strongswan/src/charon/network/socket.c

819 lines
19 KiB
C

/**
* @file socket.c
*
* @brief Implementation of socket_t.
*
*/
/*
* Copyright (C) 2006 Tobias Brunner, Daniel Roethlisberger
* Copyright (C) 2005-2006 Martin Willi
* Copyright (C) 2005 Jan Hutter
* Hochschule fuer Technik Rapperswil
*
* 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 2 of the License, or (at your
* option) any later version. See <http://www.fsf.org/copyleft/gpl.txt>.
*
* 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.
*/
#include <pthread.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <string.h>
#include <errno.h>
#include <unistd.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/ioctl.h>
#include <netinet/in.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>
#include <netinet/udp.h>
#include <linux/ipsec.h>
#include <linux/filter.h>
#include <net/if.h>
#include <ifaddrs.h>
#include "socket.h"
#include <daemon.h>
/* constants for packet handling */
#define IP_LEN sizeof(struct iphdr)
#define IP6_LEN sizeof(struct ip6_hdr)
#define UDP_LEN sizeof(struct udphdr)
#define MARKER_LEN sizeof(u_int32_t)
/* offsets for packet handling */
#define IP_PROTO_OFFSET 9
#define IP6_PROTO_OFFSET 6
#define IKE_VERSION_OFFSET 17
#define IKE_LENGTH_OFFSET 24
/* from linux/in.h */
#ifndef IP_IPSEC_POLICY
#define IP_IPSEC_POLICY 16
#endif /*IP_IPSEC_POLICY*/
/* from linux/udp.h */
#ifndef UDP_ENCAP
#define UDP_ENCAP 100
#endif /*UDP_ENCAP*/
#ifndef UDP_ENCAP_ESPINUDP
#define UDP_ENCAP_ESPINUDP 2
#endif /*UDP_ENCAP_ESPINUDP*/
/* needed for older kernel headers */
#ifndef IPV6_2292PKTINFO
#define IPV6_2292PKTINFO 2
#endif /*IPV6_2292PKTINFO*/
/* missing on uclibc */
#ifndef IPV6_IPSEC_POLICY
#define IPV6_IPSEC_POLICY 34
#endif /*IPV6_IPSEC_POLICY*/
typedef struct private_socket_t private_socket_t;
/**
* Private data of an socket_t object
*/
struct private_socket_t{
/**
* public functions
*/
socket_t public;
/**
* regular port
*/
int port;
/**
* port used for nat-t
*/
int natt_port;
/**
* raw receiver socket for IPv4
*/
int recv4;
/**
* raw receiver socket for IPv6
*/
int recv6;
/**
* send socket on regular port for IPv4
*/
int send4;
/**
* send socket on regular port for IPv6
*/
int send6;
/**
* send socket on nat-t port for IPv4
*/
int send4_natt;
/**
* send socket on nat-t port for IPv6
*/
int send6_natt;
};
/**
* implementation of socket_t.receive
*/
static status_t receiver(private_socket_t *this, packet_t **packet)
{
char buffer[MAX_PACKET];
chunk_t data;
packet_t *pkt;
struct udphdr *udp;
host_t *source = NULL, *dest = NULL;
int bytes_read = 0;
int data_offset, oldstate;
fd_set rfds;
FD_ZERO(&rfds);
if (this->recv4)
{
FD_SET(this->recv4, &rfds);
}
if (this->recv6)
{
FD_SET(this->recv6, &rfds);
}
DBG2(DBG_NET, "waiting for data on raw sockets");
pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
if (select(max(this->recv4, this->recv6) + 1, &rfds, NULL, NULL, NULL) <= 0)
{
