUB_HANDLE, "sock_notify_fn enabled");

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	return err;
	}

	int knet_handle_add_datafd(knet_handle_t knet_h, int datafd, int8_t channel)
	{
	int err = 0, savederrno = 0;
	int i;
	struct epoll_event ev;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (datafd == NULL) {
	errno = EINVAL;
	return -1;
	}

	if (channel == NULL) {
	errno = EINVAL;
	return -1;
	}

	if (*channel >= KNET_DATAFD_MAX) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	if (!knet_h->sock_notify_fn) {
	log_err(knet_h, KNET_SUB_HANDLE, "Adding datafd requires sock notify callback enabled!");
	savederrno = EINVAL;
	err = -1;
	goto out_unlock;
	}

	if (*datafd > 0) {
	for (i = 0; i < KNET_DATAFD_MAX; i++) {
	if ((knet_h->sockfd[i].in_use) && (knet_h->sockfd[i].sockfd[0] == *datafd)) {
	log_err(knet_h, KNET_SUB_HANDLE, "requested datafd: %d already exist in index: %d", *datafd, i);
	savederrno = EEXIST;
	err = -1;
	goto out_unlock;
	}
	}
	}

	/*
	* auto allocate a channel
	*/
	if (*channel < 0) {
	for (i = 0; i < KNET_DATAFD_MAX; i++) {
	if (!knet_h->sockfd[i].in_use) {
	*channel = i;
	break;
	}
	}
	if (*channel < 0) {
	savederrno = EBUSY;
	err = -1;
	goto out_unlock;
	}
	} else {
	if (knet_h->sockfd[*channel].in_use) {
	savederrno = EBUSY;
	err = -1;
	goto out_unlock;
	}
	}

	knet_h->sockfd[*channel].is_created = 0;
	knet_h->sockfd[*channel].is_socket = 0;
	knet_h->sockfd[*channel].has_error = 0;

	if (*datafd > 0) {
	int sockopt;
	socklen_t sockoptlen = sizeof(sockopt);

	if (_fdset_cloexec(*datafd)) {
	savederrno = errno;
	err = -1;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to set CLOEXEC on datafd: %s",
	strerror(savederrno));
	goto out_unlock;
	}

	if (_fdset_nonblock(*datafd)) {
	savederrno = errno;
	err = -1;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to set NONBLOCK on datafd: %s",
	strerror(savederrno));
	goto out_unlock;
	}

	knet_h->sockfd[channel].sockfd[0] = datafd;
	knet_h->sockfd[*channel].sockfd[1] = 0;

	if (!getsockopt(knet_h->sockfd[*channel].sockfd[0], SOL_SOCKET, SO_TYPE, &sockopt, &sockoptlen)) {
	knet_h->sockfd[*channel].is_socket = 1;
	}
	} else {
	if (_init_socketpair(knet_h, knet_h->sockfd[*channel].sockfd)) {
	savederrno = errno;
	err = -1;
	goto out_unlock;
	}

	knet_h->sockfd[*channel].is_created = 1;
	knet_h->sockfd[*channel].is_socket = 1;
	datafd = knet_h->sockfd[channel].sockfd[0];
	}

	memset(&ev, 0, sizeof(struct epoll_event));
	ev.events = EPOLLIN;
	ev.data.fd = knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created];

	if (epoll_ctl(knet_h->send_to_links_epollfd,
	EPOLL_CTL_ADD, knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], &ev)) {
	savederrno = errno;
	err = -1;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to add datafd %d to linkfd epoll pool: %s",
	knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], strerror(savederrno));
	if (knet_h->sockfd[*channel].is_created) {
	_close_socketpair(knet_h, knet_h->sockfd[*channel].sockfd);
	}
	goto out_unlock;
	}

	knet_h->sockfd[*channel].in_use = 1;

	out_unlock:
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	errno = savederrno;
	return err;
	}

	int knet_handle_remove_datafd(knet_handle_t knet_h, int datafd)
	{
	int err = 0, savederrno = 0;
	int8_t channel = -1;
	int i;
	struct epoll_event ev;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (datafd <= 0) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	for (i = 0; i < KNET_DATAFD_MAX; i++) {
	if ((knet_h->sockfd[i].in_use) &&
	(knet_h->sockfd[i].sockfd[0] == datafd)) {
	channel = i;
	break;
	}
	}

	if (channel < 0) {
	savederrno = EINVAL;
	err = -1;
	goto out_unlock;
	}

	if (!knet_h->sockfd[channel].has_error) {
	memset(&ev, 0, sizeof(struct epoll_event));

	if (epoll_ctl(knet_h->send_to_links_epollfd,
	EPOLL_CTL_DEL, knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], &ev)) {
	savederrno = errno;
	err = -1;
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to del datafd %d from linkfd epoll pool: %s",
	knet_h->sockfd[channel].sockfd[0], strerror(savederrno));
	goto out_unlock;
	}
	}

	if (knet_h->sockfd[channel].is_created) {
	_close_socketpair(knet_h, knet_h->sockfd[channel].sockfd);
	}

	memset(&knet_h->sockfd[channel], 0, sizeof(struct knet_sock));

	out_unlock:
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	errno = savederrno;
	return err;
	}

	int knet_handle_get_datafd(knet_handle_t knet_h, const int8_t channel, int *datafd)
	{
	int err = 0, savederrno = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if ((channel < 0) \|\| (channel >= KNET_DATAFD_MAX)) {
	errno = EINVAL;
	return -1;
	}

	if (datafd == NULL) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	if (!knet_h->sockfd[channel].in_use) {
	savederrno = EINVAL;
	err = -1;
	goto out_unlock;
	}

	*datafd = knet_h->sockfd[channel].sockfd[0];

	out_unlock:
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	errno = savederrno;
	return err;
	}

	int knet_handle_get_channel(knet_handle_t knet_h, const int datafd, int8_t *channel)
	{
	int err = 0, savederrno = 0;
	int i;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (datafd <= 0) {
	errno = EINVAL;
	return -1;
	}

	if (channel == NULL) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	*channel = -1;

	for (i = 0; i < KNET_DATAFD_MAX; i++) {
	if ((knet_h->sockfd[i].in_use) &&
	(knet_h->sockfd[i].sockfd[0] == datafd)) {
	*channel = i;
	break;
	}
	}

	if (*channel < 0) {
	savederrno = EINVAL;
	err = -1;
	goto out_unlock;
	}

	out_unlock:
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	errno = savederrno;
	return err;
	}

	int knet_handle_enable_filter(knet_handle_t knet_h,
	void *dst_host_filter_fn_private_data,
	int (*dst_host_filter_fn) (
	void *private_data,
	const unsigned char *outdata,
	ssize_t outdata_len,
	uint8_t tx_rx,
	knet_node_id_t this_host_id,
	knet_node_id_t src_node_id,
	int8_t *channel,
	knet_node_id_t *dst_host_ids,
	size_t *dst_host_ids_entries))
	{
	int savederrno = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	knet_h->dst_host_filter_fn_private_data = dst_host_filter_fn_private_data;
	knet_h->dst_host_filter_fn = dst_host_filter_fn;
	if (knet_h->dst_host_filter_fn) {
	log_debug(knet_h, KNET_SUB_HANDLE, "dst_host_filter_fn enabled");
	} else {
	log_debug(knet_h, KNET_SUB_HANDLE, "dst_host_filter_fn disabled");
	}

