n2n/edge_utils.c
realjiangms 9601e82daf More reliable punching:
1. Some nat router blocks the port if incoming traffic arrives before outcoming traffic being sent. Give edge ability to set proper TTL so that the registration packet is dropped before it arrives peer.
2. Support Symmetric NAT by predicting 15 more ports when sending registration packet
3. Purge pending mac also on P2P normal packet. This is actually more usual condition.
4. Add doc for new flag.
2019-11-08 09:21:10 +08:00

1983 lines
59 KiB
C

/**
* (C) 2007-18 - ntop.org and contributors
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not see see <http://www.gnu.org/licenses/>
*
*/
#include "n2n.h"
#include "lzoconf.h"
#ifdef WIN32
#include <process.h>
/* Multicast peers discovery disabled due to https://github.com/ntop/n2n/issues/65 */
#define SKIP_MULTICAST_PEERS_DISCOVERY
#endif
#ifdef __ANDROID_NDK__
#include "android/edge_android.h"
#include <tun2tap/tun2tap.h>
#endif /* __ANDROID_NDK__ */
#define SOCKET_TIMEOUT_INTERVAL_SECS 10
#define REGISTER_SUPER_INTERVAL_DFL 20 /* sec, usually UDP NAT entries in a firewall expire after 30 seconds */
#define IFACE_UPDATE_INTERVAL (30) /* sec. How long it usually takes to get an IP lease. */
#define TRANSOP_TICK_INTERVAL (10) /* sec */
#ifdef __ANDROID_NDK__
#define ARP_PERIOD_INTERVAL (10) /* sec */
#endif
#define ETH_FRAMESIZE 14
#define IP4_SRCOFFSET 12
#define IP4_DSTOFFSET 16
#define IP4_MIN_SIZE 20
#define UDP_SIZE 8
/* ************************************** */
static const char * supernode_ip(const n2n_edge_t * eee);
static void send_register(n2n_edge_t *eee, const n2n_sock_t *remote_peer, const n2n_mac_t peer_mac);
static void check_peer_registration_needed(n2n_edge_t * eee,
uint8_t from_supernode,
const n2n_mac_t mac,
const n2n_sock_t * peer);
static int edge_init_sockets(n2n_edge_t *eee, int udp_local_port, int mgmt_port, uint8_t tos);
static void supernode2addr(n2n_sock_t * sn, const n2n_sn_name_t addrIn);
static void check_known_peer_sock_change(n2n_edge_t * eee,
uint8_t from_supernode,
const n2n_mac_t mac,
const n2n_sock_t * peer,
time_t when);
/* ************************************** */
int edge_verify_conf(const n2n_edge_conf_t *conf) {
if(conf->community_name[0] == 0)
return(-1);
if(conf->sn_num == 0)
return(-2);
if(conf->register_interval < 1)
return(-3);
if(((conf->encrypt_key == NULL) && (conf->transop_id != N2N_TRANSFORM_ID_NULL)) ||
((conf->encrypt_key != NULL) && (conf->transop_id == N2N_TRANSFORM_ID_NULL)))
return(-4);
return(0);
}
/* ************************************** */
struct n2n_edge_stats {
uint32_t tx_p2p;
uint32_t rx_p2p;
uint32_t tx_sup;
uint32_t rx_sup;
uint32_t tx_sup_broadcast;
uint32_t rx_sup_broadcast;
};
/* ************************************** */
struct n2n_edge {
n2n_edge_conf_t conf;
/* Status */
uint8_t sn_idx; /**< Currently active supernode. */
uint8_t sn_wait; /**< Whether we are waiting for a supernode response. */
size_t sup_attempts; /**< Number of remaining attempts to this supernode. */
tuntap_dev device; /**< All about the TUNTAP device */
n2n_trans_op_t transop; /**< The transop to use when encoding */
n2n_cookie_t last_cookie; /**< Cookie sent in last REGISTER_SUPER. */
/* Sockets */
n2n_sock_t supernode;
int udp_sock;
int udp_mgmt_sock; /**< socket for status info. */
#ifndef SKIP_MULTICAST_PEERS_DISCOVERY
n2n_sock_t multicast_peer; /**< Multicast peer group (for local edges) */
int udp_multicast_sock; /**< socket for local multicast registrations. */
int multicast_joined; /**< 1 if the group has been joined.*/
#endif
/* Peers */
struct peer_info * known_peers; /**< Edges we are connected to. */
struct peer_info * pending_peers; /**< Edges we have tried to register with. */
/* Timers */
time_t last_register_req; /**< Check if time to re-register with super*/
time_t last_p2p; /**< Last time p2p traffic was received. */
time_t last_sup; /**< Last time a packet arrived from supernode. */
time_t start_time; /**< For calculating uptime */
/* Statistics */
struct n2n_edge_stats stats;
};
/* ************************************** */
static const char* transop_str(enum n2n_transform tr) {
switch(tr) {
case N2N_TRANSFORM_ID_NULL: return("null");
case N2N_TRANSFORM_ID_TWOFISH: return("twofish");
case N2N_TRANSFORM_ID_AESCBC: return("AES-CBC");
default: return("invalid");
};
}
/* ************************************** */
/** Initialise an edge to defaults.
*
* This also initialises the NULL transform operation opstruct.
*/
n2n_edge_t* edge_init(const tuntap_dev *dev, const n2n_edge_conf_t *conf, int *rv) {
n2n_transform_t transop_id = conf->transop_id;
n2n_edge_t *eee = calloc(1, sizeof(n2n_edge_t));
int rc = -1, i;
if((rc = edge_verify_conf(conf)) != 0) {
traceEvent(TRACE_ERROR, "Invalid configuration");
goto edge_init_error;
}
if(!eee) {
traceEvent(TRACE_ERROR, "Cannot allocate memory");
goto edge_init_error;
}
#ifdef WIN32
initWin32();
#endif
memcpy(&eee->conf, conf, sizeof(*conf));
memcpy(&eee->device, dev, sizeof(*dev));
eee->start_time = time(NULL);
eee->known_peers = NULL;
eee->pending_peers = NULL;
eee->sup_attempts = N2N_EDGE_SUP_ATTEMPTS;
#ifdef NOT_USED
if(lzo_init() != LZO_E_OK) {
traceEvent(TRACE_ERROR, "LZO compression error");
goto edge_init_error;
}
#endif
for(i=0; i<conf->sn_num; ++i)
traceEvent(TRACE_NORMAL, "supernode %u => %s\n", i, (conf->sn_ip_array[i]));
/* Set the active supernode */
supernode2addr(&(eee->supernode), conf->sn_ip_array[eee->sn_idx]);
/* Set active transop */
switch(transop_id) {
case N2N_TRANSFORM_ID_TWOFISH:
rc = n2n_transop_twofish_init(&eee->conf, &eee->transop);
break;
#ifdef N2N_HAVE_AES
case N2N_TRANSFORM_ID_AESCBC:
rc = n2n_transop_aes_cbc_init(&eee->conf, &eee->transop);
break;
#endif
default:
rc = n2n_transop_null_init(&eee->conf, &eee->transop);
}
if((rc < 0) || (eee->transop.fwd == NULL) || (eee->transop.transform_id != transop_id)) {
traceEvent(TRACE_ERROR, "Transop init failed");
goto edge_init_error;
}
if(eee->transop.no_encryption)
traceEvent(TRACE_WARNING, "Encryption is disabled in edge");
if(edge_init_sockets(eee, conf->local_port, conf->mgmt_port, conf->tos) < 0) {
traceEvent(TRACE_ERROR, "Error: socket setup failed");
goto edge_init_error;
}
//edge_init_success:
*rv = 0;
return(eee);
edge_init_error:
if(eee)
free(eee);
*rv = rc;
return(NULL);
}
/* ************************************** */
static int find_and_remove_peer(struct peer_info **head, const n2n_mac_t mac) {
struct peer_info *peer;
HASH_FIND_PEER(*head, mac, peer);
if(peer) {
HASH_DEL(*head, peer);
free(peer);
return(1);
}
return(0);
}
/* ************************************** */
static uint32_t localhost_v4 = 0x7f000001;
static uint8_t localhost_v6[IPV6_SIZE] = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1};
/* Exclude localhost as it may be received when an edge node runs
* in the same supernode host.
*/
static int is_valid_peer_sock(const n2n_sock_t *sock) {
if(((sock->family == AF_INET) && (*((uint32_t*)sock->addr.v4) != htonl(localhost_v4)))
|| ((sock->family == AF_INET6) && memcmp(sock->addr.v6, localhost_v6, IPV6_SIZE)))
return(1);
return(0);
}
/* ***************************************************** */
/** Resolve the supernode IP address.