pthread_setcancelstate(oldstate, NULL);
return FAILED;
}
pthread_setcancelstate(oldstate, NULL);
if (this->recv4 && FD_ISSET(this->recv4, &rfds))
{
/* IPv4 raw sockets return the IP header. We read src/dest
* information directly from the raw header */
struct iphdr *ip;
struct sockaddr_in src, dst;
bytes_read = recv(this->recv4, buffer, MAX_PACKET, 0);
if (bytes_read < 0)
{
DBG1(DBG_NET, "error reading from IPv4 socket: %m");
return FAILED;
}
DBG3(DBG_NET, "received IPv4 packet %b", buffer, bytes_read);
/* read source/dest from raw IP/UDP header */
if (bytes_read < IP_LEN + UDP_LEN + MARKER_LEN)
{
DBG1(DBG_NET, "received IPv4 packet too short (%d bytes)",
bytes_read);
return FAILED;
}
ip = (struct iphdr*) buffer;
udp = (struct udphdr*) (buffer + IP_LEN);
src.sin_family = AF_INET;
src.sin_addr.s_addr = ip->saddr;
src.sin_port = udp->source;
dst.sin_family = AF_INET;
dst.sin_addr.s_addr = ip->daddr;
dst.sin_port = udp->dest;
source = host_create_from_sockaddr((sockaddr_t*)&src);
dest = host_create_from_sockaddr((sockaddr_t*)&dst);
pkt = packet_create();
pkt->set_source(pkt, source);
pkt->set_destination(pkt, dest);
DBG2(DBG_NET, "received packet: from %#H to %#H", source, dest);
data_offset = IP_LEN + UDP_LEN;
/* remove non esp marker */
if (dest->get_port(dest) == this->natt_port)
{
data_offset += MARKER_LEN;
}
/* fill in packet */
data.len = bytes_read - data_offset;
data.ptr = malloc(data.len);
memcpy(data.ptr, buffer + data_offset, data.len);
pkt->set_data(pkt, data);
}
else if (this->recv6 && FD_ISSET(this->recv6, &rfds))
{
/* IPv6 raw sockets return no IP header. We must query
* src/dest via socket options/ancillary data */
struct msghdr msg;
struct cmsghdr *cmsgptr;
struct sockaddr_in6 src, dst;
struct iovec iov;
char ancillary[64];
msg.msg_name = &src;
msg.msg_namelen = sizeof(src);
iov.iov_base = buffer;
iov.iov_len = sizeof(buffer);
msg.msg_iov = &iov;
msg.msg_iovlen = 1;
msg.msg_control = ancillary;
msg.msg_controllen = sizeof(ancillary);
msg.msg_flags = 0;
bytes_read = recvmsg(this->recv6, &msg, 0);
if (bytes_read < 0)
{
DBG1(DBG_NET, "error reading from IPv6 socket: %m");
return FAILED;
}
DBG3(DBG_NET, "received IPv6 packet %b", buffer, bytes_read);
if (bytes_read < IP_LEN + UDP_LEN + MARKER_LEN)
{
DBG3(DBG_NET, "received IPv6 packet too short (%d bytes)",
bytes_read);
return FAILED;
}
/* read ancillary data to get destination address */
for (cmsgptr = CMSG_FIRSTHDR(&msg); cmsgptr != NULL;
cmsgptr = CMSG_NXTHDR(&msg, cmsgptr))
{
if (cmsgptr->cmsg_len == 0)
{
DBG1(DBG_NET, "error reading IPv6 ancillary data");
return FAILED;
}
if (cmsgptr->cmsg_level == SOL_IPV6 &&
cmsgptr->cmsg_type == IPV6_2292PKTINFO)
{
struct in6_pktinfo *pktinfo;
pktinfo = (struct in6_pktinfo*)CMSG_DATA(cmsgptr);
memset(&dst, 0, sizeof(dst));
memcpy(&dst.sin6_addr, &pktinfo->ipi6_addr, sizeof(dst.sin6_addr));
dst.sin6_family = AF_INET6;
udp = (struct udphdr*) (buffer);
dst.sin6_port = udp->dest;