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	return 0;
	}

	int knet_handle_setfwd(knet_handle_t knet_h, unsigned int enabled)
	{
	int savederrno = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (enabled > 1) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	knet_h->enabled = enabled;

	if (enabled) {
	log_debug(knet_h, KNET_SUB_HANDLE, "Data forwarding is enabled");
	} else {
	log_debug(knet_h, KNET_SUB_HANDLE, "Data forwarding is disabled");
	}

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	return 0;
	}

	int knet_handle_pmtud_getfreq(knet_handle_t knet_h, unsigned int *interval)
	{
	int savederrno = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (!interval) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	*interval = knet_h->pmtud_interval;

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	return 0;
	}

	int knet_handle_pmtud_setfreq(knet_handle_t knet_h, unsigned int interval)
	{
	int savederrno = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if ((!interval) \|\| (interval > 86400)) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	knet_h->pmtud_interval = interval;
	log_debug(knet_h, KNET_SUB_HANDLE, "PMTUd interval set to: %u seconds", interval);

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	return 0;
	}

	int knet_handle_enable_pmtud_notify(knet_handle_t knet_h,
	void *pmtud_notify_fn_private_data,
	void (*pmtud_notify_fn) (
	void *private_data,
	unsigned int data_mtu))
	{
	int savederrno = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	knet_h->pmtud_notify_fn_private_data = pmtud_notify_fn_private_data;
	knet_h->pmtud_notify_fn = pmtud_notify_fn;
	if (knet_h->pmtud_notify_fn) {
	log_debug(knet_h, KNET_SUB_HANDLE, "pmtud_notify_fn enabled");
	} else {
	log_debug(knet_h, KNET_SUB_HANDLE, "pmtud_notify_fn disabled");
	}

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	return 0;
	}

	int knet_handle_pmtud_get(knet_handle_t knet_h,
	unsigned int *data_mtu)
	{
	int savederrno = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (!data_mtu) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	*data_mtu = knet_h->data_mtu;

	pthread_rwlock_unlock(&knet_h->global_rwlock);

	return 0;
	}

	int knet_handle_crypto(knet_handle_t knet_h, struct knet_handle_crypto_cfg *knet_handle_crypto_cfg)
	{
	int savederrno = 0;
	int err = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (!knet_handle_crypto_cfg) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_wrlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get write lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	crypto_fini(knet_h);

	if ((!strncmp("none", knet_handle_crypto_cfg->crypto_model, 4)) \|\|
	((!strncmp("none", knet_handle_crypto_cfg->crypto_cipher_type, 4)) &&
	(!strncmp("none", knet_handle_crypto_cfg->crypto_hash_type, 4)))) {
	log_debug(knet_h, KNET_SUB_CRYPTO, "crypto is not enabled");
	err = 0;
	goto exit_unlock;
	}

	if (knet_handle_crypto_cfg->private_key_len < KNET_MIN_KEY_LEN) {
	log_debug(knet_h, KNET_SUB_CRYPTO, "private key len too short (min %d): %u",
	KNET_MIN_KEY_LEN, knet_handle_crypto_cfg->private_key_len);
	savederrno = EINVAL;
	err = -1;
	goto exit_unlock;
	}

	if (knet_handle_crypto_cfg->private_key_len > KNET_MAX_KEY_LEN) {
	log_debug(knet_h, KNET_SUB_CRYPTO, "private key len too long (max %d): %u",
	KNET_MAX_KEY_LEN, knet_handle_crypto_cfg->private_key_len);
	savederrno = EINVAL;
	err = -1;
	goto exit_unlock;
	}

	err = crypto_init(knet_h, knet_handle_crypto_cfg);

	if (err) {
	err = -2;
	}

	exit_unlock:
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	errno = savederrno;
	return err;
	}

	ssize_t knet_recv(knet_handle_t knet_h, char *buff, const size_t buff_len, const int8_t channel)
	{
	int savederrno = 0;
	ssize_t err = 0;
	struct iovec iov_in;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (buff == NULL) {
	errno = EINVAL;
	return -1;
	}

	if (buff_len <= 0) {
	errno = EINVAL;
	return -1;
	}

	if (buff_len > KNET_MAX_PACKET_SIZE) {
	errno = EINVAL;
	return -1;
	}

	if (channel < 0) {
	errno = EINVAL;
	return -1;
	}

	if (channel >= KNET_DATAFD_MAX) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	if (!knet_h->sockfd[channel].in_use) {
	savederrno = EINVAL;
	err = -1;
	goto out_unlock;
	}

	memset(&iov_in, 0, sizeof(iov_in));
	iov_in.iov_base = (void *)buff;
	iov_in.iov_len = buff_len;

	err = readv(knet_h->sockfd[channel].sockfd[0], &iov_in, 1);
	savederrno = errno;

	out_unlock:
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	errno = savederrno;
	return err;
	}

	ssize_t knet_send(knet_handle_t knet_h, const char *buff, const size_t buff_len, const int8_t channel)
	{
	int savederrno = 0;
	ssize_t err = 0;
	struct iovec iov_out[1];

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (buff == NULL) {
	errno = EINVAL;
	return -1;
	}

	if (buff_len <= 0) {
	errno = EINVAL;
	return -1;
	}

	if (buff_len > KNET_MAX_PACKET_SIZE) {
	errno = EINVAL;
	return -1;
	}

	if (channel < 0) {
	errno = EINVAL;
	return -1;
	}

	if (channel >= KNET_DATAFD_MAX) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_HANDLE, "Unable to get read lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	if (!knet_h->sockfd[channel].in_use) {
	savederrno = EINVAL;
	err = -1;
	goto out_unlock;
	}

	memset(iov_out, 0, sizeof(iov_out));

	iov_out[0].iov_base = (void *)buff;
	iov_out[0].iov_len = buff_len;

	err = writev(knet_h->sockfd[channel].sockfd[0], iov_out, 1);
	savederrno = errno;

	out_unlock:
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	errno = savederrno;
	return err;
	}
	diff --git a/libknet/internals.h b/libknet/internals.h
	index 59030ffa..91418384 100644
	--- a/libknet/internals.h
	+++ b/libknet/internals.h
	@@ -1,468 +1,468 @@
	/*
	* Copyright (C) 2010-2015 Red Hat, Inc. All rights reserved.
	*
	* Authors: Fabio M. Di Nitto <fabbione@kronosnet.org>
	* Federico Simoncelli <fsimon@kronosnet.org>
	*
	* This software licensed under GPL-2.0+, LGPL-2.0+
	*/