*
* REVISIT: This is a really bad idea. The edge will block completely while the
* hostname resolution is performed. This could take 15 seconds.
*/
static void supernode2addr(n2n_sock_t * sn, const n2n_sn_name_t addrIn) {
n2n_sn_name_t addr;
const char *supernode_host;
memcpy(addr, addrIn, N2N_EDGE_SN_HOST_SIZE);
supernode_host = strtok(addr, ":");
if(supernode_host)
{
in_addr_t sn_addr;
char *supernode_port = strtok(NULL, ":");
const struct addrinfo aihints = {0, PF_INET, 0, 0, 0, NULL, NULL, NULL};
struct addrinfo * ainfo = NULL;
int nameerr;
if(supernode_port)
sn->port = atoi(supernode_port);
else
traceEvent(TRACE_WARNING, "Bad supernode parameter (-l <host:port>) %s %s:%s",
addr, supernode_host, supernode_port);
nameerr = getaddrinfo(supernode_host, NULL, &aihints, &ainfo);
if(0 == nameerr)
{
struct sockaddr_in * saddr;
/* ainfo s the head of a linked list if non-NULL. */
if(ainfo && (PF_INET == ainfo->ai_family))
{
/* It is definitely and IPv4 address -> sockaddr_in */
saddr = (struct sockaddr_in *)ainfo->ai_addr;
memcpy(sn->addr.v4, &(saddr->sin_addr.s_addr), IPV4_SIZE);
sn->family=AF_INET;
}
else
{
/* Should only return IPv4 addresses due to aihints. */
traceEvent(TRACE_WARNING, "Failed to resolve supernode IPv4 address for %s", supernode_host);
}
freeaddrinfo(ainfo); /* free everything allocated by getaddrinfo(). */
ainfo = NULL;
} else {
traceEvent(TRACE_WARNING, "Failed to resolve supernode host %s, assuming numeric", supernode_host);
sn_addr = inet_addr(supernode_host); /* uint32_t */
memcpy(sn->addr.v4, &(sn_addr), IPV4_SIZE);
sn->family=AF_INET;
}
} else
traceEvent(TRACE_WARNING, "Wrong supernode parameter (-l <host:port>)");
}
/* ************************************** */
/***
*
* Register over multicast in case there is a peer on the same network listening
*/
static void register_with_local_peers(n2n_edge_t * eee) {
#ifndef SKIP_MULTICAST_PEERS_DISCOVERY
if(eee->multicast_joined && eee->conf.allow_p2p) {
/* send registration to the local multicast group */
traceEvent(TRACE_DEBUG, "Registering with multicast group %s:%u",
N2N_MULTICAST_GROUP, N2N_MULTICAST_PORT);
send_register(eee, &(eee->multicast_peer), NULL);
}
#else
traceEvent(TRACE_DEBUG, "Multicast peers discovery is disabled, skipping");
#endif
}
/* ************************************** */
/** Start the registration process.
*
* If the peer is already in pending_peers, ignore the request.
* If not in pending_peers, add it and send a REGISTER.
*
* If hdr is for a direct peer-to-peer packet, try to register back to sender
* even if the MAC is in pending_peers. This is because an incident direct
* packet indicates that peer-to-peer exchange should work so more aggressive
* registration can be permitted (once per incoming packet) as this should only
* last for a small number of packets..
*
* Called from the main loop when Rx a packet for our device mac.
*/
static void register_with_new_peer(n2n_edge_t * eee,
uint8_t from_supernode,
const n2n_mac_t mac,
const n2n_sock_t * peer) {
/* REVISIT: purge of pending_peers not yet done. */
struct peer_info * scan;
macstr_t mac_buf;
n2n_sock_str_t sockbuf;
HASH_FIND_PEER(eee->pending_peers, mac, scan);
/* NOTE: pending_peers are purged periodically with purge_expired_registrations */
if(scan == NULL) {
scan = calloc(1, sizeof(struct peer_info));
memcpy(scan->mac_addr, mac, N2N_MAC_SIZE);
scan->sock = *peer;
scan->timeout = REGISTER_SUPER_INTERVAL_DFL; /* TODO: should correspond to the peer supernode registration timeout */
scan->last_seen = time(NULL); /* Don't change this it marks the pending peer for removal. */
HASH_ADD_PEER(eee->pending_peers, scan);
traceEvent(TRACE_DEBUG, "=== new pending %s -> %s",
macaddr_str(mac_buf, scan->mac_addr),
sock_to_cstr(sockbuf, &(scan->sock)));
traceEvent(TRACE_DEBUG, "Pending peers list size=%u",
HASH_COUNT(eee->pending_peers));
/* trace Sending REGISTER */
if(from_supernode) {
/* UDP NAT hole punching through supernode. Send to peer first(punch local UDP hole)
* and then ask supernode to forward. Supernode then ask peer to ack. Some nat device
* drop and block ports with incoming UDP packet if out-come traffic does not exist.
* So we can alternatively set TTL so that the packet sent to peer never really reaches
* The register_ttl is basically nat level + 1. Set it to 1 means host like DMZ.
*/
if (eee->conf.register_ttl == 1) {
/* We are DMZ host or port is directly accessible. Just let peer to send back the ack */
#ifndef WIN32
} else if(eee->conf.register_ttl > 1) {
/* Setting register_ttl usually implies that the edge knows the internal net topology
* clearly, we can apply aggressive port prediction to support incoming Symmetric NAT
*/
int curTTL = 0;
socklen_t lenTTL = sizeof(int);
n2n_sock_t sock = scan->sock;
int alter = 16; /* TODO: set by command line or more reliable prediction method */
getsockopt(eee->udp_sock, IPPROTO_IP, IP_TTL, (void *)(char *)&curTTL, &lenTTL);
setsockopt(eee->udp_sock, IPPROTO_IP, IP_TTL,
(void *)(char *)&eee->conf.register_ttl,
sizeof(eee->conf.register_ttl));
for (; alter > 0; alter--, sock.port++)
{
send_register(eee, &sock, mac);
}
setsockopt(eee->udp_sock, IPPROTO_IP, IP_TTL, (void *)(char *)&curTTL, sizeof(curTTL));
#endif
} else { /* eee->conf.register_ttl <= 0 */
/* Normal STUN */
send_register(eee, &(scan->sock), mac);
}
send_register(eee, &(eee->supernode), mac);
} else {
/* P2P register, send directly */
send_register(eee, &(scan->sock), mac);
}
register_with_local_peers(eee);
} else
scan->sock = *peer;
}
/* ************************************** */
/** Update the last_seen time for this peer, or get registered. */
static void check_peer_registration_needed(n2n_edge_t * eee,
uint8_t from_supernode,
const n2n_mac_t mac,
const n2n_sock_t * peer) {
struct peer_info *scan;
HASH_FIND_PEER(eee->known_peers, mac, scan);
if(scan == NULL) {
/* Not in known_peers - start the REGISTER process. */
register_with_new_peer(eee, from_supernode, mac, peer);
} else {
/* Already in known_peers. */
time_t now = time(NULL);
if(!from_supernode)
scan->last_p2p = now;
if((now - scan->last_seen) > 0 /* >= 1 sec */) {
/* Don't register too often */
check_known_peer_sock_change(eee, from_supernode, mac, peer, now);
}
}
}
/* ************************************** */
/* Confirm that a pending peer is reachable directly via P2P.
*
* peer must be a pointer to an element of the pending_peers list.
*/
static void peer_set_p2p_confirmed(n2n_edge_t * eee,
const n2n_mac_t mac,
const n2n_sock_t * peer,
time_t now) {
struct peer_info *scan;
macstr_t mac_buf;
n2n_sock_str_t sockbuf;
HASH_FIND_PEER(eee->pending_peers, mac, scan);
if(scan) {
HASH_DEL(eee->pending_peers, scan);
/* Add scan to known_peers. */
HASH_ADD_PEER(eee->known_peers, scan);
scan->sock = *peer;
scan->last_p2p = now;
traceEvent(TRACE_NORMAL, "P2P connection established: %s [%s]",
macaddr_str(mac_buf, mac),
sock_to_cstr(sockbuf, peer));
traceEvent(TRACE_DEBUG, "=== new peer %s -> %s",
macaddr_str(mac_buf, scan->mac_addr),
sock_to_cstr(sockbuf, &(scan->sock)));
traceEvent(TRACE_DEBUG, "Pending peers list size=%u",
HASH_COUNT(eee->pending_peers));
traceEvent(TRACE_DEBUG, "Known peers list size=%u",
HASH_COUNT(eee->known_peers));
scan->last_seen = now;
} else
traceEvent(TRACE_DEBUG, "Failed to find sender in pending_peers.");
}
/* ************************************** */
int is_empty_ip_address(const n2n_sock_t * sock) {
const uint8_t * ptr=NULL;
size_t len=0;
size_t i;
if(AF_INET6 == sock->family)
{
ptr = sock->addr.v6;
len = 16;
}
else
{
ptr = sock->addr.v4;
len = 4;
}
for (i=0; i<len; ++i)
{
if(0 != ptr[i])
{
/* found a non-zero byte in address */
return 0;
}
}
return 1;
}
/* ************************************** */
static n2n_mac_t broadcast_mac = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
/** Check if a known peer socket has changed and possibly register again.