src.sin6_port = udp->source;
dest = host_create_from_sockaddr((sockaddr_t*)&dst);
}
}
/* ancillary data missing? */
if (dest == NULL)
{
DBG1(DBG_NET, "error reading IPv6 packet header");
return FAILED;
}
source = host_create_from_sockaddr((sockaddr_t*)&src);
pkt = packet_create();
pkt->set_source(pkt, source);
pkt->set_destination(pkt, dest);
DBG2(DBG_NET, "received packet: from %#H to %#H", source, dest);
data_offset = UDP_LEN;
/* remove non esp marker */
if (dest->get_port(dest) == this->natt_port)
{
data_offset += MARKER_LEN;
}
/* fill in packet */
data.len = bytes_read - data_offset;
data.ptr = malloc(data.len);
memcpy(data.ptr, buffer + data_offset, data.len);
pkt->set_data(pkt, data);
}
else
{
/* oops, shouldn't happen */
return FAILED;
}
/* return packet */
*packet = pkt;
return SUCCESS;
}
/**
* implementation of socket_t.send
*/
status_t sender(private_socket_t *this, packet_t *packet)
{
int sport, skt, family;
ssize_t bytes_sent;
chunk_t data, marked;
host_t *src, *dst;
src = packet->get_source(packet);
dst = packet->get_destination(packet);
data = packet->get_data(packet);
DBG2(DBG_NET, "sending packet: from %#H to %#H", src, dst);
/* send data */
sport = src->get_port(src);
family = dst->get_family(dst);
if (sport == this->port)
{
if (family == AF_INET)
{
skt = this->send4;
}
else
{
skt = this->send6;
}
}
else if (sport == this->natt_port)
{
if (family == AF_INET)
{
skt = this->send4_natt;
}
else
{
skt = this->send6_natt;
}
/* NAT keepalives without marker */
if (data.len != 1 || data.ptr[0] != 0xFF)
{
/* add non esp marker to packet */
if (data.len > MAX_PACKET - MARKER_LEN)
{
DBG1(DBG_NET, "unable to send packet: it's too big (%d bytes)",
data.len);
return FAILED;
}
marked = chunk_alloc(data.len + MARKER_LEN);
memset(marked.ptr, 0, MARKER_LEN);
memcpy(marked.ptr + MARKER_LEN, data.ptr, data.len);
/* let the packet do the clean up for us */
packet->set_data(packet, marked);
data = marked;
}
}
else
{
DBG1(DBG_NET, "unable to locate a send socket for port %d", sport);
return FAILED;
}
bytes_sent = sendto(skt, data.ptr, data.len, 0,
dst->get_sockaddr(dst), *(dst->get_sockaddr_len(dst)));
if (bytes_sent != data.len)
{
DBG1(DBG_NET, "error writing to socket: %m");
return FAILED;
}
return SUCCESS;
}
/**
* implements socket_t.is_local_address
*/
static bool is_local_address(private_socket_t *this, host_t *host, char **dev)
{
struct ifaddrs *list;
struct ifaddrs *cur;
bool found = FALSE;
if (getifaddrs(&list) < 0)
{
return FALSE;
}
for (cur = list; cur != NULL; cur = cur->ifa_next)
{
if (!(cur->ifa_flags & IFF_UP))
{
/* ignore interface which are down */
continue;
}
if (cur->ifa_addr == NULL ||
cur->ifa_addr->sa_family != host->get_family(host))
{
/* no match in family */
continue;
}
switch (cur->ifa_addr->sa_family)
{
case AF_INET:
{
struct sockaddr_in *listed, *requested;
listed = (struct sockaddr_in*)cur->ifa_addr;
requested = (struct sockaddr_in*)host->get_sockaddr(host);
