#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: if_stf.c,v 1.109 2022/09/03 02:47:59 thorpej Exp $");
#ifdef _KERNEL_OPT
#include "opt_inet.h"
#include "stf.h"
#endif
#ifndef INET6
#error "pseudo-device stf requires options INET6"
#endif
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/mbuf.h>
#include <sys/errno.h>
#include <sys/ioctl.h>
#include <sys/proc.h>
#include <sys/queue.h>
#include <sys/syslog.h>
#include <sys/device.h>
#include <sys/module.h>
#include <sys/cpu.h>
#include <net/if.h>
#include <net/route.h>
#include <net/if_types.h>
#include <net/if_stf.h>
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/ip_var.h>
#include <netinet/in_var.h>
#include <netinet/ip6.h>
#include <netinet6/ip6_var.h>
#include <netinet6/in6_var.h>
#include <netinet/ip_ecn.h>
#include <netinet/ip_encap.h>
#include <net/bpf.h>
#include "ioconf.h"
#define IN6_IS_ADDR_6TO4(x) (ntohs((x)->s6_addr16[0]) == 0x2002)
#define GET_V4(x) ((const struct in_addr *)(&(x)->s6_addr16[1]))
struct stf_softc {
struct ifnet sc_if;
struct route sc_ro;
const struct encaptab *encap_cookie;
LIST_ENTRY(stf_softc) sc_list;
};
static LIST_HEAD(, stf_softc) stf_softc_list;
static int stf_clone_create(struct if_clone *, int);
static int stf_clone_destroy(struct ifnet *);
struct if_clone stf_cloner =
IF_CLONE_INITIALIZER("stf", stf_clone_create, stf_clone_destroy);
static int ip_stf_ttl = STF_TTL;
extern struct domain inetdomain;
static const struct encapsw in_stf_encapsw =
{
.encapsw4 = {
.pr_input = in_stf_input,
.pr_ctlinput = NULL,
}
};
static int stf_encapcheck(struct mbuf *, int, int, void *);
static struct in6_ifaddr *stf_getsrcifa6(struct ifnet *);
static int stf_output(struct ifnet *, struct mbuf *, const struct sockaddr *,
const struct rtentry *);
static int isrfc1918addr(const struct in_addr *);
static int stf_checkaddr4(struct stf_softc *, const struct in_addr *,
struct ifnet *);
static int stf_checkaddr6(struct stf_softc *, const struct in6_addr *,
struct ifnet *);
static void stf_rtrequest(int, struct rtentry *, const struct rt_addrinfo *);
static int stf_ioctl(struct ifnet *, u_long, void *);
void
stfattach(int count)
{
}
static void
stfinit(void)
{
LIST_INIT(&stf_softc_list);
if_clone_attach(&stf_cloner);
}
static int
stfdetach(void)
{
int error = 0;
if (!LIST_EMPTY(&stf_softc_list))
error = EBUSY;
if (error == 0)
if_clone_detach(&stf_cloner);
return error;
}
static int
stf_clone_create(struct if_clone *ifc, int unit)
{
struct stf_softc *sc;
int error;
sc = malloc(sizeof(struct stf_softc), M_DEVBUF, M_WAIT|M_ZERO);
if_initname(&sc->sc_if, ifc->ifc_name, unit);
error = encap_lock_enter();
if (error) {
free(sc, M_DEVBUF);
return error;
}
if (LIST_FIRST(&stf_softc_list) != NULL) {
encap_lock_exit();
free(sc, M_DEVBUF);
return EEXIST;
}
sc->encap_cookie = encap_attach_func(AF_INET, IPPROTO_IPV6,
stf_encapcheck, &in_stf_encapsw, sc);
