101#include <net/bpf.h>
102#endif
103
104#include <net/net_osdep.h>
105#include <net/if_gre.h>
106
107/*
108 * It is not easy to calculate the right value for a GRE MTU.
109 * We leave this task to the admin and use the same default that
110 * other vendors use.
111 */
112#define GREMTU 1476
113
114#define GRENAME "gre"
115
116static MALLOC_DEFINE(M_GRE, GRENAME, "Generic Routing Encapsulation");
117
118struct gre_softc_head gre_softc_list;
119
120static int gre_clone_create __P((struct if_clone *, int));
121static void gre_clone_destroy __P((struct ifnet *));
122static int gre_ioctl(struct ifnet *, u_long, caddr_t);
123static int gre_output(struct ifnet *, struct mbuf *, struct sockaddr *,
124 struct rtentry *rt);
125
126static struct if_clone gre_cloner =
127 IF_CLONE_INITIALIZER("gre", gre_clone_create, gre_clone_destroy, 0, IF_MAXUNIT);
128
129static int gre_compute_route(struct gre_softc *sc);
130
131static void greattach __P((void));
132
133#ifdef INET
134extern struct domain inetdomain;
135static const struct protosw in_gre_protosw =
136{ SOCK_RAW, &inetdomain, IPPROTO_GRE, PR_ATOMIC|PR_ADDR,
137 (pr_input_t*)gre_input, (pr_output_t*)rip_output, rip_ctlinput, rip_ctloutput,
138 0,
139 0, 0, 0, 0,
140 &rip_usrreqs
141};
142static const struct protosw in_mobile_protosw =
143{ SOCK_RAW, &inetdomain, IPPROTO_MOBILE, PR_ATOMIC|PR_ADDR,
144 (pr_input_t*)gre_mobile_input, (pr_output_t*)rip_output, rip_ctlinput, rip_ctloutput,
145 0,
146 0, 0, 0, 0,
147 &rip_usrreqs
148};
149#endif
150
151SYSCTL_DECL(_net_link);
152SYSCTL_NODE(_net_link, IFT_OTHER, gre, CTLFLAG_RW, 0,
153 "Generic Routing Encapsulation");
154#ifndef MAX_GRE_NEST
155/*
156 * This macro controls the default upper limitation on nesting of gre tunnels.
157 * Since, setting a large value to this macro with a careless configuration
158 * may introduce system crash, we don't allow any nestings by default.
159 * If you need to configure nested gre tunnels, you can define this macro
160 * in your kernel configuration file. However, if you do so, please be
161 * careful to configure the tunnels so that it won't make a loop.
162 */
163#define MAX_GRE_NEST 1
164#endif
165static int max_gre_nesting = MAX_GRE_NEST;
166SYSCTL_INT(_net_link_gre, OID_AUTO, max_nesting, CTLFLAG_RW,
167 &max_gre_nesting, 0, "Max nested tunnels");
168
169/* ARGSUSED */
170static void
171greattach(void)
172{
173
174 LIST_INIT(&gre_softc_list);
175 if_clone_attach(&gre_cloner);
176}
177
178static int
179gre_clone_create(ifc, unit)
180 struct if_clone *ifc;
181 int unit;
182{
183 struct gre_softc *sc;
184
185 sc = malloc(sizeof(struct gre_softc), M_GRE, M_WAITOK);
186 memset(sc, 0, sizeof(struct gre_softc));
187
188 sc->sc_if.if_name = GRENAME;
189 sc->sc_if.if_softc = sc;
190 sc->sc_if.if_unit = unit;
191 sc->sc_if.if_snd.ifq_maxlen = IFQ_MAXLEN;
192 sc->sc_if.if_type = IFT_OTHER;
193 sc->sc_if.if_addrlen = 0;