if (listed->sin_addr.s_addr == requested->sin_addr.s_addr)
{
found = TRUE;
}
break;
}
case AF_INET6:
{
struct sockaddr_in6 *listed, *requested;
listed = (struct sockaddr_in6*)cur->ifa_addr;
requested = (struct sockaddr_in6*)host->get_sockaddr(host);
if (memcmp(&listed->sin6_addr, &requested->sin6_addr,
sizeof(listed->sin6_addr)) == 0)
{
found = TRUE;
}
break;
}
default:
break;
}
if (found)
{
if (dev && cur->ifa_name)
{
/* return interface name, if requested */
*dev = strdup(cur->ifa_name);
}
break;
}
}
freeifaddrs(list);
return found;
}
/**
* implements socket_t.create_local_address_list
*/
static linked_list_t* create_local_address_list(private_socket_t *this)
{
struct ifaddrs *list;
struct ifaddrs *cur;
host_t *host;
linked_list_t *result = linked_list_create();
if (getifaddrs(&list) < 0)
{
return result;
}
for (cur = list; cur != NULL; cur = cur->ifa_next)
{
if (!(cur->ifa_flags & IFF_UP))
{
/* ignore interface which are down */
continue;
}
host = host_create_from_sockaddr(cur->ifa_addr);
if (host)
{
/* we use always the IKEv2 port. This is relevant for
* natd payload hashing. */
host->set_port(host, this->port);
result->insert_last(result, host);
}
}
freeifaddrs(list);
return result;
}
/**
* open a socket to send packets
*/
static int open_send_socket(private_socket_t *this, int family, u_int16_t port)
{
int on = TRUE;
int type = UDP_ENCAP_ESPINUDP;
struct sockaddr_storage addr;
u_int sol, ipsec_policy;
struct sadb_x_policy policy;
int skt;
memset(&addr, 0, sizeof(addr));
/* precalculate constants depending on address family */
switch (family)
{
case AF_INET:
{
struct sockaddr_in *sin = (struct sockaddr_in *)&addr;
sin->sin_family = AF_INET;
sin->sin_addr.s_addr = INADDR_ANY;
sin->sin_port = htons(port);
sol = SOL_IP;
ipsec_policy = IP_IPSEC_POLICY;
break;
}
case AF_INET6:
{
struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&addr;
sin6->sin6_family = AF_INET6;
memcpy(&sin6->sin6_addr, &in6addr_any, sizeof(in6addr_any));
sin6->sin6_port = htons(port);
sol = SOL_IPV6;
ipsec_policy = IPV6_IPSEC_POLICY;
break;
}
default:
return 0;
}
skt = socket(family, SOCK_DGRAM, IPPROTO_UDP);
if (skt < 0)
{
DBG1(DBG_NET, "could not open send socket: %m");
return 0;
}
if (setsockopt(skt, SOL_SOCKET, SO_REUSEADDR, (void*)&on, sizeof(on)) < 0)
{
DBG1(DBG_NET, "unable to set SO_REUSEADDR on send socket: %m");
close(skt);
return 0;
}
/* bypass outgoung IKE traffic on send socket */
memset(&policy, 0, sizeof(policy));
policy.sadb_x_policy_len = sizeof(policy) / sizeof(u_int64_t);
policy.sadb_x_policy_exttype = SADB_X_EXT_POLICY;
policy.sadb_x_policy_type = IPSEC_POLICY_BYPASS;
policy.sadb_x_policy_dir = IPSEC_DIR_OUTBOUND;
if (setsockopt(skt, sol, ipsec_policy, &policy, sizeof(policy)) < 0)
{
DBG1(DBG_NET, "unable to set IPSEC_POLICY on send socket: %m");
close(skt);
return 0;
}
/* We don't receive packets on the send socket, but we need a INBOUND policy.