encap_lock_exit();
if (sc->encap_cookie == NULL) {
printf("%s: unable to attach encap\n", if_name(&sc->sc_if));
free(sc, M_DEVBUF);
return EIO;
}
sc->sc_if.if_mtu = STF_MTU;
sc->sc_if.if_flags = 0;
sc->sc_if.if_ioctl = stf_ioctl;
sc->sc_if.if_output = stf_output;
sc->sc_if.if_type = IFT_STF;
sc->sc_if.if_dlt = DLT_NULL;
if_attach(&sc->sc_if);
if_alloc_sadl(&sc->sc_if);
bpf_attach(&sc->sc_if, DLT_NULL, sizeof(u_int));
LIST_INSERT_HEAD(&stf_softc_list, sc, sc_list);
return 0;
}
static int
stf_clone_destroy(struct ifnet *ifp)
{
struct stf_softc *sc = (void *) ifp;
encap_lock_enter();
LIST_REMOVE(sc, sc_list);
encap_detach(sc->encap_cookie);
encap_lock_exit();
bpf_detach(ifp);
if_detach(ifp);
rtcache_free(&sc->sc_ro);
free(sc, M_DEVBUF);
return 0;
}
static int
stf_encapcheck(struct mbuf *m, int off, int proto, void *arg)
{
struct ip ip;
struct in6_ifaddr *ia6;
struct stf_softc *sc;
struct in_addr a, b;
sc = (struct stf_softc *)arg;
if (sc == NULL)
return 0;
if ((sc->sc_if.if_flags & IFF_UP) == 0)
return 0;
if ((sc->sc_if.if_flags & IFF_LINK0) != 0)
return 0;
if (proto != IPPROTO_IPV6)
return 0;
m_copydata(m, 0, sizeof(ip), (void *)&ip);
if (ip.ip_v != 4)
return 0;
ia6 = stf_getsrcifa6(&sc->sc_if);
if (ia6 == NULL)
return 0;
if (memcmp(GET_V4(&ia6->ia_addr.sin6_addr), &ip.ip_dst,
sizeof(ip.ip_dst)) != 0)
return 0;
memset(&a, 0, sizeof(a));
a.s_addr = GET_V4(&ia6->ia_addr.sin6_addr)->s_addr;
a.s_addr &= GET_V4(&ia6->ia_prefixmask.sin6_addr)->s_addr;
b = ip.ip_src;
b.s_addr &= GET_V4(&ia6->ia_prefixmask.sin6_addr)->s_addr;
if (a.s_addr != b.s_addr)
return 0;
return 32;
}
static struct in6_ifaddr *
stf_getsrcifa6(struct ifnet *ifp)
{
struct ifaddr *ifa;
struct in_ifaddr *ia4;
struct sockaddr_in6 *sin6;
struct in_addr in;
int s;
s = pserialize_read_enter();
IFADDR_READER_FOREACH(ifa, ifp) {
if (ifa->ifa_addr->sa_family != AF_INET6)
continue;
sin6 = (struct sockaddr_in6 *)ifa->ifa_addr;
if (!IN6_IS_ADDR_6TO4(&sin6->sin6_addr))
continue;
memcpy(&in, GET_V4(&sin6->sin6_addr), sizeof(in));
ia4 = in_get_ia(in);
if (ia4 == NULL)
continue;
pserialize_read_exit(s);
return (struct in6_ifaddr *)ifa;
}
pserialize_read_exit(s);
return NULL;
}
static int
stf_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
const struct rtentry *rt0)
{
struct rtentry *rt;
struct stf_softc *sc;
const struct sockaddr_in6 *dst6;
const struct in_addr *in4;
uint8_t tos;
struct ip *ip;
struct ip6_hdr *ip6;
struct in6_ifaddr *ia6;
union {
struct sockaddr dst;
struct sockaddr_in dst4;
} u;
sc = (struct stf_softc*)ifp;
dst6 = (const struct sockaddr_in6 *)dst;
if ((ifp->if_flags & IFF_UP) == 0) {
m_freem(m);
return ENETDOWN;
}
ia6 = stf_getsrcifa6(ifp);
if (ia6 == NULL) {
m_freem(m);
if_statinc(ifp, if_oerrors);
return ENETDOWN;
}
if (m->m_len < sizeof(*ip6)) {
m = m_pullup(m, sizeof(*ip6));
if (m == NULL) {
if_statinc(ifp, if_oerrors);