*/
static void check_known_peer_sock_change(n2n_edge_t * eee,
uint8_t from_supernode,
const n2n_mac_t mac,
const n2n_sock_t * peer,
time_t when) {
struct peer_info *scan;
n2n_sock_str_t sockbuf1;
n2n_sock_str_t sockbuf2; /* don't clobber sockbuf1 if writing two addresses to trace */
macstr_t mac_buf;
if(is_empty_ip_address(peer))
return;
if(!memcmp(mac, broadcast_mac, N2N_MAC_SIZE))
return;
/* Search the peer in known_peers */
HASH_FIND_PEER(eee->known_peers, mac, scan);
if(!scan)
/* Not in known_peers */
return;
if(!sock_equal(&(scan->sock), peer)) {
if(!from_supernode) {
/* This is a P2P packet */
traceEvent(TRACE_NORMAL, "Peer changed %s: %s -> %s",
macaddr_str(mac_buf, scan->mac_addr),
sock_to_cstr(sockbuf1, &(scan->sock)),
sock_to_cstr(sockbuf2, peer));
/* The peer has changed public socket. It can no longer be assumed to be reachable. */
HASH_DEL(eee->known_peers, scan);
free(scan);
register_with_new_peer(eee, from_supernode, mac, peer);
} else {
/* Don't worry about what the supernode reports, it could be seeing a different socket. */
}
} else
scan->last_seen = when;
}
/* ************************************** */
/** Send a datagram to a socket defined by a n2n_sock_t */
static ssize_t sendto_sock(int fd, const void * buf,
size_t len, const n2n_sock_t * dest) {
struct sockaddr_in peer_addr;
ssize_t sent;
fill_sockaddr((struct sockaddr *) &peer_addr,
sizeof(peer_addr),
dest);
sent = sendto(fd, buf, len, 0/*flags*/,
(struct sockaddr *)&peer_addr, sizeof(struct sockaddr_in));
if(sent < 0)
{
char * c = strerror(errno);
traceEvent(TRACE_ERROR, "sendto failed (%d) %s", errno, c);
}
else
{
traceEvent(TRACE_DEBUG, "sendto sent=%d to ", (signed int)sent);
}
return sent;
}
/* ************************************** */
/* Bind eee->udp_multicast_sock to multicast group */
static void check_join_multicast_group(n2n_edge_t *eee) {
#ifndef SKIP_MULTICAST_PEERS_DISCOVERY
if(!eee->multicast_joined) {
struct ip_mreq mreq;
mreq.imr_multiaddr.s_addr = inet_addr(N2N_MULTICAST_GROUP);
mreq.imr_interface.s_addr = htonl(INADDR_ANY);
if(setsockopt(eee->udp_multicast_sock, IPPROTO_IP, IP_ADD_MEMBERSHIP, (char *)&mreq, sizeof(mreq)) < 0) {
traceEvent(TRACE_WARNING, "Failed to bind to local multicast group %s:%u [errno %u]",
N2N_MULTICAST_GROUP, N2N_MULTICAST_PORT, errno);
#ifdef WIN32
traceEvent(TRACE_ERROR, "WSAGetLastError(): %u", WSAGetLastError());
#endif
} else {
traceEvent(TRACE_NORMAL, "Successfully joined multicast group %s:%u",
N2N_MULTICAST_GROUP, N2N_MULTICAST_PORT);
eee->multicast_joined = 1;
}
}
#endif
}
/* ************************************** */
/** Send a REGISTER_SUPER packet to the current supernode. */
static void send_register_super(n2n_edge_t * eee,
const n2n_sock_t * supernode) {
uint8_t pktbuf[N2N_PKT_BUF_SIZE] = {0};
size_t idx;
/* ssize_t sent; */
n2n_common_t cmn;
n2n_REGISTER_SUPER_t reg;
n2n_sock_str_t sockbuf;
memset(&cmn, 0, sizeof(cmn));
memset(&reg, 0, sizeof(reg));
cmn.ttl=N2N_DEFAULT_TTL;
cmn.pc = n2n_register_super;
cmn.flags = 0;
memcpy(cmn.community, eee->conf.community_name, N2N_COMMUNITY_SIZE);
for(idx=0; idx < N2N_COOKIE_SIZE; ++idx)
eee->last_cookie[idx] = rand() % 0xff;
memcpy(reg.cookie, eee->last_cookie, N2N_COOKIE_SIZE);
reg.auth.scheme=0; /* No auth yet */
idx=0;
encode_mac(reg.edgeMac, &idx, eee->device.mac_addr);
idx=0;
encode_REGISTER_SUPER(pktbuf, &idx, &cmn, &reg);
traceEvent(TRACE_DEBUG, "send REGISTER_SUPER to %s",
sock_to_cstr(sockbuf, supernode));
/* sent = */ sendto_sock(eee->udp_sock, pktbuf, idx, supernode);
}
/* ************************************** */
/** Send a QUERY_PEER packet to the current supernode. */
static void send_query_peer( n2n_edge_t * eee,
const n2n_mac_t dstMac) {
uint8_t pktbuf[N2N_PKT_BUF_SIZE];
size_t idx;
n2n_common_t cmn = {0};
n2n_QUERY_PEER_t query = {{0}};
cmn.ttl=N2N_DEFAULT_TTL;
cmn.pc = n2n_query_peer;
cmn.flags = 0;
memcpy( cmn.community, eee->conf.community_name, N2N_COMMUNITY_SIZE );
idx=0;
encode_mac( query.srcMac, &idx, eee->device.mac_addr );
idx=0;
encode_mac( query.targetMac, &idx, dstMac );
idx=0;
encode_QUERY_PEER( pktbuf, &idx, &cmn, &query );
traceEvent( TRACE_DEBUG, "send QUERY_PEER to supernode" );
sendto_sock( eee->udp_sock, pktbuf, idx, &(eee->supernode) );
}
/** Send a REGISTER packet to another edge. */
static void send_register(n2n_edge_t * eee,
const n2n_sock_t * remote_peer,
const n2n_mac_t peer_mac) {
uint8_t pktbuf[N2N_PKT_BUF_SIZE];
size_t idx;
/* ssize_t sent; */
n2n_common_t cmn;
n2n_REGISTER_t reg;
n2n_sock_str_t sockbuf;
if(!eee->conf.allow_p2p) {
traceEvent(TRACE_DEBUG, "Skipping register as P2P is disabled");
return;
}
memset(&cmn, 0, sizeof(cmn));
memset(&reg, 0, sizeof(reg));
cmn.ttl=N2N_DEFAULT_TTL;
cmn.pc = n2n_register;
cmn.flags = 0;
memcpy(cmn.community, eee->conf.community_name, N2N_COMMUNITY_SIZE);
idx=0;
encode_uint32(reg.cookie, &idx, 123456789);
idx=0;
encode_mac(reg.srcMac, &idx, eee->device.mac_addr);
if(peer_mac) {
/* Can be NULL for multicast registrations */
idx=0;
encode_mac(reg.dstMac, &idx, peer_mac);
}
idx=0;
encode_REGISTER(pktbuf, &idx, &cmn, &reg);
traceEvent(TRACE_INFO, "Send REGISTER to %s",
sock_to_cstr(sockbuf, remote_peer));
/* sent = */ sendto_sock(eee->udp_sock, pktbuf, idx, remote_peer);
}
/* ************************************** */
/** Send a REGISTER_ACK packet to a peer edge. */
static void send_register_ack(n2n_edge_t * eee,
const n2n_sock_t * remote_peer,
const n2n_REGISTER_t * reg) {
uint8_t pktbuf[N2N_PKT_BUF_SIZE];
size_t idx;
/* ssize_t sent; */
n2n_common_t cmn;
n2n_REGISTER_ACK_t ack;
n2n_sock_str_t sockbuf;
if(!eee->conf.allow_p2p) {
traceEvent(TRACE_DEBUG, "Skipping register ACK as P2P is disabled");
return;
}
memset(&cmn, 0, sizeof(cmn));
memset(&ack, 0, sizeof(reg));
cmn.ttl=N2N_DEFAULT_TTL;
cmn.pc = n2n_register_ack;
cmn.flags = 0;
memcpy(cmn.community, eee->conf.community_name, N2N_COMMUNITY_SIZE);
memset(&ack, 0, sizeof(ack));
memcpy(ack.cookie, reg->cookie, N2N_COOKIE_SIZE);
memcpy(ack.srcMac, eee->device.mac_addr, N2N_MAC_SIZE);
memcpy(ack.dstMac, reg->srcMac, N2N_MAC_SIZE);
idx=0;
encode_REGISTER_ACK(pktbuf, &idx, &cmn, &ack);
traceEvent(TRACE_INFO, "send REGISTER_ACK %s",
sock_to_cstr(sockbuf, remote_peer));
/* sent = */ sendto_sock(eee->udp_sock, pktbuf, idx, remote_peer);
}
/* ************************************** */
/** @brief Check to see if we should re-register with the supernode.
