#include "opt_inet.h"
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/bus.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/proc.h>
#include <sys/caps.h>
#include <sys/protosw.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/systm.h>
#include <sys/thread2.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_types.h>
#include <net/ifq_var.h>
#include <net/route.h>
#include <net/if_clone.h>
#include <net/netmsg2.h>
#include <net/netisr2.h>
#ifdef INET
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/in_var.h>
#include <netinet/ip.h>
#include <netinet/ip_gre.h>
#include <netinet/ip_var.h>
#include <netinet/ip_encap.h>
#else
#error "Huh? if_gre without inet?"
#endif
#include <net/bpf.h>
#include <net/net_osdep.h>
#include "if_gre.h"
#define GREMTU 1476
#define GRENAME "gre"
#define MTAG_GRE 1709049137
static MALLOC_DEFINE(M_GRE, GRENAME, "Generic Routing Encapsulation");
struct gre_softc_head gre_softc_list;
static int gre_clone_create(struct if_clone *, int, caddr_t, caddr_t);
static int gre_clone_destroy(struct ifnet *);
static int gre_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
static int gre_output(struct ifnet *, struct mbuf *, struct sockaddr *,
struct rtentry *rt);
static struct if_clone gre_cloner = IF_CLONE_INITIALIZER("gre",
gre_clone_create, gre_clone_destroy, 0, IF_MAXUNIT);
static int gre_compute_route(struct gre_softc *sc, struct route *);
static int gre_check_route(struct gre_softc *sc);
static void greattach(void);
#ifdef INET
extern struct domain inetdomain;
static const struct protosw in_gre_protosw =
{
.pr_type = SOCK_RAW,
.pr_domain = &inetdomain,
.pr_protocol = IPPROTO_GRE,
.pr_flags = PR_ATOMIC|PR_ADDR,
.pr_input = gre_input,
.pr_output = rip_output,
.pr_ctlinput = NULL,
.pr_ctloutput = rip_ctloutput,
.pr_ctlport = NULL,
.pr_usrreqs = &rip_usrreqs
};
static const struct protosw in_mobile_protosw =
{
.pr_type = SOCK_RAW,
.pr_domain = &inetdomain,
.pr_protocol = IPPROTO_MOBILE,
.pr_flags = PR_ATOMIC|PR_ADDR,
.pr_input = gre_mobile_input,
.pr_output = rip_output,
.pr_ctlinput = NULL,
.pr_ctloutput = rip_ctloutput,
.pr_ctlport = NULL,
.pr_usrreqs = &rip_usrreqs
};
#endif
SYSCTL_DECL(_net_link);
SYSCTL_NODE(_net_link, IFT_OTHER, gre, CTLFLAG_RW, 0,
"Generic Routing Encapsulation");
#ifndef MAX_GRE_NEST
#define MAX_GRE_NEST 1
#endif
static int max_gre_nesting = MAX_GRE_NEST;
SYSCTL_INT(_net_link_gre, OID_AUTO, max_nesting, CTLFLAG_RW,
&max_gre_nesting, 0, "Max nested tunnels");
static void
greattach(void)
{
LIST_INIT(&gre_softc_list);
if_clone_attach(&gre_cloner);
}
static int
gre_clone_create(struct if_clone *ifc, int unit,
caddr_t params __unused, caddr_t data __unused)
{
struct gre_softc *sc;
sc = kmalloc(sizeof(struct gre_softc), M_GRE, M_WAITOK);
memset(sc, 0, sizeof(struct gre_softc));
sc->sc_if.if_softc = sc;
if_initname(&(sc->sc_if), GRENAME, unit);
ifq_set_maxlen(&sc->sc_if.if_snd, IFQ_MAXLEN);
sc->sc_if.if_type = IFT_OTHER;
sc->sc_if.if_addrlen = 0;
sc->sc_if.if_hdrlen = 24;
sc->sc_if.if_mtu = GREMTU;
sc->sc_if.if_flags = IFF_POINTOPOINT|IFF_MULTICAST;
sc->sc_if.if_output = gre_output;
sc->sc_if.if_ioctl = gre_ioctl;
sc->g_dst.s_addr = sc->g_src.s_addr = INADDR_ANY;
sc->g_proto = IPPROTO_GRE;
sc->sc_if.if_flags |= IFF_LINK0;
sc->encap = NULL;
sc->route_pcpu = kmalloc(netisr_ncpus * sizeof(struct route), M_GRE,
M_WAITOK | M_ZERO);