* Otherwise, UDP decapsulation does not work!!! */
policy.sadb_x_policy_dir = IPSEC_DIR_INBOUND;
if (setsockopt(skt, sol, ipsec_policy, &policy, sizeof(policy)) < 0)
{
DBG1(DBG_NET, "unable to set IPSEC_POLICY on send socket: %m");
close(skt);
return 0;
}
/* bind the send socket */
if (bind(skt, (struct sockaddr *)&addr, sizeof(addr)) < 0)
{
DBG1(DBG_NET, "unable to bind send socket: %m");
close(skt);
return 0;
}
if (family == AF_INET)
{
/* enable UDP decapsulation globally, only for one socket needed */
if (setsockopt(skt, SOL_UDP, UDP_ENCAP, &type, sizeof(type)) < 0)
{
DBG1(DBG_NET, "unable to set UDP_ENCAP: %m; NAT-T may fail");
}
}
return skt;
}
/**
* open a socket to receive packets
*/
static int open_recv_socket(private_socket_t *this, int family)
{
int skt;
int on = TRUE;
u_int proto_offset, ip_len, sol, ipsec_policy, udp_header, ike_header;
struct sadb_x_policy policy;
/* precalculate constants depending on address family */
switch (family)
{
case AF_INET:
proto_offset = IP_PROTO_OFFSET;
ip_len = IP_LEN;
sol = SOL_IP;
ipsec_policy = IP_IPSEC_POLICY;
break;
case AF_INET6:
proto_offset = IP6_PROTO_OFFSET;
ip_len = 0; /* IPv6 raw sockets contain no IP header */
sol = SOL_IPV6;
ipsec_policy = IPV6_IPSEC_POLICY;
break;
default:
return 0;
}
udp_header = ip_len;
ike_header = ip_len + UDP_LEN;
/* This filter code filters out all non-IKEv2 traffic on
* a SOCK_RAW IP_PROTP_UDP socket. Handling of other
* IKE versions is done in pluto.
*/
struct sock_filter ikev2_filter_code[] =
{
/* Destination Port must be either port or natt_port */
BPF_STMT(BPF_LD+BPF_H+BPF_ABS, udp_header + 2),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, this->port, 1, 0),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, this->natt_port, 5, 12),
/* port */
/* IKE version must be 2.0 */
BPF_STMT(BPF_LD+BPF_B+BPF_ABS, ike_header + IKE_VERSION_OFFSET),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, 0x20, 0, 10),
/* packet length is length in IKEv2 header + ip header + udp header */
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, ike_header + IKE_LENGTH_OFFSET),
BPF_STMT(BPF_ALU+BPF_ADD+BPF_K, ip_len + UDP_LEN),
BPF_STMT(BPF_RET+BPF_A, 0),
/* natt_port */
/* nat-t: check for marker */
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, ike_header),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, 0, 0, 5),
/* nat-t: IKE version must be 2.0 */
BPF_STMT(BPF_LD+BPF_B+BPF_ABS, ike_header + MARKER_LEN + IKE_VERSION_OFFSET),
BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, 0x20, 0, 3),
/* nat-t: packet length is length in IKEv2 header + ip header + udp header + non esp marker */
BPF_STMT(BPF_LD+BPF_W+BPF_ABS, ike_header + MARKER_LEN + IKE_LENGTH_OFFSET),
BPF_STMT(BPF_ALU+BPF_ADD+BPF_K, ip_len + UDP_LEN + MARKER_LEN),
BPF_STMT(BPF_RET+BPF_A, 0),
/* packet doesn't match, ignore */
BPF_STMT(BPF_RET+BPF_K, 0),
};
/* Filter struct to use with setsockopt */
struct sock_fprog ikev2_filter = {
sizeof(ikev2_filter_code) / sizeof(struct sock_filter),
ikev2_filter_code
};
/* set up a raw socket */
skt = socket(family, SOCK_RAW, IPPROTO_UDP);
if (skt < 0)
{
DBG1(DBG_NET, "unable to create raw socket: %m");
return 0;
}