*
* This is frequently called by the main loop.
*/
static void update_supernode_reg(n2n_edge_t * eee, time_t nowTime) {
u_int sn_idx;
if(eee->sn_wait && (nowTime > (eee->last_register_req + (eee->conf.register_interval/10)))) {
/* fall through */
traceEvent(TRACE_DEBUG, "update_supernode_reg: doing fast retry.");
} else if(nowTime < (eee->last_register_req + eee->conf.register_interval))
return; /* Too early */
check_join_multicast_group(eee);
if(0 == eee->sup_attempts) {
/* Give up on that supernode and try the next one. */
++(eee->sn_idx);
if (eee->sn_idx >= eee->conf.sn_num) {
/* Got to end of list, go back to the start. Also works for list of one entry. */
eee->sn_idx=0;
}
traceEvent(TRACE_WARNING, "Supernode not responding, now trying %s", supernode_ip(eee));
eee->sup_attempts = N2N_EDGE_SUP_ATTEMPTS;
}
else
--(eee->sup_attempts);
for(sn_idx=0; sn_idx<eee->conf.sn_num; sn_idx++) {
supernode2addr(&(eee->supernode), eee->conf.sn_ip_array[sn_idx]);
traceEvent(TRACE_INFO, "Registering with supernode [id: %u/%u][%s][attempts left %u]",
sn_idx+1, eee->conf.sn_num,
supernode_ip(eee), (unsigned int)eee->sup_attempts);
send_register_super(eee, &(eee->supernode));
}
register_with_local_peers(eee);
eee->sn_wait=1;
/* REVISIT: turn-on gratuitous ARP with config option. */
/* send_grat_arps(sock_fd, is_udp_sock); */
eee->last_register_req = nowTime;
}
/* ************************************** */
/** NOT IMPLEMENTED
*
* This would send a DEREGISTER packet to a peer edge or supernode to indicate
* the edge is going away.
*/
static void send_deregister(n2n_edge_t * eee,
n2n_sock_t * remote_peer) {
/* Marshall and send message */
}
/* ************************************** */
/** Return the IP address of the current supernode in the ring. */
static const char * supernode_ip(const n2n_edge_t * eee) {
return (eee->conf.sn_ip_array)[eee->sn_idx];
}
/* ************************************** */
/** A PACKET has arrived containing an encapsulated ethernet datagram - usually
* encrypted. */
static int handle_PACKET(n2n_edge_t * eee,
const n2n_common_t * cmn,
const n2n_PACKET_t * pkt,
const n2n_sock_t * orig_sender,
uint8_t * payload,
size_t psize) {
ssize_t data_sent_len;
uint8_t from_supernode;
uint8_t * eth_payload=NULL;
int retval = -1;
time_t now;
ether_hdr_t * eh;
ipstr_t ip_buf;
now = time(NULL);
traceEvent(TRACE_DEBUG, "handle_PACKET size %u transform %u",
(unsigned int)psize, (unsigned int)pkt->transform);
/* hexdump(payload, psize); */
from_supernode= cmn->flags & N2N_FLAGS_FROM_SUPERNODE;
if(from_supernode)
{
if(!memcmp(pkt->dstMac, broadcast_mac, 6))
++(eee->stats.rx_sup_broadcast);
++(eee->stats.rx_sup);
eee->last_sup=now;
}
else
{
++(eee->stats.rx_p2p);
eee->last_p2p=now;
}
/* Update the sender in peer table entry */
check_peer_registration_needed(eee, from_supernode, pkt->srcMac, orig_sender);
/* Handle transform. */
{
uint8_t decodebuf[N2N_PKT_BUF_SIZE];
size_t eth_size;
n2n_transform_t rx_transop_id;
rx_transop_id = (n2n_transform_t)pkt->transform;
if(rx_transop_id == eee->conf.transop_id) {
eth_payload = decodebuf;
eh = (ether_hdr_t*)eth_payload;
eth_size = eee->transop.rev(&eee->transop,
eth_payload, N2N_PKT_BUF_SIZE,
payload, psize, pkt->srcMac);
++(eee->transop.rx_cnt); /* stats */
if(!(eee->conf.allow_routing)) {
if((ntohs(eh->type) == 0x0800) && (eth_size >= ETH_FRAMESIZE + IP4_MIN_SIZE)) {
uint32_t *dst = (uint32_t*)&eth_payload[ETH_FRAMESIZE + IP4_DSTOFFSET];
u_int8_t *dst_mac = (u_int8_t*)eth_payload;
/* Note: all elements of the_ip are in network order */
if(!memcmp(dst_mac, broadcast_mac, 6))
traceEvent(TRACE_DEBUG, "Broadcast packet [%s]",
intoa(ntohl(*dst), ip_buf, sizeof(ip_buf)));
else if((*dst != eee->device.ip_addr)) {
/* This is a packet that needs to be routed */
traceEvent(TRACE_INFO, "Discarding routed packet [%s]",
intoa(ntohl(*dst), ip_buf, sizeof(ip_buf)));
return(-1);
} else {
/* This packet is directed to us */
/* traceEvent(TRACE_INFO, "Sending non-routed packet"); */
}
}
}
/* Write ethernet packet to tap device. */
traceEvent(TRACE_DEBUG, "sending to TAP %u", (unsigned int)eth_size);
data_sent_len = tuntap_write(&(eee->device), eth_payload, eth_size);
if (data_sent_len == eth_size)
{
retval = 0;
}
}
else
{
traceEvent(TRACE_ERROR, "invalid transop ID: expected %s(%u), got %s(%u)",
transop_str(eee->conf.transop_id), eee->conf.transop_id,
transop_str(rx_transop_id), rx_transop_id);
}
}
return retval;
}
/* ************************************** */
/** Read a datagram from the management UDP socket and take appropriate
* action. */
static void readFromMgmtSocket(n2n_edge_t * eee, int * keep_running) {
uint8_t udp_buf[N2N_PKT_BUF_SIZE]; /* Compete UDP packet */
ssize_t recvlen;
/* ssize_t sendlen; */
struct sockaddr_in sender_sock;
socklen_t i;
size_t msg_len;
time_t now;
now = time(NULL);
i = sizeof(sender_sock);
recvlen = recvfrom(eee->udp_mgmt_sock, udp_buf, N2N_PKT_BUF_SIZE, 0/*flags*/,
(struct sockaddr *)&sender_sock, (socklen_t*)&i);
if(recvlen < 0)
{
traceEvent(TRACE_ERROR, "mgmt recvfrom failed with %s", strerror(errno));
return; /* failed to receive data from UDP */
}
if(recvlen >= 4)
{
if(0 == memcmp(udp_buf, "stop", 4))
{
traceEvent(TRACE_ERROR, "stop command received.");
*keep_running = 0;
return;
}
if(0 == memcmp(udp_buf, "help", 4))
{
msg_len=0;
setTraceLevel(getTraceLevel()+1);
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"Help for edge management console:\n"
" stop Gracefully exit edge\n"
" help This help message\n"
" +verb Increase verbosity of logging\n"
" -verb Decrease verbosity of logging\n"
" <enter> Display statistics\n\n");
sendto(eee->udp_mgmt_sock, udp_buf, msg_len, 0/*flags*/,
(struct sockaddr *)&sender_sock, sizeof(struct sockaddr_in));
return;
}
}
if(recvlen >= 5)
{
if(0 == memcmp(udp_buf, "+verb", 5))
{
msg_len=0;
setTraceLevel(getTraceLevel()+1);
traceEvent(TRACE_ERROR, "+verb traceLevel=%u", (unsigned int)getTraceLevel());
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"> +OK traceLevel=%u\n", (unsigned int)getTraceLevel());
sendto(eee->udp_mgmt_sock, udp_buf, msg_len, 0/*flags*/,
(struct sockaddr *)&sender_sock, sizeof(struct sockaddr_in));
return;
}
if(0 == memcmp(udp_buf, "-verb", 5))
{
msg_len=0;
if(getTraceLevel() > 0)
{
setTraceLevel(getTraceLevel()-1);
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"> -OK traceLevel=%u\n", getTraceLevel());
}
else
{
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"> -NOK traceLevel=%u\n", getTraceLevel());
}
traceEvent(TRACE_ERROR, "-verb traceLevel=%u", (unsigned int)getTraceLevel());
sendto(eee->udp_mgmt_sock, udp_buf, msg_len, 0/*flags*/,
(struct sockaddr *)&sender_sock, sizeof(struct sockaddr_in));
return;
}
}
traceEvent(TRACE_DEBUG, "mgmt status rq");
msg_len=0;
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"Statistics for edge\n");
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"uptime %lu\n",
time(NULL) - eee->start_time);
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"paths super:%u,%u p2p:%u,%u\n",
(unsigned int)eee->stats.tx_sup,
(unsigned int)eee->stats.rx_sup,
(unsigned int)eee->stats.tx_p2p,
(unsigned int)eee->stats.rx_p2p);
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"transop |%6u|%6u|\n",
(unsigned int)eee->transop.tx_cnt,
(unsigned int)eee->transop.rx_cnt);
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"peers pend:%u full:%u\n",
HASH_COUNT(eee->pending_peers),
HASH_COUNT(eee->known_peers));
msg_len += snprintf((char *)(udp_buf+msg_len), (N2N_PKT_BUF_SIZE-msg_len),
"last super:%lu(%ld sec ago) p2p:%lu(%ld sec ago)\n",
eee->last_sup, (now-eee->last_sup), eee->last_p2p,
(now-eee->last_p2p));
traceEvent(TRACE_DEBUG, "mgmt status sending: %s", udp_buf);
/* sendlen = */ sendto(eee->udp_mgmt_sock, udp_buf, msg_len, 0/*flags*/,
(struct sockaddr *)&sender_sock, sizeof(struct sockaddr_in));
}
/* ************************************** */
/** Destination MAC 33:33:0:00:00:00 - 33:33:FF:FF:FF:FF is reserved for IPv6
* neighbour discovery.