	#ifndef __KNET_INTERNALS_H__
	#define __KNET_INTERNALS_H__

	/*
	* NOTE: you shouldn't need to include this header normally
	*/

	#include "libknet.h"
	#include "onwire.h"
	#include "compat.h"

	#define KNET_DATABUFSIZE KNET_MAX_PACKET_SIZE + KNET_HEADER_ALL_SIZE
	#define KNET_DATABUFSIZE_CRYPT_PAD 1024
	#define KNET_DATABUFSIZE_CRYPT KNET_DATABUFSIZE + KNET_DATABUFSIZE_CRYPT_PAD

	#define KNET_RING_RCVBUFF 8388608

	#define PCKT_FRAG_MAX UINT8_MAX

	#define KNET_EPOLL_MAX_EVENTS KNET_DATAFD_MAX

	typedef void knet_transport_link_t; / per link transport handle */
	typedef void knet_transport_t; / per knet_h transport handle */
	struct knet_transport_ops; /* Forward because of circular dependancy */

	struct knet_mmsghdr {
	struct msghdr msg_hdr; /* Message header */
	unsigned int msg_len; /* Number of bytes transmitted */
	};

	struct knet_link {
	/* required */
	struct sockaddr_storage src_addr;
	struct sockaddr_storage dst_addr;
	/* configurable */
	unsigned int dynamic; /* see KNET_LINK_DYN_ define above */
	uint8_t priority; /* higher priority == preferred for A/P */
	unsigned long long ping_interval; /* interval */
	unsigned long long pong_timeout; /* timeout */
	unsigned int latency_fix; /* precision */
	uint8_t pong_count; /* how many ping/pong to send/receive before link is up */
	/* status */
	struct knet_link_status status;
	/* internals */
	uint8_t link_id;
	uint8_t transport_type; /* #defined constant from API */
	knet_transport_link_t transport_link; /* link_info_t from transport */
	int outsock;
	unsigned int configured:1; /* set to 1 if src/dst have been configured transport initialized on this link*/
	unsigned int transport_connected:1; /* set to 1 if lower level transport is connected */
	unsigned int latency_exp;
	uint8_t received_pong;
	struct timespec ping_last;
	/* used by PMTUD thread as temp per-link variables and should always contain the onwire_len value! */
	uint32_t proto_overhead;
	struct timespec pmtud_last;
	uint32_t last_ping_size;
	uint32_t last_good_mtu;
	uint32_t last_bad_mtu;
	uint32_t last_sent_mtu;
	uint32_t last_recv_mtu;
	uint8_t has_valid_mtu;
	};

	#define KNET_CBUFFER_SIZE 4096

	struct knet_host_defrag_buf {
	char buf[KNET_DATABUFSIZE];
	uint8_t in_use; /* 0 buffer is free, 1 is in use */
	seq_num_t pckt_seq; /* identify the pckt we are receiving */
	uint8_t frag_recv; /* how many frags did we receive */
	uint8_t frag_map[PCKT_FRAG_MAX];/* bitmap of what we received? */
	uint8_t last_first; /* special case if we receive the last fragment first */
	uint16_t frag_size; /* normal frag size (not the last one) */
	uint16_t last_frag_size; /* the last fragment might not be aligned with MTU size */
	struct timespec last_update; /* keep time of the last pckt */
	};

	struct knet_host {
	/* required */
	knet_node_id_t host_id;
	/* configurable */
	uint8_t link_handler_policy;
	char name[KNET_MAX_HOST_LEN];
	/* status */
	struct knet_host_status status;
	/* internals */
	char circular_buffer[KNET_CBUFFER_SIZE];
	seq_num_t rx_seq_num;
	seq_num_t untimed_rx_seq_num;
	seq_num_t timed_rx_seq_num;
	uint8_t got_data;
	/* defrag/reassembly buffers */
	struct knet_host_defrag_buf defrag_buf[KNET_MAX_LINK];
	char circular_buffer_defrag[KNET_CBUFFER_SIZE];
	/* link stuff */
	struct knet_link link[KNET_MAX_LINK];
	uint8_t active_link_entries;
	uint8_t active_links[KNET_MAX_LINK];
	struct knet_host *next;
	};

	struct knet_sock {
	int sockfd[2]; /* sockfd[0] will always be application facing
	* and sockfd[1] internal if sockpair has been created by knet */
	int is_socket; /* check if it's a socket for recvmmsg usage */
	int is_created; /* knet created this socket and has to clean up on exit/del */
	int in_use; /* set to 1 if it's use, 0 if free */
	int has_error; /* set to 1 if there were errors reading from the sock
	* and socket has been removed from epoll */
	};

	struct knet_fd_trackers {
	uint8_t transport; /* transport type (UDP/SCTP...) */
	uint8_t data_type; /* internal use for transport to define what data are associated
	* to this fd */
	void data; / pointer to the data */
	};