return ENOBUFS;
}
}
ip6 = mtod(m, struct ip6_hdr *);
tos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
if (IN6_IS_ADDR_6TO4(&ip6->ip6_dst))
in4 = GET_V4(&ip6->ip6_dst);
else if (IN6_IS_ADDR_6TO4(&dst6->sin6_addr))
in4 = GET_V4(&dst6->sin6_addr);
else {
m_freem(m);
if_statinc(ifp, if_oerrors);
return ENETUNREACH;
}
bpf_mtap_af(ifp, AF_INET6, m, BPF_D_OUT);
M_PREPEND(m, sizeof(struct ip), M_DONTWAIT);
if (m && m->m_len < sizeof(struct ip))
m = m_pullup(m, sizeof(struct ip));
if (m == NULL) {
if_statinc(ifp, if_oerrors);
return ENOBUFS;
}
ip = mtod(m, struct ip *);
memset(ip, 0, sizeof(*ip));
bcopy(GET_V4(&((struct sockaddr_in6 *)&ia6->ia_addr)->sin6_addr),
&ip->ip_src, sizeof(ip->ip_src));
memcpy(&ip->ip_dst, in4, sizeof(ip->ip_dst));
ip->ip_p = IPPROTO_IPV6;
ip->ip_ttl = ip_stf_ttl;
ip->ip_len = htons(m->m_pkthdr.len);
if (ifp->if_flags & IFF_LINK1)
ip_ecn_ingress(ECN_ALLOWED, &ip->ip_tos, &tos);
else
ip_ecn_ingress(ECN_NOCARE, &ip->ip_tos, &tos);
sockaddr_in_init(&u.dst4, &ip->ip_dst, 0);
if ((rt = rtcache_lookup(&sc->sc_ro, &u.dst)) == NULL) {
m_freem(m);
if_statinc(ifp, if_oerrors);
return ENETUNREACH;
}
if (rt->rt_ifp == ifp) {
rtcache_unref(rt, &sc->sc_ro);
rtcache_free(&sc->sc_ro);
m_freem(m);
if_statinc(ifp, if_oerrors);
return ENETUNREACH;
}
rtcache_unref(rt, &sc->sc_ro);
if_statadd2(ifp, if_opackets, 1,
if_obytes, m->m_pkthdr.len - sizeof(struct ip));
return ip_output(m, NULL, &sc->sc_ro, 0, NULL, NULL);
}
static int
isrfc1918addr(const struct in_addr *in)
{
if ((ntohl(in->s_addr) & 0xff000000) >> 24 == 10 ||
(ntohl(in->s_addr) & 0xfff00000) >> 16 == 172 * 256 + 16 ||
(ntohl(in->s_addr) & 0xffff0000) >> 16 == 192 * 256 + 168)
return 1;
return 0;
}
static int
stf_checkaddr4(struct stf_softc *sc, const struct in_addr *in,
struct ifnet *inifp )
{
struct in_ifaddr *ia4;
if (IN_MULTICAST(in->s_addr))
return -1;
switch ((ntohl(in->s_addr) & 0xff000000) >> 24) {
case 0: case 127: case 255:
return -1;
}
if (isrfc1918addr(in))
return -1;
if (((ntohl(in->s_addr) & 0xff000000) >> 24) == 169 &&
((ntohl(in->s_addr) & 0x00ff0000) >> 16) == 254)
return -1;
IN_ADDRLIST_READER_FOREACH(ia4) {
if ((ia4->ia_ifa.ifa_ifp->if_flags & IFF_BROADCAST) == 0)
continue;
if (in->s_addr == ia4->ia_broadaddr.sin_addr.s_addr)
return -1;
}
if (sc && (sc->sc_if.if_flags & IFF_LINK2) == 0 && inifp) {
struct sockaddr_in sin;
struct rtentry *rt;
memset(&sin, 0, sizeof(sin));
sin.sin_family = AF_INET;
sin.sin_len = sizeof(struct sockaddr_in);
sin.sin_addr = *in;
rt = rtalloc1((struct sockaddr *)&sin, 0);
if (!rt || rt->rt_ifp != inifp) {
#if 0
log(LOG_WARNING, "%s: packet from 0x%x dropped "
"due to ingress filter\n", if_name(&sc->sc_if),
(uint32_t)ntohl(sin.sin_addr.s_addr));
#endif
if (rt)
rt_unref(rt);
return -1;
}
rt_unref(rt);
}
return 0;
}
static int
stf_checkaddr6(struct stf_softc *sc, const struct in6_addr *in6,
struct ifnet *inifp )