*/
static int is_ip6_discovery(const void * buf, size_t bufsize) {
int retval = 0;
if(bufsize >= sizeof(ether_hdr_t)) {
/* copy to aligned memory */
ether_hdr_t eh;
memcpy(&eh, buf, sizeof(ether_hdr_t));
if((0x33 == eh.dhost[0]) && (0x33 == eh.dhost[1]))
retval = 1; /* This is an IPv6 multicast packet [RFC2464]. */
}
return retval;
}
/* ************************************** */
/** Destination 01:00:5E:00:00:00 - 01:00:5E:7F:FF:FF is multicast ethernet.
*/
static int is_ethMulticast(const void * buf, size_t bufsize) {
int retval = 0;
/* Match 01:00:5E:00:00:00 - 01:00:5E:7F:FF:FF */
if(bufsize >= sizeof(ether_hdr_t)) {
/* copy to aligned memory */
ether_hdr_t eh;
memcpy(&eh, buf, sizeof(ether_hdr_t));
if((0x01 == eh.dhost[0]) &&
(0x00 == eh.dhost[1]) &&
(0x5E == eh.dhost[2]) &&
(0 == (0x80 & eh.dhost[3])))
retval = 1; /* This is an ethernet multicast packet [RFC1112]. */
}
return retval;
}
/* ************************************** */
static int check_query_peer_info(n2n_edge_t *eee, time_t now, n2n_mac_t mac) {
struct peer_info *scan;
HASH_FIND_PEER(eee->pending_peers, mac, scan);
if(!scan) {
scan = calloc(1, sizeof(struct peer_info));
memcpy(scan->mac_addr, mac, N2N_MAC_SIZE);
scan->timeout = REGISTER_SUPER_INTERVAL_DFL; /* TODO: should correspond to the peer supernode registration timeout */
scan->last_seen = now; /* Don't change this it marks the pending peer for removal. */
HASH_ADD_PEER(eee->pending_peers, scan);
}
if(now - scan->last_sent_query > REGISTER_SUPER_INTERVAL_DFL) {
send_query_peer(eee, scan->mac_addr);
scan->last_sent_query = now;
return(0);
}
return(1);
}
/* ************************************** */
/* @return 1 if destination is a peer, 0 if destination is supernode */
static int find_peer_destination(n2n_edge_t * eee,
n2n_mac_t mac_address,
n2n_sock_t * destination) {
struct peer_info *scan;
macstr_t mac_buf;
n2n_sock_str_t sockbuf;
int retval=0;
time_t now = time(NULL);
if(!memcmp(mac_address, broadcast_mac, 6)) {
traceEvent(TRACE_DEBUG, "Broadcast destination peer, using supernode");
memcpy(destination, &(eee->supernode), sizeof(struct sockaddr_in));
return(0);
}
traceEvent(TRACE_DEBUG, "Searching destination peer for MAC %02X:%02X:%02X:%02X:%02X:%02X",
mac_address[0] & 0xFF, mac_address[1] & 0xFF, mac_address[2] & 0xFF,
mac_address[3] & 0xFF, mac_address[4] & 0xFF, mac_address[5] & 0xFF);
HASH_FIND_PEER(eee->known_peers, mac_address, scan);
if(scan && (scan->last_seen > 0)) {
if((now - scan->last_p2p) >= (scan->timeout / 2)) {
/* Too much time passed since we saw the peer, need to register again
* since the peer address may have changed. */
traceEvent(TRACE_DEBUG, "Refreshing idle known peer");
HASH_DEL(eee->known_peers, scan);
free(scan);
/* NOTE: registration will be performed upon the receival of the next response packet */
} else {
/* Valid known peer found */
memcpy(destination, &scan->sock, sizeof(n2n_sock_t));
retval=1;
}
}
if(retval == 0) {
memcpy(destination, &(eee->supernode), sizeof(struct sockaddr_in));
traceEvent(TRACE_DEBUG, "P2P Peer [MAC=%02X:%02X:%02X:%02X:%02X:%02X] not found, using supernode",
mac_address[0] & 0xFF, mac_address[1] & 0xFF, mac_address[2] & 0xFF,
mac_address[3] & 0xFF, mac_address[4] & 0xFF, mac_address[5] & 0xFF);
check_query_peer_info(eee, now, mac_address);
}
traceEvent(TRACE_DEBUG, "find_peer_address (%s) -> [%s]",
macaddr_str(mac_buf, mac_address),
sock_to_cstr(sockbuf, destination));
return retval;
}
/* ***************************************************** */
/** Send an ecapsulated ethernet PACKET to a destination edge or broadcast MAC
* address. */
static int send_packet(n2n_edge_t * eee,
n2n_mac_t dstMac,
const uint8_t * pktbuf,
size_t pktlen) {
int is_p2p;
/*ssize_t s; */
n2n_sock_str_t sockbuf;
n2n_sock_t destination;
macstr_t mac_buf;
/* hexdump(pktbuf, pktlen); */
is_p2p = find_peer_destination(eee, dstMac, &destination);
if(is_p2p)
++(eee->stats.tx_p2p);
else {
++(eee->stats.tx_sup);
if(!memcmp(dstMac, broadcast_mac, 6))
++(eee->stats.tx_sup_broadcast);
}
traceEvent(TRACE_INFO, "Tx PACKET to %s (dest=%s) [%u B]",
sock_to_cstr(sockbuf, &destination),
macaddr_str(mac_buf, dstMac), pktlen);
/* s = */ sendto_sock(eee->udp_sock, pktbuf, pktlen, &destination);
return 0;
}
/* ************************************** */
/** A layer-2 packet was received at the tunnel and needs to be sent via UDP. */
static void send_packet2net(n2n_edge_t * eee,
uint8_t *tap_pkt, size_t len) {
ipstr_t ip_buf;
n2n_mac_t destMac;
n2n_common_t cmn;
n2n_PACKET_t pkt;
uint8_t pktbuf[N2N_PKT_BUF_SIZE];
size_t idx=0;
n2n_transform_t tx_transop_idx = eee->transop.transform_id;
ether_hdr_t eh;
/* tap_pkt is not aligned so we have to copy to aligned memory */
memcpy(&eh, tap_pkt, sizeof(ether_hdr_t));
/* Discard IP packets that are not originated by this hosts */
if(!(eee->conf.allow_routing)) {
if(ntohs(eh.type) == 0x0800) {
/* This is an IP packet from the local source address - not forwarded. */
uint32_t *src = (uint32_t*)&tap_pkt[ETH_FRAMESIZE + IP4_SRCOFFSET];
/* Note: all elements of the_ip are in network order */
if(*src != eee->device.ip_addr) {
/* This is a packet that needs to be routed */
traceEvent(TRACE_INFO, "Discarding routed packet [%s]",
intoa(ntohl(*src), ip_buf, sizeof(ip_buf)));
return;
} else {
/* This packet is originated by us */
/* traceEvent(TRACE_INFO, "Sending non-routed packet"); */
}
}
}
/* Optionally compress then apply transforms, eg encryption. */
/* Once processed, send to destination in PACKET */
memcpy(destMac, tap_pkt, N2N_MAC_SIZE); /* dest MAC is first in ethernet header */
memset(&cmn, 0, sizeof(cmn));
cmn.ttl = N2N_DEFAULT_TTL;
cmn.pc = n2n_packet;
cmn.flags=0; /* no options, not from supernode, no socket */
memcpy(cmn.community, eee->conf.community_name, N2N_COMMUNITY_SIZE);
memset(&pkt, 0, sizeof(pkt));
memcpy(pkt.srcMac, eee->device.mac_addr, N2N_MAC_SIZE);
memcpy(pkt.dstMac, destMac, N2N_MAC_SIZE);
pkt.sock.family=0; /* do not encode sock */
pkt.transform = tx_transop_idx;
idx=0;
encode_PACKET(pktbuf, &idx, &cmn, &pkt);
idx += eee->transop.fwd(&eee->transop,
pktbuf+idx, N2N_PKT_BUF_SIZE-idx,
tap_pkt, len, pkt.dstMac);
traceEvent(TRACE_DEBUG, "Encode %u B PACKET [%u B data, %u B overhead] transform %u",
(u_int)idx, (u_int)len, (u_int)(idx-len), tx_transop_idx);
#ifdef MTU_ASSERT_VALUE
{
const u_int eth_udp_overhead = ETH_FRAMESIZE + IP4_MIN_SIZE + UDP_SIZE;
// MTU assertion which avoids fragmentation by N2N
assert(idx + eth_udp_overhead <= MTU_ASSERT_VALUE);
}
#endif
eee->transop.tx_cnt++; /* stats */
send_packet(eee, destMac, pktbuf, idx); /* to peer or supernode */
}
/* ************************************** */
/** Read a single packet from the TAP interface, process it and write out the
* corresponding packet to the cooked socket.