	#define KNET_MAX_FDS KNET_MAX_HOST * KNET_MAX_LINK * 4

	struct knet_handle {
	knet_node_id_t host_id;
	unsigned int enabled:1;
	struct knet_sock sockfd[KNET_DATAFD_MAX];
	int logfd;
	uint8_t log_levels[KNET_MAX_SUBSYSTEMS];
	int hostsockfd[2];
	int dstsockfd[2];
	int send_to_links_epollfd;
	int recv_from_links_epollfd;
	int dst_link_handler_epollfd;
	unsigned int pmtud_interval;
	unsigned int data_mtu; /* contains the max data size that we can send onwire
	* without frags */
	struct knet_host *host_head;
	struct knet_host *host_index[KNET_MAX_HOST];
	knet_transport_t transports[KNET_MAX_TRANSPORTS+1];
	struct knet_transport_ops *transport_ops[KNET_MAX_TRANSPORTS+1];
	struct knet_fd_trackers knet_transport_fd_tracker[KNET_MAX_FDS]; /* track status for each fd handled by transports */
	uint32_t reconnect_int;
	knet_node_id_t host_ids[KNET_MAX_HOST];
	size_t host_ids_entries;
	- struct knet_header *recv_from_sock_buf[PCKT_FRAG_MAX];
	+ struct knet_header *recv_from_sock_buf;
	struct knet_header *send_to_links_buf[PCKT_FRAG_MAX];
	struct knet_header *recv_from_links_buf[PCKT_FRAG_MAX];
	struct knet_header *pingbuf;
	struct knet_header *pmtudbuf;
	pthread_t send_to_links_thread;
	pthread_t recv_from_links_thread;
	pthread_t heartbt_thread;
	pthread_t dst_link_handler_thread;
	pthread_t pmtud_link_handler_thread;
	int lock_init_done;
	pthread_rwlock_t global_rwlock; /* global config lock */
	pthread_mutex_t pmtud_mutex; /* pmtud mutex to handle conditional send/recv + timeout */
	pthread_cond_t pmtud_cond; /* conditional for above */
	pthread_mutex_t tx_mutex; /* used to protect knet_send_sync and TX thread */
	pthread_mutex_t hb_mutex; /* used to protect heartbeat thread and seq_num broadcasting */
	struct crypto_instance *crypto_instance;
	uint16_t sec_header_size;
	uint16_t sec_block_size;
	uint16_t sec_hash_size;
	uint16_t sec_salt_size;
	unsigned char *send_to_links_buf_crypt[PCKT_FRAG_MAX];
	unsigned char *recv_from_links_buf_crypt;
	unsigned char *recv_from_links_buf_decrypt;
	unsigned char *pingbuf_crypt;
	unsigned char *pmtudbuf_crypt;
	seq_num_t tx_seq_num;
	pthread_mutex_t tx_seq_num_mutex;
	void *dst_host_filter_fn_private_data;
	int (*dst_host_filter_fn) (
	void *private_data,
	const unsigned char *outdata,
	ssize_t outdata_len,
	uint8_t tx_rx,
	knet_node_id_t this_host_id,
	knet_node_id_t src_node_id,
	int8_t *channel,
	knet_node_id_t *dst_host_ids,
	size_t *dst_host_ids_entries);
	void *pmtud_notify_fn_private_data;
	void (*pmtud_notify_fn) (
	void *private_data,
	unsigned int data_mtu);
	void *host_status_change_notify_fn_private_data;
	void (*host_status_change_notify_fn) (
	void *private_data,
	knet_node_id_t host_id,
	uint8_t reachable,
	uint8_t remote,
	uint8_t external);
	void *sock_notify_fn_private_data;
	void (*sock_notify_fn) (
	void *private_data,
	int datafd,
	int8_t channel,
	uint8_t tx_rx,
	int error,
	int errorno);
	int fini_in_progress;
	};

	/*
	* NOTE: every single operation must be implementend
	* for every protocol.
	*/

	typedef struct knet_transport_ops {
	/*
	* transport generic information
	*/
	const char *transport_name;
	const uint8_t transport_id;
	uint32_t transport_mtu_overhead;
	/*
	* transport init must allocate the new transport
	* and perform all internal initializations
	* (threads, lists, etc).
	*/
	int (*transport_init)(knet_handle_t knet_h);
	/*
	* transport free must releases _all_ resources
	* allocated by tranport_init
	*/
	int (*transport_free)(knet_handle_t knet_h);

	/*
	* link operations should take care of all the
	* sockets and epoll management for a given link/transport set
	* transport_link_disable should return err = -1 and errno = EBUSY
	* if listener is still in use, and any other errno in case
	* the link cannot be disabled.
	*
	* set_config/clear_config are invoked in global write lock context
	*/
	int (transport_link_set_config)(knet_handle_t knet_h, struct knet_link link);
	int (transport_link_clear_config)(knet_handle_t knet_h, struct knet_link link);

	/*
	* transport callback for incoming dynamic connections
	* this is called in global read lock context
	*/
	int (transport_link_dyn_connect)(knet_handle_t knet_h, int sockfd, struct knet_link link);

	/*
	* per transport error handling of recvmmsg
	* (see _handle_recv_from_links comments for details)
	*/

	/*
	* transport_rx_sock_error is invoked when recvmmsg returns <= 0
	*
	* transport_rx_sock_error is invoked with both global_rdlock
	*/

	int (*transport_rx_sock_error)(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno);

	/*
	* transport_tx_sock_error is invoked with global_rwlock and
	* it's invoked when sendto or sendmmsg returns =< 0
	*
	* it should return:
	* -1 on internal error
	* 0 ignore error and continue
	* 1 retry
	* any sleep or wait action should happen inside the transport code
	*/
	int (*transport_tx_sock_error)(knet_handle_t knet_h, int sockfd, int recv_err, int recv_errno);

	/*
	* this function is called on _every_ received packet
	* to verify if the packet is data or internal protocol error handling
	*
	* it should return:
	* -1 on error
	* 0 packet is not data and we should continue the packet process loop
	* 1 packet is not data and we should STOP the packet process loop
	* 2 packet is data and should be parsed as such
	*
	* transport_rx_is_data is invoked with both global_rwlock
	* and fd_tracker read lock (from RX thread)
	*/
	int (transport_rx_is_data)(knet_handle_t knet_h, int sockfd, struct knet_mmsghdr msg);
	} knet_transport_ops_t;

	socklen_t sockaddr_len(const struct sockaddr_storage *ss);

	/**
	* This is a kernel style list implementation.
	*
	* @author Steven Dake <sdake@redhat.com>
	*/

	struct knet_list_head {
	struct knet_list_head *next;
	struct knet_list_head *prev;
	};

	/**
	* @def KNET_LIST_DECLARE()
	* Declare and initialize a list head.
	*/
	#define KNET_LIST_DECLARE(name) \
	struct knet_list_head name = { &(name), &(name) }

	#define KNET_INIT_LIST_HEAD(ptr) do { \
	(ptr)->next = (ptr); (ptr)->prev = (ptr); \
	} while (0)

	/**
	* Initialize the list entry.
	*
	* Points next and prev pointers to head.
	* @param head pointer to the list head
	*/
	static inline void knet_list_init(struct knet_list_head *head)
	{
	head->next = head;
	head->prev = head;
	}

	/**
	* Add this element to the list.
	*
	* @param element the new element to insert.
	* @param head pointer to the list head
	*/
	static inline void knet_list_add(struct knet_list_head *element,
	struct knet_list_head *head)
	{
	head->next->prev = element;
	element->next = head->next;
	element->prev = head;
	head->next = element;
	}

	/**
	* Add to the list (but at the end of the list).
	*
	* @param element pointer to the element to add
	* @param head pointer to the list head
	* @see knet_list_add()
	*/
	static inline void knet_list_add_tail(struct knet_list_head *element,
	struct knet_list_head *head)
	{
	head->prev->next = element;
	element->next = head;
	element->prev = head->prev;
	head->prev = element;
	}

	/**
	* Delete an entry from the list.
	*
	* @param _remove the list item to remove
	*/
	static inline void knet_list_del(struct knet_list_head *_remove)
	{
	_remove->next->prev = _remove->prev;
	_remove->prev->next = _remove->next;
	}

	/**
	* Replace old entry by new one
	* @param old: the element to be replaced
	* @param new: the new element to insert
	*/
	static inline void knet_list_replace(struct knet_list_head *old,
	struct knet_list_head *new)
	{
	new->next = old->next;
	new->next->prev = new;
	new->prev = old->prev;
	new->prev->next = new;
	}