194 sc->sc_if.if_hdrlen = 24; /* IP + GRE */
195 sc->sc_if.if_mtu = GREMTU;
196 sc->sc_if.if_flags = IFF_POINTOPOINT|IFF_MULTICAST;
197 sc->sc_if.if_output = gre_output;
198 sc->sc_if.if_ioctl = gre_ioctl;
199 sc->g_dst.s_addr = sc->g_src.s_addr = INADDR_ANY;
200 sc->g_proto = IPPROTO_GRE;
201 sc->sc_if.if_flags |= IFF_LINK0;
202 sc->encap = NULL;
203 sc->called = 0;
204 if_attach(&sc->sc_if);
205#if NBPF
206 bpfattach(&sc->sc_if, DLT_NULL, sizeof(u_int32_t));
207#endif
208 LIST_INSERT_HEAD(&gre_softc_list, sc, sc_list);
209 return (0);
210}
211
212static void
213gre_clone_destroy(ifp)
214 struct ifnet *ifp;
215{
216 struct gre_softc *sc = ifp->if_softc;
217
218#ifdef INET
219 if (sc->encap != NULL)
220 encap_detach(sc->encap);
221#endif
222 LIST_REMOVE(sc, sc_list);
223#if NBPF
224 bpfdetach(ifp);
225#endif
226 if_detach(ifp);
227 free(sc, M_GRE);
228}
229
230/*
231 * The output routine. Takes a packet and encapsulates it in the protocol
232 * given by sc->g_proto. See also RFC 1701 and RFC 2004
233 */
234static int
235gre_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
236 struct rtentry *rt)
237{
238 int error = 0;
239 struct gre_softc *sc = ifp->if_softc;
240 struct greip *gh;
241 struct ip *ip;
242 u_char osrc;
243 u_short etype = 0;
244 struct mobile_h mob_h;
245
246 /*
247 * gre may cause infinite recursion calls when misconfigured.
248 * We'll prevent this by introducing upper limit.
249 */
250 if (++(sc->called) > max_gre_nesting) {
251 printf("%s: gre_output: recursively called too many "
252 "times(%d)\n", if_name(&sc->sc_if), sc->called);
253 m_freem(m);
254 error = EIO; /* is there better errno? */
255 goto end;
256 }
257
258 if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) == 0 ||
259 sc->g_src.s_addr == INADDR_ANY || sc->g_dst.s_addr == INADDR_ANY) {
260 m_freem(m);
261 error = ENETDOWN;
262 goto end;
263 }
264
265 gh = NULL;
266 ip = NULL;
267 osrc = 0;
268
269#if NBPF
270 if (ifp->if_bpf) {
271 /* see comment of other if_foo.c files */
272 struct mbuf m0;
273 u_int32_t af = dst->sa_family;
274
275 m0.m_next = m;
276 m0.m_len = 4;
277 m0.m_data = (char *)⁡
278
279 bpf_mtap(ifp, &m0);
280 }
281#endif
282
283 m->m_flags &= ~(M_BCAST|M_MCAST);
284
285 if (sc->g_proto == IPPROTO_MOBILE) {
286 if (dst->sa_family == AF_INET) {
287 struct mbuf *m0;
288 int msiz;
289
290 ip = mtod(m, struct ip *);
291
292 /*
293 * RFC2004 specifies that fragmented diagrams shouldn't
294 * be encapsulated.
295 */
296 if ((ip->ip_off & IP_MF) != 0) {
297 _IF_DROP(&ifp->if_snd);
298 m_freem(m);
299 error = EINVAL; /* is there better errno? */
300 goto end;
301 }
302 memset(&mob_h, 0, MOB_H_SIZ_L);
303 mob_h.proto = (ip->ip_p) << 8;
304 mob_h.odst = ip->ip_dst.s_addr;
305 ip->ip_dst.s_addr = sc->g_dst.s_addr;
306
307 /*
308 * If the packet comes from our host, we only change
309 * the destination address in the IP header.