*/
static void readFromTAPSocket(n2n_edge_t * eee) {
/* tun -> remote */
uint8_t eth_pkt[N2N_PKT_BUF_SIZE];
macstr_t mac_buf;
ssize_t len;
#ifdef __ANDROID_NDK__
if (uip_arp_len != 0) {
len = uip_arp_len;
memcpy(eth_pkt, uip_arp_buf, MIN(uip_arp_len, N2N_PKT_BUF_SIZE));
traceEvent(TRACE_DEBUG, "ARP reply packet to send");
}
else
{
#endif /* #ifdef __ANDROID_NDK__ */
len = tuntap_read( &(eee->device), eth_pkt, N2N_PKT_BUF_SIZE );
#ifdef __ANDROID_NDK__
}
#endif /* #ifdef __ANDROID_NDK__ */
if((len <= 0) || (len > N2N_PKT_BUF_SIZE))
{
traceEvent(TRACE_WARNING, "read()=%d [%d/%s]",
(signed int)len, errno, strerror(errno));
}
else
{
const uint8_t * mac = eth_pkt;
traceEvent(TRACE_DEBUG, "### Rx TAP packet (%4d) for %s",
(signed int)len, macaddr_str(mac_buf, mac));
if(eee->conf.drop_multicast &&
(is_ip6_discovery(eth_pkt, len) ||
is_ethMulticast(eth_pkt, len)
)
)
{
traceEvent(TRACE_DEBUG, "Dropping multicast");
}
else
{
send_packet2net(eee, eth_pkt, len);
}
}
}
/* ************************************** */
#ifdef WIN32
struct tunread_arg {
n2n_edge_t *eee;
int *keep_running;
};
static DWORD* tunReadThread(LPVOID lpArg) {
struct tunread_arg *arg = (struct tunread_arg*)lpArg;
while(*arg->keep_running)
readFromTAPSocket(arg->eee);
return((DWORD*)NULL);
}
/* ************************************** */
/** Start a second thread in Windows because TUNTAP interfaces do not expose
* file descriptors. */
static HANDLE startTunReadThread(struct tunread_arg *arg) {
DWORD dwThreadId;
return(CreateThread(NULL, /* security attributes */
0, /* use default stack size */
(LPTHREAD_START_ROUTINE)tunReadThread, /* thread function */
(void*)arg, /* argument to thread function */
0, /* thread creation flags */
&dwThreadId)); /* thread id out */
}
#endif
/* ************************************** */
/** Read a datagram from the main UDP socket to the internet. */
static void readFromIPSocket(n2n_edge_t * eee, int in_sock) {
n2n_common_t cmn; /* common fields in the packet header */
n2n_sock_str_t sockbuf1;
n2n_sock_str_t sockbuf2; /* don't clobber sockbuf1 if writing two addresses to trace */
macstr_t mac_buf1;
macstr_t mac_buf2;
uint8_t udp_buf[N2N_PKT_BUF_SIZE]; /* Compete UDP packet */
ssize_t recvlen;
size_t rem;
size_t idx;
size_t msg_type;
uint8_t from_supernode;
struct sockaddr_in sender_sock;
n2n_sock_t sender;
n2n_sock_t * orig_sender=NULL;
time_t now=0;
size_t i;
i = sizeof(sender_sock);
recvlen = recvfrom(in_sock, udp_buf, N2N_PKT_BUF_SIZE, 0/*flags*/,
(struct sockaddr *)&sender_sock, (socklen_t*)&i);
if(recvlen < 0) {
#ifdef WIN32
if(WSAGetLastError() != WSAECONNRESET)
#endif
{
traceEvent(TRACE_ERROR, "recvfrom() failed %d errno %d (%s)", recvlen, errno, strerror(errno));
#ifdef WIN32
traceEvent(TRACE_ERROR, "WSAGetLastError(): %u", WSAGetLastError());
#endif
}
return; /* failed to receive data from UDP */
}
/* REVISIT: when UDP/IPv6 is supported we will need a flag to indicate which
* IP transport version the packet arrived on. May need to UDP sockets. */
sender.family = AF_INET; /* UDP socket was opened PF_INET v4 */
sender.port = ntohs(sender_sock.sin_port);
memcpy(&(sender.addr.v4), &(sender_sock.sin_addr.s_addr), IPV4_SIZE);
/* The packet may not have an orig_sender socket spec. So default to last
* hop as sender. */
orig_sender=&sender;
traceEvent(TRACE_DEBUG, "### Rx N2N UDP (%d) from %s",
(signed int)recvlen, sock_to_cstr(sockbuf1, &sender));
/* hexdump(udp_buf, recvlen); */
rem = recvlen; /* Counts down bytes of packet to protect against buffer overruns. */
idx = 0; /* marches through packet header as parts are decoded. */
if(decode_common(&cmn, udp_buf, &rem, &idx) < 0)
{
traceEvent(TRACE_ERROR, "Failed to decode common section in N2N_UDP");
return; /* failed to decode packet */
}
now = time(NULL);
msg_type = cmn.pc; /* packet code */
from_supernode= cmn.flags & N2N_FLAGS_FROM_SUPERNODE;
if(0 == memcmp(cmn.community, eee->conf.community_name, N2N_COMMUNITY_SIZE)) {
switch(msg_type) {
case MSG_TYPE_PACKET:
{
/* process PACKET - most frequent so first in list. */
n2n_PACKET_t pkt;
decode_PACKET(&pkt, &cmn, udp_buf, &rem, &idx);
if(is_valid_peer_sock(&pkt.sock))
orig_sender = &(pkt.sock);
if(!from_supernode) {
/* This is a P2P packet from the peer. We purge a pending
* registration towards the possibly nat-ted peer address as we now have
* a valid channel. We still use check_peer_registration_needed in
* handle_PACKET to double check this.