	/**
	* Tests whether list is the last entry in list head
	* @param list: the entry to test
	* @param head: the head of the list
	* @return boolean true/false
	*/
	static inline int knet_list_is_last(const struct knet_list_head *list,
	const struct knet_list_head *head)
	{
	return list->next == head;
	}

	/**
	* A quick test to see if the list is empty (pointing to it's self).
	* @param head pointer to the list head
	* @return boolean true/false
	*/
	static inline int32_t knet_list_empty(const struct knet_list_head *head)
	{
	return head->next == head;
	}


	/**
	* Get the struct for this entry
	* @param ptr: the &struct list_head pointer.
	* @param type: the type of the struct this is embedded in.
	* @param member: the name of the list_struct within the struct.
	*/
	#define knet_list_entry(ptr,type,member)\
	((type )((char )(ptr)-(char)(&((type )0)->member)))

	/**
	* Get the first element from a list
	* @param ptr: the &struct list_head pointer.
	* @param type: the type of the struct this is embedded in.
	* @param member: the name of the list_struct within the struct.
	*/
	#define knet_list_first_entry(ptr, type, member) \
	knet_list_entry((ptr)->next, type, member)

	/**
	* Iterate over a list
	* @param pos: the &struct list_head to use as a loop counter.
	* @param head: the head for your list.
	*/
	#define knet_list_for_each(pos, head) \
	for (pos = (head)->next; pos != (head); pos = pos->next)

	/**
	* Iterate over a list backwards
	* @param pos: the &struct list_head to use as a loop counter.
	* @param head: the head for your list.
	*/
	#define knet_list_for_each_reverse(pos, head) \
	for (pos = (head)->prev; pos != (head); pos = pos->prev)

	/**
	* Iterate over a list safe against removal of list entry
	* @param pos: the &struct list_head to use as a loop counter.
	* @param n: another &struct list_head to use as temporary storage
	* @param head: the head for your list.
	*/
	#define knet_list_for_each_safe(pos, n, head) \
	for (pos = (head)->next, n = pos->next; pos != (head); \
	pos = n, n = pos->next)

	/**
	* Iterate over list of given type
	* @param pos: the type * to use as a loop counter.
	* @param head: the head for your list.
	* @param member: the name of the list_struct within the struct.
	*/
	#define knet_list_for_each_entry(pos, head, member) \
	for (pos = knet_list_entry((head)->next, typeof(*pos), member); \
	&pos->member != (head); \
	pos = knet_list_entry(pos->member.next, typeof(*pos), member))


	#endif
	diff --git a/libknet/threads_tx.c b/libknet/threads_tx.c
	index e87cd6f5..7c754389 100644
	--- a/libknet/threads_tx.c
	+++ b/libknet/threads_tx.c
	@@ -1,611 +1,589 @@
	/*
	* Copyright (C) 2010-2017 Red Hat, Inc. All rights reserved.
	*
	* Authors: Fabio M. Di Nitto <fabbione@kronosnet.org>
	* Federico Simoncelli <fsimon@kronosnet.org>
	*
	* This software licensed under GPL-2.0+, LGPL-2.0+
	*/

	#include "config.h"

	#include <math.h>
	#include <string.h>
	#include <pthread.h>
	#include <sys/uio.h>
	#include <errno.h>

	#include "compat.h"
	#include "crypto.h"
	#include "host.h"
	#include "link.h"
	#include "logging.h"
	#include "transports.h"
	#include "threads_common.h"
	#include "threads_heartbeat.h"
	#include "threads_tx.h"
	#include "netutils.h"

	/*
	* SEND
	*/

	static int _dispatch_to_links(knet_handle_t knet_h, struct knet_host dst_host, struct knet_mmsghdr msg, int msgs_to_send)
	{
	int link_idx, msg_idx, sent_msgs, prev_sent, progress;
	int err = 0, savederrno = 0;
	struct knet_mmsghdr *cur;

	for (link_idx = 0; link_idx < dst_host->active_link_entries; link_idx++) {
	sent_msgs = 0;
	prev_sent = 0;
	progress = 1;

	msg_idx = 0;
	while (msg_idx < msgs_to_send) {
	msg[msg_idx].msg_hdr.msg_name = &dst_host->link[dst_host->active_links[link_idx]].dst_addr;
	msg_idx++;
	}

	retry:
	cur = &msg[prev_sent];

	sent_msgs = _sendmmsg(dst_host->link[dst_host->active_links[link_idx]].outsock,
	&cur[0], msgs_to_send - prev_sent, MSG_DONTWAIT \| MSG_NOSIGNAL);
	savederrno = errno;

	err = knet_h->transport_ops[dst_host->link[dst_host->active_links[link_idx]].transport_type]->transport_tx_sock_error(knet_h, dst_host->link[dst_host->active_links[link_idx]].outsock, sent_msgs, savederrno);
	switch(err) {
	case -1: /* unrecoverable error */
	goto out_unlock;
	break;
	case 0: /* ignore error and continue */
	break;
	case 1: /* retry to send those same data */
	goto retry;
	break;
	}

	prev_sent = prev_sent + sent_msgs;

	if ((sent_msgs >= 0) && (prev_sent < msgs_to_send)) {
	if ((sent_msgs) \|\| (progress)) {
	if (sent_msgs) {
	progress = 1;
	} else {
	progress = 0;
	}
	#ifdef DEBUG
	log_debug(knet_h, KNET_SUB_TX, "Unable to send all (%d/%d) data packets to host %s (%u) link %s:%s (%u)",
	sent_msgs, msg_idx,
	dst_host->name, dst_host->host_id,
	dst_host->link[dst_host->active_links[link_idx]].status.dst_ipaddr,
	dst_host->link[dst_host->active_links[link_idx]].status.dst_port,
	dst_host->link[dst_host->active_links[link_idx]].link_id);
	#endif
	goto retry;
	}
	if (!progress) {
	savederrno = EAGAIN;
	err = -1;
	goto out_unlock;
	}
	}

	if ((dst_host->link_handler_policy == KNET_LINK_POLICY_RR) &&
	(dst_host->active_link_entries > 1)) {
	uint8_t cur_link_id = dst_host->active_links[0];

	memmove(&dst_host->active_links[0], &dst_host->active_links[1], KNET_MAX_LINK - 1);
	dst_host->active_links[dst_host->active_link_entries - 1] = cur_link_id;

	break;
	}
	}

	out_unlock:
	errno = savederrno;
	return err;
	}

	-static int _parse_recv_from_sock(knet_handle_t knet_h, int buf_idx, ssize_t inlen, int8_t channel, int is_sync)
	+static int _parse_recv_from_sock(knet_handle_t knet_h, ssize_t inlen, int8_t channel, int is_sync)
	{
	ssize_t outlen, frag_len;
	struct knet_host *dst_host;
	knet_node_id_t dst_host_ids_temp[KNET_MAX_HOST];
	size_t dst_host_ids_entries_temp = 0;
	knet_node_id_t dst_host_ids[KNET_MAX_HOST];
	size_t dst_host_ids_entries = 0;
	int bcast = 1;
	struct knet_hostinfo *knet_hostinfo;
	struct iovec iov_out[PCKT_FRAG_MAX][2];
	int iovcnt_out = 2;
	uint8_t frag_idx;
	unsigned int temp_data_mtu;
	size_t host_idx;
	int send_mcast = 0;
	struct knet_header *inbuf;
	int savederrno = 0;
	int err = 0;
	seq_num_t tx_seq_num;
	struct knet_mmsghdr msg[PCKT_FRAG_MAX];
	int msgs_to_send, msg_idx;