if_attach(&sc->sc_if, NULL);
bpfattach(&sc->sc_if, DLT_NULL, sizeof(u_int32_t));
LIST_INSERT_HEAD(&gre_softc_list, sc, sc_list);
return (0);
}
static int
gre_clone_destroy(struct ifnet *ifp)
{
struct gre_softc *sc = ifp->if_softc;
int cpu;
#ifdef INET
if (sc->encap != NULL)
encap_detach(sc->encap);
#endif
LIST_REMOVE(sc, sc_list);
bpfdetach(ifp);
if_detach(ifp);
for (cpu = 0; cpu < netisr_ncpus; ++cpu) {
if (sc->route_pcpu[cpu].ro_rt != NULL) {
rtfree_async(sc->route_pcpu[cpu].ro_rt);
sc->route_pcpu[cpu].ro_rt = NULL;
}
}
kfree(sc->route_pcpu, M_GRE);
kfree(sc, M_GRE);
return 0;
}
static int
gre_output_serialized(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
struct rtentry *rt)
{
int error = 0;
struct gre_softc *sc = ifp->if_softc;
struct greip *gh;
struct ip *ip;
u_short etype = 0;
struct mobile_h mob_h;
struct route *ro;
struct sockaddr_in *ro_dst;
ASSERT_NETISR_NCPUS(mycpuid);
error = if_tunnel_check_nesting(ifp, m, MTAG_GRE, max_gre_nesting);
if (error != 0) {
m_freem(m);
goto end;
}
if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) == 0 ||
sc->g_src.s_addr == INADDR_ANY || sc->g_dst.s_addr == INADDR_ANY) {
m_freem(m);
error = ENETDOWN;
goto end;
}
ro = &sc->route_pcpu[mycpuid];
ro_dst = (struct sockaddr_in *)&ro->ro_dst;
if (ro->ro_rt != NULL &&
((ro->ro_rt->rt_flags & RTF_UP) == 0 ||
ro_dst->sin_addr.s_addr != sc->g_dst.s_addr)) {
RTFREE(ro->ro_rt);
ro->ro_rt = NULL;
}
if (ro->ro_rt == NULL) {
error = gre_compute_route(sc, ro);
if (error) {
m_freem(m);
goto end;
}
}
gh = NULL;
ip = NULL;
if (ifp->if_bpf) {
bpf_gettoken();
if (ifp->if_bpf) {
uint32_t af = dst->sa_family;
bpf_ptap(ifp->if_bpf, m, &af, sizeof(af));
}
bpf_reltoken();
}
m->m_flags &= ~(M_BCAST|M_MCAST);
if (sc->g_proto == IPPROTO_MOBILE) {
if (dst->sa_family == AF_INET) {
struct mbuf *m0;
int msiz;
ip = mtod(m, struct ip *);
if (ip->ip_off & htons(IP_MF | IP_OFFMASK)) {
m_freem(m);
error = EINVAL;
goto end;
}
memset(&mob_h, 0, MOB_H_SIZ_L);
mob_h.proto = (ip->ip_p) << 8;
mob_h.odst = ip->ip_dst.s_addr;
ip->ip_dst.s_addr = sc->g_dst.s_addr;
if (in_hosteq(ip->ip_src, sc->g_src)) {
msiz = MOB_H_SIZ_S;
} else {
mob_h.proto |= MOB_H_SBIT;
mob_h.osrc = ip->ip_src.s_addr;
ip->ip_src.s_addr = sc->g_src.s_addr;
msiz = MOB_H_SIZ_L;
}
mob_h.proto = htons(mob_h.proto);
mob_h.hcrc = gre_in_cksum((u_short *)&mob_h, msiz);
if ((m->m_data - msiz) < m->m_pktdat) {
MGETHDR(m0, M_NOWAIT, MT_HEADER);
if (m0 == NULL) {
m_freem(m);
error = ENOBUFS;
goto end;
}
m0->m_next = m;
m->m_data += sizeof(struct ip);
m->m_len -= sizeof(struct ip);
m0->m_pkthdr.len = m->m_pkthdr.len + msiz;
m0->m_len = msiz + sizeof(struct ip);
m0->m_data += max_linkhdr;
memcpy(mtod(m0, caddr_t), (caddr_t)ip,
sizeof(struct ip));
m = m0;
} else {
m->m_data -= msiz;
m->m_len += msiz;
m->m_pkthdr.len += msiz;
bcopy(ip, mtod(m, caddr_t),
sizeof(struct ip));
}
ip = mtod(m, struct ip *);
memcpy((caddr_t)(ip + 1), &mob_h, (unsigned)msiz);
ip->ip_len = htons(ntohs(ip->ip_len) + msiz);
} else {
m_freem(m);
error = EINVAL;
goto end;
}
} else if (sc->g_proto == IPPROTO_GRE) {
switch (dst->sa_family) {
case AF_INET:
ip = mtod(m, struct ip *);
etype = ETHERTYPE_IP;
break;
default:
m_freem(m);
error = EAFNOSUPPORT;
goto end;
}
M_PREPEND(m, sizeof(struct greip), M_NOWAIT);
} else {
m_freem(m);
error = EINVAL;
goto end;
}
if (m == NULL) {
error = ENOBUFS;
goto end;