{
if (IN6_IS_ADDR_6TO4(in6))
return stf_checkaddr4(sc, GET_V4(in6), inifp);
if (IN6_IS_ADDR_V4COMPAT(in6) || IN6_IS_ADDR_V4MAPPED(in6))
return -1;
if (IN6_IS_ADDR_LINKLOCAL(in6) || IN6_IS_ADDR_SITELOCAL(in6))
return -1;
if (IN6_IS_ADDR_MC_NODELOCAL(in6) || IN6_IS_ADDR_MC_LINKLOCAL(in6))
return -1;
return 0;
}
void
in_stf_input(struct mbuf *m, int off, int proto, void *eparg)
{
int s;
struct stf_softc *sc = eparg;
struct ip *ip;
struct ip6_hdr *ip6;
uint8_t otos, itos;
struct ifnet *ifp;
size_t pktlen;
KASSERT(sc != NULL);
if (proto != IPPROTO_IPV6) {
m_freem(m);
return;
}
ip = mtod(m, struct ip *);
if ((sc->sc_if.if_flags & IFF_UP) == 0) {
m_freem(m);
return;
}
ifp = &sc->sc_if;
if (stf_checkaddr4(sc, &ip->ip_dst, NULL) < 0 ||
stf_checkaddr4(sc, &ip->ip_src, m_get_rcvif_NOMPSAFE(m)) < 0) {
m_freem(m);
return;
}
otos = ip->ip_tos;
m_adj(m, off);
if (m->m_len < sizeof(*ip6)) {
m = m_pullup(m, sizeof(*ip6));
if (!m)
return;
}
ip6 = mtod(m, struct ip6_hdr *);
if (stf_checkaddr6(sc, &ip6->ip6_dst, NULL) < 0 ||
stf_checkaddr6(sc, &ip6->ip6_src, m_get_rcvif_NOMPSAFE(m)) < 0) {
m_freem(m);
return;
}
itos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
if ((ifp->if_flags & IFF_LINK1) != 0)
ip_ecn_egress(ECN_ALLOWED, &otos, &itos);
else
ip_ecn_egress(ECN_NOCARE, &otos, &itos);
ip6->ip6_flow &= ~htonl(0xff << 20);
ip6->ip6_flow |= htonl((uint32_t)itos << 20);
pktlen = m->m_pkthdr.len;
m_set_rcvif(m, ifp);
bpf_mtap_af(ifp, AF_INET6, m, BPF_D_IN);
s = splnet();
if (__predict_true(pktq_enqueue(ip6_pktq, m, 0))) {
if_statadd2(ifp, if_ipackets, 1, if_ibytes, pktlen);
} else {
m_freem(m);
}
splx(s);
return;
}
static void
stf_rtrequest(int cmd, struct rtentry *rt,
const struct rt_addrinfo *info)
{
if (rt != NULL) {
struct stf_softc *sc;
sc = LIST_FIRST(&stf_softc_list);
rt->rt_rmx.rmx_mtu = (sc != NULL) ? sc->sc_if.if_mtu : STF_MTU;
}
}
static int
stf_ioctl(struct ifnet *ifp, u_long cmd, void *data)
{
struct ifaddr *ifa;
struct ifreq *ifr = data;
struct sockaddr_in6 *sin6;
int error;
error = 0;
switch (cmd) {
case SIOCINITIFADDR:
ifa = (struct ifaddr *)data;
if (ifa == NULL || ifa->ifa_addr->sa_family != AF_INET6) {
error = EAFNOSUPPORT;
break;
}
sin6 = (struct sockaddr_in6 *)ifa->ifa_addr;
if (IN6_IS_ADDR_6TO4(&sin6->sin6_addr) &&
!isrfc1918addr(GET_V4(&sin6->sin6_addr))) {
ifa->ifa_rtrequest = stf_rtrequest;
ifp->if_flags |= IFF_UP;
} else
error = EINVAL;
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
if (ifr != NULL &&
ifreq_getaddr(cmd, ifr)->sa_family == AF_INET6)
;
else
error = EAFNOSUPPORT;
break;
case SIOCSIFMTU:
if (ifr->ifr_mtu < STF_MTU_MIN || ifr->ifr_mtu > STF_MTU_MAX)
return EINVAL;
else if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
error = 0;
break;
default:
error = ifioctl_common(ifp, cmd, data);
break;
}
return error;
}
#include "if_module.h"
IF_MODULE(MODULE_CLASS_DRIVER, stf, NULL)