	- inbuf = knet_h->recv_from_sock_buf[buf_idx];
	+ inbuf = knet_h->recv_from_sock_buf;

	if ((knet_h->enabled != 1) &&
	(inbuf->kh_type != KNET_HEADER_TYPE_HOST_INFO)) { /* data forward is disabled */
	log_debug(knet_h, KNET_SUB_TX, "Received data packet but forwarding is disabled");
	savederrno = ECANCELED;
	err = -1;
	goto out_unlock;
	}

	/*
	* move this into a separate function to expand on
	* extra switching rules
	*/
	switch(inbuf->kh_type) {
	case KNET_HEADER_TYPE_DATA:
	if (knet_h->dst_host_filter_fn) {
	bcast = knet_h->dst_host_filter_fn(
	knet_h->dst_host_filter_fn_private_data,
	(const unsigned char *)inbuf->khp_data_userdata,
	inlen,
	KNET_NOTIFY_TX,
	knet_h->host_id,
	knet_h->host_id,
	&channel,
	dst_host_ids_temp,
	&dst_host_ids_entries_temp);
	if (bcast < 0) {
	log_debug(knet_h, KNET_SUB_TX, "Error from dst_host_filter_fn: %d", bcast);
	savederrno = EFAULT;
	err = -1;
	goto out_unlock;
	}

	if ((!bcast) && (!dst_host_ids_entries_temp)) {
	log_debug(knet_h, KNET_SUB_TX, "Message is unicast but no dst_host_ids_entries");
	savederrno = EINVAL;
	err = -1;
	goto out_unlock;
	}
	}
	break;
	case KNET_HEADER_TYPE_HOST_INFO:
	knet_hostinfo = (struct knet_hostinfo *)inbuf->khp_data_userdata;
	if (knet_hostinfo->khi_bcast == KNET_HOSTINFO_UCAST) {
	bcast = 0;
	dst_host_ids_temp[0] = knet_hostinfo->khi_dst_node_id;
	dst_host_ids_entries_temp = 1;
	knet_hostinfo->khi_dst_node_id = htons(knet_hostinfo->khi_dst_node_id);
	}
	break;
	default:
	log_warn(knet_h, KNET_SUB_TX, "Receiving unknown messages from socket");
	savederrno = ENOMSG;
	err = -1;
	goto out_unlock;
	break;
	}

	if (is_sync) {
	if ((bcast) \|\|
	((!bcast) && (dst_host_ids_entries_temp > 1))) {
	log_debug(knet_h, KNET_SUB_TX, "knet_send_sync is only supported with unicast packets for one destination");
	savederrno = E2BIG;
	err = -1;
	goto out_unlock;
	}
	}

	/*
	* check destinations hosts before spending time
	* in fragmenting/encrypting packets to save
	* time processing data for unrechable hosts.
	* for unicast, also remap the destination data
	* to skip unreachable hosts.
	*/

	if (!bcast) {
	dst_host_ids_entries = 0;
	for (host_idx = 0; host_idx < dst_host_ids_entries_temp; host_idx++) {
	dst_host = knet_h->host_index[dst_host_ids_temp[host_idx]];
	if (!dst_host) {
	continue;
	}
	if (dst_host->status.reachable) {
	dst_host_ids[dst_host_ids_entries] = dst_host_ids_temp[host_idx];
	dst_host_ids_entries++;
	}
	}
	if (!dst_host_ids_entries) {
	savederrno = EHOSTDOWN;
	err = -1;
	goto out_unlock;
	}
	} else {
	send_mcast = 0;
	for (dst_host = knet_h->host_head; dst_host != NULL; dst_host = dst_host->next) {
	if (dst_host->status.reachable) {
	send_mcast = 1;
	break;
	}
	}
	if (!send_mcast) {
	savederrno = EHOSTDOWN;
	err = -1;
	goto out_unlock;
	}
	}

	if (!knet_h->data_mtu) {
	/*
	* using MIN_MTU_V4 for data mtu is not completely accurate but safe enough
	*/
	log_debug(knet_h, KNET_SUB_TX,
	"Received data packet but data MTU is still unknown."
	" Packet might not be delivered."
	" Assuming mininum IPv4 mtu (%d)",
	KNET_PMTUD_MIN_MTU_V4);
	temp_data_mtu = KNET_PMTUD_MIN_MTU_V4;
	} else {
	/*
	* take a copy of the mtu to avoid value changing under
	* our feet while we are sending a fragmented pckt
	*/
	temp_data_mtu = knet_h->data_mtu;
	}

	/*
	* prepare the outgoing buffers
	*/

	frag_len = inlen;
	frag_idx = 0;

	inbuf->khp_data_bcast = bcast;
	inbuf->khp_data_frag_num = ceil((float)inlen / temp_data_mtu);
	inbuf->khp_data_channel = channel;

	if (pthread_mutex_lock(&knet_h->tx_seq_num_mutex)) {
	log_debug(knet_h, KNET_SUB_TX, "Unable to get seq mutex lock");
	goto out_unlock;
	}
	knet_h->tx_seq_num++;
	/*
	* force seq_num 0 to detect a node that has crashed and rejoining
	* the knet instance. seq_num 0 will clear the buffers in the RX
	* thread
	*/
	if (knet_h->tx_seq_num == 0) {
	knet_h->tx_seq_num++;
	}
	/*
	* cache the value in locked context
	*/
	tx_seq_num = knet_h->tx_seq_num;
	inbuf->khp_data_seq_num = htons(knet_h->tx_seq_num);
	pthread_mutex_unlock(&knet_h->tx_seq_num_mutex);

	/*
	* forcefully broadcast a ping to all nodes every SEQ_MAX / 8
	* pckts.
	* this solves 2 problems:
	* 1) on TX socket overloads we generate extra pings to keep links alive
	* 2) in 3+ nodes setup, where all the traffic is flowing between node 1 and 2,
	* node 3+ will be able to keep in sync on the TX seq_num even without
	* receiving traffic or pings in betweens. This avoids issues with
	* rollover of the circular buffer
	*/

	if (tx_seq_num % (SEQ_MAX / 8) == 0) {
	_send_pings(knet_h, 0);
	}

	if (inbuf->khp_data_frag_num > 1) {
	while (frag_idx < inbuf->khp_data_frag_num) {
	/*
	* set the iov_base
	*/
	iov_out[frag_idx][0].iov_base = (void *)knet_h->send_to_links_buf[frag_idx];
	iov_out[frag_idx][0].iov_len = KNET_HEADER_DATA_SIZE;
	iov_out[frag_idx][1].iov_base = inbuf->khp_data_userdata + (temp_data_mtu * frag_idx);