310 * Else we also need to save and change the source
311 */
312 if (in_hosteq(ip->ip_src, sc->g_src)) {
313 msiz = MOB_H_SIZ_S;
314 } else {
315 mob_h.proto |= MOB_H_SBIT;
316 mob_h.osrc = ip->ip_src.s_addr;
317 ip->ip_src.s_addr = sc->g_src.s_addr;
318 msiz = MOB_H_SIZ_L;
319 }
320 mob_h.proto = htons(mob_h.proto);
321 mob_h.hcrc = gre_in_cksum((u_short *)&mob_h, msiz);
322
323 if ((m->m_data - msiz) < m->m_pktdat) {
324 /* need new mbuf */
325 MGETHDR(m0, M_DONTWAIT, MT_HEADER);
326 if (m0 == NULL) {
327 _IF_DROP(&ifp->if_snd);
328 m_freem(m);
329 error = ENOBUFS;
330 goto end;
331 }
332 m0->m_next = m;
333 m->m_data += sizeof(struct ip);
334 m->m_len -= sizeof(struct ip);
335 m0->m_pkthdr.len = m->m_pkthdr.len + msiz;
336 m0->m_len = msiz + sizeof(struct ip);
337 m0->m_data += max_linkhdr;
338 memcpy(mtod(m0, caddr_t), (caddr_t)ip,
339 sizeof(struct ip));
340 m = m0;
341 } else { /* we have some space left in the old one */
342 m->m_data -= msiz;
343 m->m_len += msiz;
344 m->m_pkthdr.len += msiz;
345 bcopy(ip, mtod(m, caddr_t),
346 sizeof(struct ip));
347 }
348 ip = mtod(m, struct ip *);
349 memcpy((caddr_t)(ip + 1), &mob_h, (unsigned)msiz);
350 ip->ip_len = ntohs(ip->ip_len) + msiz;
351 } else { /* AF_INET */
352 _IF_DROP(&ifp->if_snd);
353 m_freem(m);
354 error = EINVAL;
355 goto end;
356 }
357 } else if (sc->g_proto == IPPROTO_GRE) {
358 switch (dst->sa_family) {
359 case AF_INET:
360 ip = mtod(m, struct ip *);
361 etype = ETHERTYPE_IP;
362 break;
363#ifdef NETATALK
364 case AF_APPLETALK:
365 etype = ETHERTYPE_ATALK;
366 break;
367#endif
368#ifdef NS
369 case AF_NS:
370 etype = ETHERTYPE_NS;
371 break;
372#endif
373 default:
374 _IF_DROP(&ifp->if_snd);
375 m_freem(m);
376 error = EAFNOSUPPORT;
377 goto end;
378 }
379 M_PREPEND(m, sizeof(struct greip), M_DONTWAIT);
380 } else {
381 _IF_DROP(&ifp->if_snd);
382 m_freem(m);
383 error = EINVAL;
384 goto end;
385 }
386
387 if (m == NULL) { /* impossible */
388 _IF_DROP(&ifp->if_snd);
389 error = ENOBUFS;
390 goto end;
391 }
392
393 gh = mtod(m, struct greip *);
394 if (sc->g_proto == IPPROTO_GRE) {
395 /* we don't have any GRE flags for now */
396
397 memset((void *)&gh->gi_g, 0, sizeof(struct gre_h));
398 gh->gi_ptype = htons(etype);
399 }
400
401 gh->gi_pr = sc->g_proto;
402 if (sc->g_proto != IPPROTO_MOBILE) {
403 gh->gi_src = sc->g_src;
404 gh->gi_dst = sc->g_dst;
405 ((struct ip*)gh)->ip_hl = (sizeof(struct ip)) >> 2;
406 ((struct ip*)gh)->ip_ttl = GRE_TTL;
407 ((struct ip*)gh)->ip_tos = ip->ip_tos;
408 ((struct ip*)gh)->ip_id = ip->ip_id;
409 gh->gi_len = m->m_pkthdr.len;
410 }
411
412 ifp->if_opackets++;
413 ifp->if_obytes += m->m_pkthdr.len;
414 /* send it off */
415 error = ip_output(m, NULL, &sc->route, 0, NULL);
416 end:
417 sc->called = 0;
418 if (error)
419 ifp->if_oerrors++;
420 return (error);
421}
422
423static int