*/
traceEvent(TRACE_DEBUG, "Got P2P packet");
find_and_remove_peer(&eee->pending_peers, pkt.srcMac);
}
traceEvent(TRACE_INFO, "Rx PACKET from %s (sender=%s) [%u B]",
sock_to_cstr(sockbuf1, &sender),
sock_to_cstr(sockbuf2, orig_sender),
recvlen);
handle_PACKET(eee, &cmn, &pkt, orig_sender, udp_buf+idx, recvlen-idx);
break;
}
case MSG_TYPE_REGISTER:
{
/* Another edge is registering with us */
n2n_REGISTER_t reg;
n2n_mac_t null_mac = { '\0' };
int via_multicast;
decode_REGISTER(&reg, &cmn, udp_buf, &rem, &idx);
if(is_valid_peer_sock(&reg.sock))
orig_sender = &(reg.sock);
via_multicast = !memcmp(reg.dstMac, null_mac, 6);
if(via_multicast && !memcmp(reg.srcMac, eee->device.mac_addr, 6)) {
traceEvent(TRACE_DEBUG, "Skipping REGISTER from self");
break;
}
if(!via_multicast && memcmp(reg.dstMac, eee->device.mac_addr, 6)) {
traceEvent(TRACE_DEBUG, "Skipping REGISTER for other peer");
break;
}
if(!from_supernode) {
/* This is a P2P registration from the peer. We purge a pending
* registration towards the possibly nat-ted peer address as we now have
* a valid channel. We still use check_peer_registration_needed below
* to double check this.
*/
traceEvent(TRACE_DEBUG, "Got P2P register");
find_and_remove_peer(&eee->pending_peers, reg.srcMac);
/* NOTE: only ACK to peers */
send_register_ack(eee, orig_sender, &reg);
}
traceEvent(TRACE_INFO, "Rx REGISTER src=%s dst=%s from peer %s (%s)",
macaddr_str(mac_buf1, reg.srcMac),
macaddr_str(mac_buf2, reg.dstMac),
sock_to_cstr(sockbuf1, &sender),
sock_to_cstr(sockbuf2, orig_sender));
check_peer_registration_needed(eee, from_supernode, reg.srcMac, orig_sender);
break;
}
case MSG_TYPE_REGISTER_ACK:
{
/* Peer edge is acknowledging our register request */
n2n_REGISTER_ACK_t ra;
decode_REGISTER_ACK(&ra, &cmn, udp_buf, &rem, &idx);
if(is_valid_peer_sock(&ra.sock))
orig_sender = &(ra.sock);
traceEvent(TRACE_INFO, "Rx REGISTER_ACK src=%s dst=%s from peer %s (%s)",
macaddr_str(mac_buf1, ra.srcMac),
macaddr_str(mac_buf2, ra.dstMac),
sock_to_cstr(sockbuf1, &sender),
sock_to_cstr(sockbuf2, orig_sender));
peer_set_p2p_confirmed(eee, ra.srcMac, &sender, now);
break;
}
case MSG_TYPE_REGISTER_SUPER_ACK:
{
n2n_REGISTER_SUPER_ACK_t ra;
if(eee->sn_wait)
{
decode_REGISTER_SUPER_ACK(&ra, &cmn, udp_buf, &rem, &idx);
if(is_valid_peer_sock(&ra.sock))
orig_sender = &(ra.sock);
traceEvent(TRACE_INFO, "Rx REGISTER_SUPER_ACK myMAC=%s [%s] (external %s). Attempts %u",
macaddr_str(mac_buf1, ra.edgeMac),
sock_to_cstr(sockbuf1, &sender),
sock_to_cstr(sockbuf2, orig_sender),
(unsigned int)eee->sup_attempts);
if(0 == memcmp(ra.cookie, eee->last_cookie, N2N_COOKIE_SIZE))
{
if(ra.num_sn > 0)
{
traceEvent(TRACE_NORMAL, "Rx REGISTER_SUPER_ACK backup supernode at %s",
sock_to_cstr(sockbuf1, &(ra.sn_bak)));
}
eee->last_sup = now;
eee->sn_wait=0;
eee->sup_attempts = N2N_EDGE_SUP_ATTEMPTS; /* refresh because we got a response */
/* NOTE: the register_interval should be chosen by the edge node
* based on its NAT configuration. */
//eee->conf.register_interval = ra.lifetime;
}
else
{
traceEvent(TRACE_INFO, "Rx REGISTER_SUPER_ACK with wrong or old cookie.");
}
}
else
{
traceEvent(TRACE_INFO, "Rx REGISTER_SUPER_ACK with no outstanding REGISTER_SUPER.");
}
break;
} case MSG_TYPE_PEER_INFO: {
n2n_PEER_INFO_t pi;
struct peer_info * scan;
decode_PEER_INFO( &pi, &cmn, udp_buf, &rem, &idx );
if(!is_valid_peer_sock(&pi.sock)) {
traceEvent(TRACE_DEBUG, "Skip invalid PEER_INFO %s [%s]",
sock_to_cstr(sockbuf1, &pi.sock),
macaddr_str(mac_buf1, pi.mac) );
break;
}
HASH_FIND_PEER(eee->pending_peers, pi.mac, scan);
if (scan) {
scan->sock = pi.sock;
traceEvent(TRACE_INFO, "Rx PEER_INFO for %s: is at %s",
macaddr_str(mac_buf1, pi.mac),
sock_to_cstr(sockbuf1, &pi.sock));
send_register(eee, &scan->sock, scan->mac_addr);
} else {
traceEvent(TRACE_INFO, "Rx PEER_INFO unknown peer %s",
macaddr_str(mac_buf1, pi.mac) );
}
break;
}
default:
/* Not a known message type */
traceEvent(TRACE_WARNING, "Unable to handle packet type %d: ignored", (signed int)msg_type);
return;
} /* switch(msg_type) */
} else if(from_supernode) /* if (community match) */
traceEvent(TRACE_WARNING, "Received packet with unknown community");
else
traceEvent(TRACE_INFO, "Ignoring packet with unknown community");
}
/* ************************************** */
void print_edge_stats(const n2n_edge_t *eee) {
const struct n2n_edge_stats *s = &eee->stats;
traceEvent(TRACE_NORMAL, "**********************************");
traceEvent(TRACE_NORMAL, "Packet stats:");
traceEvent(TRACE_NORMAL, " TX P2P: %u pkts", s->tx_p2p);
traceEvent(TRACE_NORMAL, " RX P2P: %u pkts", s->rx_p2p);
traceEvent(TRACE_NORMAL, " TX Supernode: %u pkts (%u broadcast)", s->tx_sup, s->tx_sup_broadcast);
traceEvent(TRACE_NORMAL, " RX Supernode: %u pkts (%u broadcast)", s->rx_sup, s->rx_sup_broadcast);
traceEvent(TRACE_NORMAL, "**********************************");
}
/* ************************************** */
int run_edge_loop(n2n_edge_t * eee, int *keep_running) {
size_t numPurged;
time_t lastIfaceCheck=0;
time_t lastTransop=0;
time_t last_purge_known = 0;
time_t last_purge_pending = 0;
#ifdef __ANDROID_NDK__
time_t lastArpPeriod=0;
#endif
#ifdef WIN32
struct tunread_arg arg;
arg.eee = eee;
arg.keep_running = keep_running;
HANDLE tun_read_thread = startTunReadThread(&arg);
#endif
*keep_running = 1;
update_supernode_reg(eee, time(NULL));
/* Main loop
*
* select() is used to wait for input on either the TAP fd or the UDP/TCP
* socket. When input is present the data is read and processed by either
* readFromIPSocket() or readFromTAPSocket()
*/
while(*keep_running) {
int rc, max_sock = 0;
fd_set socket_mask;
struct timeval wait_time;
time_t nowTime;
FD_ZERO(&socket_mask);
FD_SET(eee->udp_sock, &socket_mask);
FD_SET(eee->udp_mgmt_sock, &socket_mask);
max_sock = max(eee->udp_sock, eee->udp_mgmt_sock);
#ifndef SKIP_MULTICAST_PEERS_DISCOVERY
FD_SET(eee->udp_multicast_sock, &socket_mask);
max_sock = max(eee->udp_sock, eee->udp_multicast_sock);
#endif
#ifndef WIN32
FD_SET(eee->device.fd, &socket_mask);
max_sock = max(max_sock, eee->device.fd);
#endif
wait_time.tv_sec = SOCKET_TIMEOUT_INTERVAL_SECS; wait_time.tv_usec = 0;
rc = select(max_sock+1, &socket_mask, NULL, NULL, &wait_time);
nowTime=time(NULL);
/* Make sure ciphers are updated before the packet is treated. */
if((nowTime - lastTransop) > TRANSOP_TICK_INTERVAL) {
lastTransop = nowTime;
eee->transop.tick(&eee->transop, nowTime);
}
if(rc > 0) {
/* Any or all of the FDs could have input; check them all. */
if(FD_ISSET(eee->udp_sock, &socket_mask)) {
/* Read a cooked socket from the internet socket (unicast). Writes on the TAP