}
gh = mtod(m, struct greip *);
if (sc->g_proto == IPPROTO_GRE) {
memset((void *)&gh->gi_g, 0, sizeof(struct gre_h));
gh->gi_ptype = htons(etype);
}
gh->gi_pr = sc->g_proto;
if (sc->g_proto != IPPROTO_MOBILE) {
gh->gi_src = sc->g_src;
gh->gi_dst = sc->g_dst;
((struct ip*)gh)->ip_hl = (sizeof(struct ip)) >> 2;
((struct ip*)gh)->ip_ttl = GRE_TTL;
((struct ip*)gh)->ip_tos = ip->ip_tos;
((struct ip*)gh)->ip_id = ip->ip_id;
gh->gi_len = htons(m->m_pkthdr.len);
}
IFNET_STAT_INC(ifp, opackets, 1);
IFNET_STAT_INC(ifp, obytes, m->m_pkthdr.len);
error = ip_output(m, NULL, ro, IP_DEBUGROUTE, NULL, NULL);
end:
if (error)
IFNET_STAT_INC(ifp, oerrors, 1);
return (error);
}
static int
gre_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
struct rtentry *rt)
{
struct ifaltq_subque *ifsq = ifq_get_subq_default(&ifp->if_snd);
int error;
ifsq_serialize_hw(ifsq);
error = gre_output_serialized(ifp, m, dst, rt);
ifsq_deserialize_hw(ifsq);
return error;
}
static int
gre_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
{
struct ifreq *ifr = (struct ifreq *)data;
struct if_laddrreq *lifr = (struct if_laddrreq *)data;
struct in_aliasreq *aifr = (struct in_aliasreq *)data;
struct gre_softc *sc = ifp->if_softc;
struct sockaddr_in si;
struct sockaddr *sa = NULL;
int error;
struct sockaddr_in sp, sm, dp, dm;
error = 0;
crit_enter();
switch (cmd) {
case SIOCSIFADDR:
ifp->if_flags |= IFF_UP;
break;
case SIOCSIFDSTADDR:
break;
case SIOCSIFFLAGS:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
if ((ifr->ifr_flags & IFF_LINK0) != 0)
sc->g_proto = IPPROTO_GRE;
else
sc->g_proto = IPPROTO_MOBILE;
goto recompute;
case SIOCSIFMTU:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
if (ifr->ifr_mtu < 576) {
error = EINVAL;
break;
}
ifp->if_mtu = ifr->ifr_mtu;
break;
case SIOCGIFMTU:
ifr->ifr_mtu = sc->sc_if.if_mtu;
break;
case SIOCADDMULTI:
case SIOCDELMULTI:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
if (ifr == NULL) {
error = EAFNOSUPPORT;
break;
}
switch (ifr->ifr_addr.sa_family) {
#ifdef INET
case AF_INET:
break;
#endif
default:
error = EAFNOSUPPORT;
break;
}
break;
case GRESPROTO:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
sc->g_proto = ifr->ifr_flags;
switch (sc->g_proto) {
case IPPROTO_GRE:
ifp->if_flags |= IFF_LINK0;
break;
case IPPROTO_MOBILE:
ifp->if_flags &= ~IFF_LINK0;
break;
default:
error = EPROTONOSUPPORT;
break;
}
goto recompute;
case GREGPROTO:
ifr->ifr_flags = sc->g_proto;
break;
case GRESADDRS:
case GRESADDRD:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
sa = &ifr->ifr_addr;
if (cmd == GRESADDRS)
sc->g_src = (satosin(sa))->sin_addr;
if (cmd == GRESADDRD)
sc->g_dst = (satosin(sa))->sin_addr;
recompute:
#ifdef INET
if (sc->encap != NULL) {
encap_detach(sc->encap);
sc->encap = NULL;
}
#endif
if ((sc->g_src.s_addr != INADDR_ANY) &&
(sc->g_dst.s_addr != INADDR_ANY)) {
bzero(&sp, sizeof(sp));
bzero(&sm, sizeof(sm));
bzero(&dp, sizeof(dp));
bzero(&dm, sizeof(dm));
sp.sin_len = sm.sin_len = dp.sin_len = dm.sin_len =
sizeof(struct sockaddr_in);
sp.sin_family = sm.sin_family = dp.sin_family =
dm.sin_family = AF_INET;
sp.sin_addr = sc->g_src;
dp.sin_addr = sc->g_dst;
sm.sin_addr.s_addr = dm.sin_addr.s_addr =
INADDR_BROADCAST;
#ifdef INET
sc->encap = encap_attach(AF_INET, sc->g_proto,
sintosa(&sp), sintosa(&sm), sintosa(&dp),
sintosa(&dm), (sc->g_proto == IPPROTO_GRE) ?