if (setsockopt(skt, SOL_SOCKET, SO_ATTACH_FILTER,
&ikev2_filter, sizeof(ikev2_filter)) < 0)
{
DBG1(DBG_NET, "unable to attach IKEv2 filter to raw socket: %m");
close(skt);
return 0;
}
if (family == AF_INET6 &&
/* we use IPV6_2292PKTINFO, as IPV6_PKTINFO is defined as
* 2 or 50 depending on kernel header version */
setsockopt(skt, sol, IPV6_2292PKTINFO, &on, sizeof(on)) < 0)
{
DBG1(DBG_NET, "unable to set IPV6_PKTINFO on raw socket: %m");
close(skt);
return 0;
}
/* bypass incomining IKE traffic on this socket */
memset(&policy, 0, sizeof(policy));
policy.sadb_x_policy_len = sizeof(policy) / sizeof(u_int64_t);
policy.sadb_x_policy_exttype = SADB_X_EXT_POLICY;
policy.sadb_x_policy_type = IPSEC_POLICY_BYPASS;
policy.sadb_x_policy_dir = IPSEC_DIR_INBOUND;
if (setsockopt(skt, sol, ipsec_policy, &policy, sizeof(policy)) < 0)
{
DBG1(DBG_NET, "unable to set IPSEC_POLICY on raw socket: %m");
close(skt);
return 0;
}
return skt;
}
/**
* implementation of socket_t.destroy
*/
static void destroy(private_socket_t *this)
{
if (this->recv4)
{
close(this->recv4);
}
if (this->recv6)
{
close(this->recv6);
}
if (this->send4)
{
close(this->send4);
}
if (this->send6)
{
close(this->send6);
}
if (this->send4_natt)
{
close(this->send4_natt);
}
if (this->send6_natt)
{
close(this->send6_natt);
}
free(this);
}
/*
* See header for description
*/
socket_t *socket_create(u_int16_t port, u_int16_t natt_port)
{
private_socket_t *this = malloc_thing(private_socket_t);
/* public functions */
this->public.send = (status_t(*)(socket_t*, packet_t*))sender;
this->public.receive = (status_t(*)(socket_t*, packet_t**))receiver;
this->public.is_local_address = (bool(*)(socket_t*, host_t*,char**))is_local_address;
this->public.create_local_address_list = (linked_list_t*(*)(socket_t*))create_local_address_list;
this->public.destroy = (void(*)(socket_t*)) destroy;
this->port = port;
this->natt_port = natt_port;
this->recv4 = 0;
this->recv6 = 0;
this->send4 = 0;
this->send6 = 0;
this->send4_natt = 0;
this->send6_natt = 0;
this->recv4 = open_recv_socket(this, AF_INET);
if (this->recv4 == 0)
{
DBG1(DBG_NET, "could not open IPv4 receive socket, IPv4 disabled");
}
else
{
this->send4 = open_send_socket(this, AF_INET, this->port);
if (this->send4 == 0)
{
DBG1(DBG_NET, "could not open IPv4 send socket, IPv4 disabled");
close(this->recv4);
}
else
{
this->send4_natt = open_send_socket(this, AF_INET, this->natt_port);
if (this->send4_natt == 0)
{
DBG1(DBG_NET, "could not open IPv4 NAT-T send socket");
}
}
}
this->recv6 = open_recv_socket(this, AF_INET6);
if (this->recv6 == 0)
{
DBG1(DBG_NET, "could not open IPv6 receive socket, IPv6 disabled");
}
else
{
this->send6 = open_send_socket(this, AF_INET6, this->port);
if (this->send6 == 0)
{
DBG1(DBG_NET, "could not open IPv6 send socket, IPv6 disabled");
close(this->recv6);
}
else
{
this->send6_natt = open_send_socket(this, AF_INET6, this->natt_port);
if (this->send6_natt == 0)
{
DBG1(DBG_NET, "could not open IPv6 NAT-T send socket");
}
}
}
if (!(this->send4 || this->send6) || !(this->recv4 || this->recv6))
{
DBG1(DBG_NET, "could not create any sockets");
destroy(this);
charon->kill(charon, "socket initialization failed");
}
return (socket_t*)this;
}