* socket. */
readFromIPSocket(eee, eee->udp_sock);
}
#ifndef SKIP_MULTICAST_PEERS_DISCOVERY
if(FD_ISSET(eee->udp_multicast_sock, &socket_mask)) {
/* Read a cooked socket from the internet socket (multicast). Writes on the TAP
* socket. */
traceEvent(TRACE_DEBUG, "Received packet from multicast socket");
readFromIPSocket(eee, eee->udp_multicast_sock);
}
#endif
#ifdef __ANDROID_NDK__
if (uip_arp_len != 0) {
readFromTAPSocket(eee);
uip_arp_len = 0;
}
#endif /* #ifdef __ANDROID_NDK__ */
if(FD_ISSET(eee->udp_mgmt_sock, &socket_mask)) {
/* Read a cooked socket from the internet socket. Writes on the TAP
* socket. */
readFromMgmtSocket(eee, keep_running);
}
#ifndef WIN32
if(FD_ISSET(eee->device.fd, &socket_mask)) {
/* Read an ethernet frame from the TAP socket. Write on the IP
* socket. */
readFromTAPSocket(eee);
}
#endif
}
/* Finished processing select data. */
update_supernode_reg(eee, nowTime);
numPurged = purge_expired_registrations(&eee->known_peers, &last_purge_known);
numPurged += purge_expired_registrations(&eee->pending_peers, &last_purge_pending);
if(numPurged > 0) {
traceEvent(TRACE_INFO, "%u peers removed. now: pending=%u, operational=%u",
numPurged,
HASH_COUNT(eee->pending_peers),
HASH_COUNT(eee->known_peers));
}
if(eee->conf.dyn_ip_mode &&
((nowTime - lastIfaceCheck) > IFACE_UPDATE_INTERVAL)) {
traceEvent(TRACE_NORMAL, "Re-checking dynamic IP address.");
tuntap_get_address(&(eee->device));
lastIfaceCheck = nowTime;
}
#ifdef __ANDROID_NDK__
if ((nowTime - lastArpPeriod) > ARP_PERIOD_INTERVAL) {
uip_arp_timer();
lastArpPeriod = nowTime;
}
#endif /* #ifdef __ANDROID_NDK__ */
} /* while */
#ifdef WIN32
WaitForSingleObject(tun_read_thread, INFINITE);
#endif
send_deregister(eee, &(eee->supernode));
closesocket(eee->udp_sock);
return(0);
}
/* ************************************** */
/** Deinitialise the edge and deallocate any owned memory. */
void edge_term(n2n_edge_t * eee) {
if(eee->udp_sock >= 0)
closesocket(eee->udp_sock);
if(eee->udp_mgmt_sock >= 0)
closesocket(eee->udp_mgmt_sock);
#ifndef SKIP_MULTICAST_PEERS_DISCOVERY
if(eee->udp_multicast_sock >= 0)
closesocket(eee->udp_multicast_sock);
#endif
clear_peer_list(&eee->pending_peers);
clear_peer_list(&eee->known_peers);
eee->transop.deinit(&eee->transop);
free(eee);
}
/* ************************************** */
static int edge_init_sockets(n2n_edge_t *eee, int udp_local_port, int mgmt_port, uint8_t tos) {
int sockopt;
if(udp_local_port > 0)
traceEvent(TRACE_NORMAL, "Binding to local port %d", udp_local_port);
eee->udp_sock = open_socket(udp_local_port, 1 /* bind ANY */);
if(eee->udp_sock < 0) {
traceEvent(TRACE_ERROR, "Failed to bind main UDP port %u", udp_local_port);
return(-1);
}
if(tos) {
/* https://www.tucny.com/Home/dscp-tos */
sockopt = tos;
if(setsockopt(eee->udp_sock, IPPROTO_IP, IP_TOS, &sockopt, sizeof(sockopt)) == 0)
traceEvent(TRACE_NORMAL, "TOS set to 0x%x", tos);
else
traceEvent(TRACE_ERROR, "Could not set TOS 0x%x[%d]: %s", tos, errno, strerror(errno));
}
#ifdef IP_PMTUDISC_DO
sockopt = (eee->conf.disable_pmtu_discovery) ? IP_PMTUDISC_DONT : IP_PMTUDISC_DO;
if(setsockopt(eee->udp_sock, IPPROTO_IP, IP_MTU_DISCOVER, &sockopt, sizeof(sockopt)) < 0)
traceEvent(TRACE_WARNING, "Could not %s PMTU discovery[%d]: %s",
(eee->conf.disable_pmtu_discovery) ? "disable" : "enable", errno, strerror(errno));
else
traceEvent(TRACE_DEBUG, "PMTU discovery %s", (eee->conf.disable_pmtu_discovery) ? "disabled" : "enabled");
#endif
eee->udp_mgmt_sock = open_socket(mgmt_port, 0 /* bind LOOPBACK */);
if(eee->udp_mgmt_sock < 0) {
traceEvent(TRACE_ERROR, "Failed to bind management UDP port %u", mgmt_port);
return(-2);
}
#ifndef SKIP_MULTICAST_PEERS_DISCOVERY
/* Populate the multicast group for local edge */
eee->multicast_peer.family = AF_INET;
eee->multicast_peer.port = N2N_MULTICAST_PORT;
eee->multicast_peer.addr.v4[0] = 224; /* N2N_MULTICAST_GROUP */
eee->multicast_peer.addr.v4[1] = 0;
eee->multicast_peer.addr.v4[2] = 0;
eee->multicast_peer.addr.v4[3] = 68;
eee->udp_multicast_sock = open_socket(N2N_MULTICAST_PORT, 1 /* bind ANY */);
if(eee->udp_multicast_sock < 0)
return(-3);
else {
u_int enable_reuse = 1;
/* allow multiple sockets to use the same PORT number */
setsockopt(eee->udp_multicast_sock, SOL_SOCKET, SO_REUSEADDR, (char *)&enable_reuse, sizeof(enable_reuse));
#ifdef SO_REUSEPORT /* no SO_REUSEPORT in Windows / old linux versions */
setsockopt(eee->udp_multicast_sock, SOL_SOCKET, SO_REUSEPORT, &enable_reuse, sizeof(enable_reuse));
#endif
}
#endif
return(0);
}
/* ************************************** */
void edge_init_conf_defaults(n2n_edge_conf_t *conf) {
memset(conf, 0, sizeof(*conf));
conf->local_port = 0 /* any port */;
conf->mgmt_port = N2N_EDGE_MGMT_PORT; /* 5644 by default */
conf->transop_id = N2N_TRANSFORM_ID_NULL;
conf->drop_multicast = 1;
conf->allow_p2p = 1;
conf->disable_pmtu_discovery = 1;
conf->register_interval = REGISTER_SUPER_INTERVAL_DFL;
if(getenv("N2N_KEY")) {
conf->encrypt_key = strdup(getenv("N2N_KEY"));
conf->transop_id = N2N_TRANSFORM_ID_TWOFISH;
}
}
/* ************************************** */
const n2n_edge_conf_t* edge_get_conf(const n2n_edge_t *eee) {
return(&eee->conf);
}
/* ************************************** */
int edge_conf_add_supernode(n2n_edge_conf_t *conf, const char *ip_and_port) {
if(conf->sn_num >= N2N_EDGE_NUM_SUPERNODES)
return(-1);
strncpy((conf->sn_ip_array[conf->sn_num]), ip_and_port, N2N_EDGE_SN_HOST_SIZE);
traceEvent(TRACE_NORMAL, "Adding supernode[%u] = %s", (unsigned int)conf->sn_num, (conf->sn_ip_array[conf->sn_num]));
conf->sn_num++;
return(0);
}
/* ************************************** */
int quick_edge_init(char *device_name, char *community_name,
char *encrypt_key, char *device_mac,
char *local_ip_address,
char *supernode_ip_address_port,
int *keep_on_running) {
tuntap_dev tuntap;
n2n_edge_t *eee;
n2n_edge_conf_t conf;
int rv;
/* Setup the configuration */
edge_init_conf_defaults(&conf);
conf.encrypt_key = encrypt_key;
conf.transop_id = N2N_TRANSFORM_ID_TWOFISH;
snprintf((char*)conf.community_name, sizeof(conf.community_name), "%s", community_name);
edge_conf_add_supernode(&conf, supernode_ip_address_port);
/* Validate configuration */
if(edge_verify_conf(&conf) != 0)
return(-1);
/* Open the tuntap device */
if(tuntap_open(&tuntap, device_name, "static",
local_ip_address, "255.255.255.0",
device_mac, DEFAULT_MTU) < 0)
return(-2);
/* Init edge */
if((eee = edge_init(&tuntap, &conf, &rv)) == NULL)
goto quick_edge_init_end;
rv = run_edge_loop(eee, keep_on_running);
edge_term(eee);
quick_edge_init_end:
tuntap_close(&tuntap);
return(rv);
}