&in_gre_protosw : &in_mobile_protosw, sc);
if (sc->encap == NULL)
kprintf("%s: unable to attach encap\n",
if_name(&sc->sc_if));
#endif
ifnet_deserialize_all(ifp);
error = gre_check_route(sc);
ifnet_serialize_all(ifp);
if (!error)
ifp->if_flags |= IFF_RUNNING;
else
ifp->if_flags &= ~IFF_RUNNING;
}
break;
case GREGADDRS:
memset(&si, 0, sizeof(si));
si.sin_family = AF_INET;
si.sin_len = sizeof(struct sockaddr_in);
si.sin_addr.s_addr = sc->g_src.s_addr;
sa = sintosa(&si);
ifr->ifr_addr = *sa;
break;
case GREGADDRD:
memset(&si, 0, sizeof(si));
si.sin_family = AF_INET;
si.sin_len = sizeof(struct sockaddr_in);
si.sin_addr.s_addr = sc->g_dst.s_addr;
sa = sintosa(&si);
ifr->ifr_addr = *sa;
break;
case SIOCSIFPHYADDR:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
if (aifr->ifra_addr.sin_family != AF_INET ||
aifr->ifra_dstaddr.sin_family != AF_INET)
{
error = EAFNOSUPPORT;
break;
}
if (aifr->ifra_addr.sin_len != sizeof(si) ||
aifr->ifra_dstaddr.sin_len != sizeof(si))
{
error = EINVAL;
break;
}
sc->g_src = aifr->ifra_addr.sin_addr;
sc->g_dst = aifr->ifra_dstaddr.sin_addr;
goto recompute;
case SIOCSLIFPHYADDR:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
if (lifr->addr.ss_family != AF_INET ||
lifr->dstaddr.ss_family != AF_INET) {
error = EAFNOSUPPORT;
break;
}
if (lifr->addr.ss_len != sizeof(si) ||
lifr->dstaddr.ss_len != sizeof(si)) {
error = EINVAL;
break;
}
sc->g_src = (satosin((struct sockadrr *)&lifr->addr))->sin_addr;
sc->g_dst =
(satosin((struct sockadrr *)&lifr->dstaddr))->sin_addr;
goto recompute;
case SIOCDIFPHYADDR:
error = caps_priv_check(cr, SYSCAP_RESTRICTEDROOT |
__SYSCAP_NULLCRED);
if (error)
break;
sc->g_src.s_addr = INADDR_ANY;
sc->g_dst.s_addr = INADDR_ANY;
goto recompute;
case SIOCGLIFPHYADDR:
if (sc->g_src.s_addr == INADDR_ANY ||
sc->g_dst.s_addr == INADDR_ANY) {
error = EADDRNOTAVAIL;
break;
}
memset(&si, 0, sizeof(si));
si.sin_family = AF_INET;
si.sin_len = sizeof(struct sockaddr_in);
si.sin_addr.s_addr = sc->g_src.s_addr;
memcpy(&lifr->addr, &si, sizeof(si));
si.sin_addr.s_addr = sc->g_dst.s_addr;
memcpy(&lifr->dstaddr, &si, sizeof(si));
break;
case SIOCGIFPSRCADDR:
if (sc->g_src.s_addr == INADDR_ANY) {
error = EADDRNOTAVAIL;
break;
}
memset(&si, 0, sizeof(si));
si.sin_family = AF_INET;
si.sin_len = sizeof(struct sockaddr_in);
si.sin_addr.s_addr = sc->g_src.s_addr;
bcopy(&si, &ifr->ifr_addr, sizeof(ifr->ifr_addr));
break;
case SIOCGIFPDSTADDR:
if (sc->g_dst.s_addr == INADDR_ANY) {
error = EADDRNOTAVAIL;
break;
}
memset(&si, 0, sizeof(si));
si.sin_family = AF_INET;
si.sin_len = sizeof(struct sockaddr_in);
si.sin_addr.s_addr = sc->g_dst.s_addr;
bcopy(&si, &ifr->ifr_addr, sizeof(ifr->ifr_addr));
break;
default:
error = EINVAL;
break;
}
crit_exit();
return (error);
}
static int
gre_compute_route(struct gre_softc *sc, struct route *ro)
{
#ifdef DIAGNOSTIC
char abuf[INET_ADDRSTRLEN];
#endif
u_int32_t a, b, c;
ASSERT_NETISR_NCPUS(mycpuid);
KASSERT(ro == &sc->route_pcpu[mycpuid], ("route mismatch"));
KASSERT(ro->ro_rt == NULL, ("rtentry not freed"));
memset(ro, 0, sizeof(struct route));
((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sc->g_dst;
ro->ro_dst.sa_family = AF_INET;
ro->ro_dst.sa_len = sizeof(ro->ro_dst);
if ((sc->sc_if.if_flags & IFF_LINK1) == 0) {
a = ntohl(sc->g_dst.s_addr);
b = a & 0x01;
c = a & 0xfffffffe;
b = b ^ 0x01;
a = b | c;
((struct sockaddr_in *)&ro->ro_dst)->sin_addr.s_addr
= htonl(a);
}
#ifdef DIAGNOSTIC
kprintf("%s: searching a route to %s", if_name(&sc->sc_if),
kinet_ntoa(((struct sockaddr_in *)&ro->ro_dst)->sin_addr, abuf));
#endif
rtalloc(ro);
if (ro->ro_rt == NULL || ro->ro_rt->rt_ifp->if_softc == sc) {
#ifdef DIAGNOSTIC
if (ro->ro_rt == NULL)
kprintf(" - no route found!\n");
else
kprintf(" - route loops back to ourself!\n");
#endif
return EADDRNOTAVAIL;
}
if ((sc->sc_if.if_flags & IFF_LINK1) == 0)
((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sc->g_dst;
#ifdef DIAGNOSTIC
kprintf(", choosing %s with gateway %s", if_name(ro->ro_rt->rt_ifp),
kinet_ntoa(
((struct sockaddr_in *)(ro->ro_rt->rt_gateway))->sin_addr, abuf));
kprintf("\n");
#endif
return 0;
}
static void
gre_check_route_handler(netmsg_t msg)
{
struct gre_softc *sc = msg->base.lmsg.u.ms_resultp;
struct route *ro;
int error;
ASSERT_NETISR0;
ro = &sc->route_pcpu[mycpuid];
if (ro->ro_rt != NULL) {
RTFREE(ro->ro_rt);
ro->ro_rt = NULL;
}
error = gre_compute_route(sc, ro);
netisr_replymsg(&msg->base, error);
}
static int
gre_check_route(struct gre_softc *sc)
{
struct netmsg_base msg;
netmsg_init(&msg, NULL, &curthread->td_msgport, MSGF_PRIORITY,
gre_check_route_handler);
msg.lmsg.u.ms_resultp = sc;
return (netisr_domsg(&msg, 0));
}
u_short
gre_in_cksum(u_short *p, u_int len)
{
u_int sum = 0;
int nwords = len >> 1;
while (nwords-- != 0)
sum += *p++;
if (len & 1) {
union {
u_short w;
u_char c[2];
} u;
u.c[0] = *(u_char *)p;
u.c[1] = 0;
sum += u.w;
}
sum = (sum >> 16) + (sum & 0xffff);
sum += (sum >> 16);
return (~sum);
}
static int
gremodevent(module_t mod, int type, void *data)
{
switch (type) {
case MOD_LOAD:
greattach();
break;
case MOD_UNLOAD:
if_clone_detach(&gre_cloner);
while (!LIST_EMPTY(&gre_softc_list))
gre_clone_destroy(&LIST_FIRST(&gre_softc_list)->sc_if);
break;
}
return 0;
}
static moduledata_t gre_mod = {
"if_gre",
gremodevent,
0
};
DECLARE_MODULE(if_gre, gre_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
MODULE_VERSION(if_gre, 1);