	/*
	* set the len
	*/
	if (frag_len > temp_data_mtu) {
	iov_out[frag_idx][1].iov_len = temp_data_mtu;
	} else {
	iov_out[frag_idx][1].iov_len = frag_len;
	}

	/*
	* copy the frag info on all buffers
	*/
	knet_h->send_to_links_buf[frag_idx]->kh_type = inbuf->kh_type;
	knet_h->send_to_links_buf[frag_idx]->khp_data_seq_num = inbuf->khp_data_seq_num;
	knet_h->send_to_links_buf[frag_idx]->khp_data_frag_num = inbuf->khp_data_frag_num;
	knet_h->send_to_links_buf[frag_idx]->khp_data_bcast = inbuf->khp_data_bcast;
	knet_h->send_to_links_buf[frag_idx]->khp_data_channel = inbuf->khp_data_channel;

	frag_len = frag_len - temp_data_mtu;
	frag_idx++;
	}
	iovcnt_out = 2;
	} else {
	iov_out[frag_idx][0].iov_base = (void *)inbuf;
	iov_out[frag_idx][0].iov_len = frag_len + KNET_HEADER_DATA_SIZE;
	iovcnt_out = 1;
	}

	if (knet_h->crypto_instance) {
	frag_idx = 0;
	while (frag_idx < inbuf->khp_data_frag_num) {
	if (crypto_encrypt_and_signv(
	knet_h,
	iov_out[frag_idx], iovcnt_out,
	knet_h->send_to_links_buf_crypt[frag_idx],
	&outlen) < 0) {
	log_debug(knet_h, KNET_SUB_TX, "Unable to encrypt packet");
	savederrno = ECHILD;
	err = -1;
	goto out_unlock;
	}
	iov_out[frag_idx][0].iov_base = knet_h->send_to_links_buf_crypt[frag_idx];
	iov_out[frag_idx][0].iov_len = outlen;
	frag_idx++;
	}
	iovcnt_out = 1;
	}

	memset(&msg, 0, sizeof(msg));

	msgs_to_send = inbuf->khp_data_frag_num;

	msg_idx = 0;

	while (msg_idx < msgs_to_send) {
	msg[msg_idx].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage);
	msg[msg_idx].msg_hdr.msg_iov = &iov_out[msg_idx][0];
	msg[msg_idx].msg_hdr.msg_iovlen = iovcnt_out;
	msg_idx++;
	}

	if (!bcast) {
	for (host_idx = 0; host_idx < dst_host_ids_entries; host_idx++) {
	dst_host = knet_h->host_index[dst_host_ids[host_idx]];

	err = _dispatch_to_links(knet_h, dst_host, &msg[0], msgs_to_send);
	savederrno = errno;
	if (err) {
	goto out_unlock;
	}
	}
	} else {
	for (dst_host = knet_h->host_head; dst_host != NULL; dst_host = dst_host->next) {
	if (dst_host->status.reachable) {
	err = _dispatch_to_links(knet_h, dst_host, &msg[0], msgs_to_send);
	savederrno = errno;
	if (err) {
	goto out_unlock;
	}
	}
	}
	}

	out_unlock:
	errno = savederrno;
	return err;
	}

	int knet_send_sync(knet_handle_t knet_h, const char *buff, const size_t buff_len, const int8_t channel)
	{
	int savederrno = 0, err = 0;

	if (!knet_h) {
	errno = EINVAL;
	return -1;
	}

	if (buff == NULL) {
	errno = EINVAL;
	return -1;
	}

	if (buff_len <= 0) {
	errno = EINVAL;
	return -1;
	}

	if (buff_len > KNET_MAX_PACKET_SIZE) {
	errno = EINVAL;
	return -1;
	}

	if (channel < 0) {
	errno = EINVAL;
	return -1;
	}

	if (channel >= KNET_DATAFD_MAX) {
	errno = EINVAL;
	return -1;
	}

	savederrno = pthread_rwlock_rdlock(&knet_h->global_rwlock);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_TX, "Unable to get read lock: %s",
	strerror(savederrno));
	errno = savederrno;
	return -1;
	}

	if (!knet_h->sockfd[channel].in_use) {
	savederrno = EINVAL;
	err = -1;
	goto out;
	}

	savederrno = pthread_mutex_lock(&knet_h->tx_mutex);
	if (savederrno) {
	log_err(knet_h, KNET_SUB_TX, "Unable to get TX mutex lock: %s",
	strerror(savederrno));
	err = -1;
	goto out;
	}

	- knet_h->recv_from_sock_buf[0]->kh_type = KNET_HEADER_TYPE_DATA;
	- memmove(knet_h->recv_from_sock_buf[0]->khp_data_userdata, buff, buff_len);
	- err = _parse_recv_from_sock(knet_h, 0, buff_len, channel, 1);
	+ knet_h->recv_from_sock_buf->kh_type = KNET_HEADER_TYPE_DATA;
	+ memmove(knet_h->recv_from_sock_buf->khp_data_userdata, buff, buff_len);
	+ err = _parse_recv_from_sock(knet_h, buff_len, channel, 1);
	savederrno = errno;

	pthread_mutex_unlock(&knet_h->tx_mutex);

	out:
	pthread_rwlock_unlock(&knet_h->global_rwlock);

	errno = savederrno;
	return err;
	}

	-static void _handle_send_to_links(knet_handle_t knet_h, int sockfd, int8_t channel, struct knet_mmsghdr *msg, int type)
	+static void _handle_send_to_links(knet_handle_t knet_h, struct msghdr *msg, int sockfd, int8_t channel, int type)
	{
	ssize_t inlen = 0;
	- struct iovec iov_in;
	- int msg_recv, i;
	int savederrno = 0, docallback = 0;