424gre_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
425{
426 struct ifreq *ifr = (struct ifreq *)data;
427 struct if_laddrreq *lifr = (struct if_laddrreq *)data;
428 struct in_aliasreq *aifr = (struct in_aliasreq *)data;
429 struct gre_softc *sc = ifp->if_softc;
430 int s;
431 struct sockaddr_in si;
432 struct sockaddr *sa = NULL;
433 int error;
434 struct sockaddr_in sp, sm, dp, dm;
435
436 error = 0;
437
438 s = splnet();
439 switch (cmd) {
440 case SIOCSIFADDR:
441 ifp->if_flags |= IFF_UP;
442 break;
443 case SIOCSIFDSTADDR:
444 break;
445 case SIOCSIFFLAGS:
446 if ((error = suser(curthread)) != 0)
447 break;
448 if ((ifr->ifr_flags & IFF_LINK0) != 0)
449 sc->g_proto = IPPROTO_GRE;
450 else
451 sc->g_proto = IPPROTO_MOBILE;
452 goto recompute;
453 case SIOCSIFMTU:
454 if ((error = suser(curthread)) != 0)
455 break;
456 if (ifr->ifr_mtu < 576) {
457 error = EINVAL;
458 break;
459 }
460 ifp->if_mtu = ifr->ifr_mtu;
461 break;
462 case SIOCGIFMTU:
463 ifr->ifr_mtu = sc->sc_if.if_mtu;
464 break;
465 case SIOCADDMULTI:
466 case SIOCDELMULTI:
467 if ((error = suser(curthread)) != 0)
468 break;
469 if (ifr == 0) {
470 error = EAFNOSUPPORT;
471 break;
472 }
473 switch (ifr->ifr_addr.sa_family) {
474#ifdef INET
475 case AF_INET:
476 break;
477#endif
478 default:
479 error = EAFNOSUPPORT;
480 break;
481 }
482 break;
483 case GRESPROTO:
484 if ((error = suser(curthread)) != 0)
485 break;
486 sc->g_proto = ifr->ifr_flags;
487 switch (sc->g_proto) {
488 case IPPROTO_GRE:
489 ifp->if_flags |= IFF_LINK0;
490 break;
491 case IPPROTO_MOBILE:
492 ifp->if_flags &= ~IFF_LINK0;
493 break;
494 default:
495 error = EPROTONOSUPPORT;
496 break;
497 }
498 goto recompute;
499 case GREGPROTO:
500 ifr->ifr_flags = sc->g_proto;
501 break;
502 case GRESADDRS:
503 case GRESADDRD:
504 if ((error = suser(curthread)) != 0)
505 break;
506 /*
507 * set tunnel endpoints, compute a less specific route
508 * to the remote end and mark if as up
509 */
510 sa = &ifr->ifr_addr;
511 if (cmd == GRESADDRS)
512 sc->g_src = (satosin(sa))->sin_addr;
513 if (cmd == GRESADDRD)
514 sc->g_dst = (satosin(sa))->sin_addr;
515 recompute:
516#ifdef INET
517 if (sc->encap != NULL) {
518 encap_detach(sc->encap);
519 sc->encap = NULL;
520 }
521#endif
522 if ((sc->g_src.s_addr != INADDR_ANY) &&
523 (sc->g_dst.s_addr != INADDR_ANY)) {
524 bzero(&sp, sizeof(sp));
525 bzero(&sm, sizeof(sm));
526 bzero(&dp, sizeof(dp));
527 bzero(&dm, sizeof(dm));
528 sp.sin_len = sm.sin_len = dp.sin_len = dm.sin_len =
529 sizeof(struct sockaddr_in);
530 sp.sin_family = sm.sin_family = dp.sin_family =
531 dm.sin_family = AF_INET;
532 sp.sin_addr = sc->g_src;
533 dp.sin_addr = sc->g_dst;
534 sm.sin_addr.s_addr = dm.sin_addr.s_addr =
535 INADDR_BROADCAST;
536#ifdef INET
537 sc->encap = encap_attach(AF_INET, sc->g_proto,
538 sintosa(&sp), sintosa(&sm), sintosa(&dp),
539 sintosa(&dm), (sc->g_proto == IPPROTO_GRE) ?