	if ((channel >= 0) &&
	(channel < KNET_DATAFD_MAX) &&
	(!knet_h->sockfd[channel].is_socket)) {
	- memset(&iov_in, 0, sizeof(iov_in));
	- iov_in.iov_base = (void *)knet_h->recv_from_sock_buf[0]->khp_data_userdata;
	- iov_in.iov_len = KNET_MAX_PACKET_SIZE;
	-
	- inlen = readv(sockfd, &iov_in, 1);
	-
	- if (inlen <= 0) {
	- savederrno = errno;
	- docallback = 1;
	- goto out;
	- }
	-
	- msg_recv = 1;
	- knet_h->recv_from_sock_buf[0]->kh_type = type;
	- _parse_recv_from_sock(knet_h, 0, inlen, channel, 0);
	+ inlen = readv(sockfd, msg->msg_iov, 1);
	} else {
	- msg_recv = _recvmmsg(sockfd, &msg[0], PCKT_FRAG_MAX, MSG_DONTWAIT \| MSG_NOSIGNAL);
	- if (msg_recv < 0) {
	- inlen = msg_recv;
	- savederrno = errno;
	- docallback = 1;
	- goto out;
	- }
	- for (i = 0; i < msg_recv; i++) {
	- inlen = msg[i].msg_len;
	- if (inlen == 0) {
	- savederrno = 0;
	- docallback = 1;
	- goto out;
	- break;
	- }
	- knet_h->recv_from_sock_buf[i]->kh_type = type;
	- _parse_recv_from_sock(knet_h, i, inlen, channel, 0);
	- }
	+ inlen = recvmsg(sockfd, msg, MSG_DONTWAIT \| MSG_NOSIGNAL);
	+ }
	+
	+ if (inlen == 0) {
	+ savederrno = 0;
	+ docallback = 1;
	+ goto out;
	}
	+ if (inlen < 0) {
	+ savederrno = errno;
	+ docallback = 1;
	+ goto out;
	+ }
	+
	+ knet_h->recv_from_sock_buf->kh_type = type;
	+ _parse_recv_from_sock(knet_h, inlen, channel, 0);

	out:
	if (inlen < 0) {
	struct epoll_event ev;

	memset(&ev, 0, sizeof(struct epoll_event));

	if (epoll_ctl(knet_h->send_to_links_epollfd,
	EPOLL_CTL_DEL, knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created], &ev)) {
	log_err(knet_h, KNET_SUB_TX, "Unable to del datafd %d from linkfd epoll pool: %s",
	knet_h->sockfd[channel].sockfd[0], strerror(savederrno));
	} else {
	knet_h->sockfd[channel].has_error = 1;
	}

	}

	if (docallback) {
	knet_h->sock_notify_fn(knet_h->sock_notify_fn_private_data,
	knet_h->sockfd[channel].sockfd[0],
	channel,
	KNET_NOTIFY_TX,
	inlen,
	savederrno);
	}
	}

	void _handle_send_to_links_thread(void data)
	{
	knet_handle_t knet_h = (knet_handle_t) data;
	struct epoll_event events[KNET_EPOLL_MAX_EVENTS];
	- struct sockaddr_storage address[PCKT_FRAG_MAX];
	- struct knet_mmsghdr msg[PCKT_FRAG_MAX];
	- struct iovec iov_in[PCKT_FRAG_MAX];
	int i, nev, type;
	int8_t channel;
	+ struct iovec iov_in;
	+ struct msghdr msg;
	+ struct sockaddr_storage address;

	- memset(&msg, 0, sizeof(msg));
	-
	- /* preparing data buffer */
	- for (i = 0; i < PCKT_FRAG_MAX; i++) {
	- iov_in[i].iov_base = (void *)knet_h->recv_from_sock_buf[i]->khp_data_userdata;
	- iov_in[i].iov_len = KNET_MAX_PACKET_SIZE;
	-
	- memset(&msg[i].msg_hdr, 0, sizeof(struct msghdr));
	+ memset(&iov_in, 0, sizeof(iov_in));
	+ iov_in.iov_base = (void *)knet_h->recv_from_sock_buf->khp_data_userdata;
	+ iov_in.iov_len = KNET_MAX_PACKET_SIZE;

	- msg[i].msg_hdr.msg_name = &address[i];
	- msg[i].msg_hdr.msg_namelen = sizeof(struct sockaddr_storage);
	- msg[i].msg_hdr.msg_iov = &iov_in[i];
	- msg[i].msg_hdr.msg_iovlen = 1;
	+ memset(&msg, 0, sizeof(struct msghdr));
	+ msg.msg_name = &address;
	+ msg.msg_namelen = sizeof(struct sockaddr_storage);
	+ msg.msg_iov = &iov_in;
	+ msg.msg_iovlen = 1;

	- knet_h->recv_from_sock_buf[i]->kh_version = KNET_HEADER_VERSION;
	- knet_h->recv_from_sock_buf[i]->khp_data_frag_seq = 0;
	- knet_h->recv_from_sock_buf[i]->kh_node = htons(knet_h->host_id);
	+ knet_h->recv_from_sock_buf->kh_version = KNET_HEADER_VERSION;
	+ knet_h->recv_from_sock_buf->khp_data_frag_seq = 0;
	+ knet_h->recv_from_sock_buf->kh_node = htons(knet_h->host_id);

	+ for (i = 0; i < PCKT_FRAG_MAX; i++) {
	knet_h->send_to_links_buf[i]->kh_version = KNET_HEADER_VERSION;
	knet_h->send_to_links_buf[i]->khp_data_frag_seq = i + 1;
	knet_h->send_to_links_buf[i]->kh_node = htons(knet_h->host_id);
	}

	while (!shutdown_in_progress(knet_h)) {
	nev = epoll_wait(knet_h->send_to_links_epollfd, events, KNET_EPOLL_MAX_EVENTS + 1, -1);

	if (pthread_rwlock_rdlock(&knet_h->global_rwlock) != 0) {
	log_debug(knet_h, KNET_SUB_TX, "Unable to get read lock");
	continue;
	}

	for (i = 0; i < nev; i++) {
	if (events[i].data.fd == knet_h->hostsockfd[0]) {
	type = KNET_HEADER_TYPE_HOST_INFO;
	channel = -1;
	} else {
	type = KNET_HEADER_TYPE_DATA;
	for (channel = 0; channel < KNET_DATAFD_MAX; channel++) {
	if ((knet_h->sockfd[channel].in_use) &&
	(knet_h->sockfd[channel].sockfd[knet_h->sockfd[channel].is_created] == events[i].data.fd)) {
	break;
	}
	}
	}
	if (pthread_mutex_lock(&knet_h->tx_mutex) != 0) {
	log_debug(knet_h, KNET_SUB_TX, "Unable to get mutex lock");
	continue;
	}
	- _handle_send_to_links(knet_h, events[i].data.fd, channel, &msg[0], type);
	+ _handle_send_to_links(knet_h, &msg, events[i].data.fd, channel, type);
	pthread_mutex_unlock(&knet_h->tx_mutex);
	}
	pthread_rwlock_unlock(&knet_h->global_rwlock);
	}

	return NULL;
	}

File Metadata

Mime Type: text/x-diff
Expires: Mon, Apr 21, 2:18 PM (1 d, 7 h)
Storage Engine: blob
Storage Format: Raw Data
Storage Handle: 1664909
Default Alt Text: (66 KB)

No OneTemporaryActions

View Options

File Metadata

Event Timeline

No OneTemporary
Actions