540 &in_gre_protosw : &in_mobile_protosw, sc);
541 if (sc->encap == NULL)
542 printf("%s: unable to attach encap\n",
543 if_name(&sc->sc_if));
544#endif
545 if (sc->route.ro_rt != 0) /* free old route */
546 RTFREE(sc->route.ro_rt);
547 if (gre_compute_route(sc) == 0)
548 ifp->if_flags |= IFF_RUNNING;
549 else
550 ifp->if_flags &= ~IFF_RUNNING;
551 }
552 break;
553 case GREGADDRS:
554 memset(&si, 0, sizeof(si));
555 si.sin_family = AF_INET;
556 si.sin_len = sizeof(struct sockaddr_in);
557 si.sin_addr.s_addr = sc->g_src.s_addr;
558 sa = sintosa(&si);
559 ifr->ifr_addr = *sa;
560 break;
561 case GREGADDRD:
562 memset(&si, 0, sizeof(si));
563 si.sin_family = AF_INET;
564 si.sin_len = sizeof(struct sockaddr_in);
565 si.sin_addr.s_addr = sc->g_dst.s_addr;
566 sa = sintosa(&si);
567 ifr->ifr_addr = *sa;
568 break;
569 case SIOCSIFPHYADDR:
570 if ((error = suser(curthread)) != 0)
571 break;
572 if (aifr->ifra_addr.sin_family != AF_INET ||
573 aifr->ifra_dstaddr.sin_family != AF_INET) {
574 error = EAFNOSUPPORT;
575 break;
576 }
577 if (aifr->ifra_addr.sin_len != sizeof(si) ||
578 aifr->ifra_dstaddr.sin_len != sizeof(si)) {
579 error = EINVAL;
580 break;
581 }
582 sc->g_src = aifr->ifra_addr.sin_addr;
583 sc->g_dst = aifr->ifra_dstaddr.sin_addr;
584 goto recompute;
585 case SIOCSLIFPHYADDR:
586 if ((error = suser(curthread)) != 0)
587 break;
588 if (lifr->addr.ss_family != AF_INET ||
589 lifr->dstaddr.ss_family != AF_INET) {
590 error = EAFNOSUPPORT;
591 break;
592 }
593 if (lifr->addr.ss_len != sizeof(si) ||
594 lifr->dstaddr.ss_len != sizeof(si)) {
595 error = EINVAL;
596 break;
597 }
598 sc->g_src = (satosin((struct sockadrr *)&lifr->addr))->sin_addr;
599 sc->g_dst =
600 (satosin((struct sockadrr *)&lifr->dstaddr))->sin_addr;
601 goto recompute;
602 case SIOCDIFPHYADDR:
603 if ((error = suser(curthread)) != 0)
604 break;
605 sc->g_src.s_addr = INADDR_ANY;
606 sc->g_dst.s_addr = INADDR_ANY;
607 goto recompute;
608 case SIOCGLIFPHYADDR:
609 if (sc->g_src.s_addr == INADDR_ANY ||
610 sc->g_dst.s_addr == INADDR_ANY) {
611 error = EADDRNOTAVAIL;
612 break;
613 }
614 memset(&si, 0, sizeof(si));
615 si.sin_family = AF_INET;
616 si.sin_len = sizeof(struct sockaddr_in);
617 si.sin_addr.s_addr = sc->g_src.s_addr;
618 memcpy(&lifr->addr, &si, sizeof(si));
619 si.sin_addr.s_addr = sc->g_dst.s_addr;
620 memcpy(&lifr->dstaddr, &si, sizeof(si));
621 break;
622 case SIOCGIFPSRCADDR:
623 if (sc->g_src.s_addr == INADDR_ANY) {
624 error = EADDRNOTAVAIL;
625 break;
626 }
627 memset(&si, 0, sizeof(si));
628 si.sin_family = AF_INET;
629 si.sin_len = sizeof(struct sockaddr_in);
630 si.sin_addr.s_addr = sc->g_src.s_addr;
631 bcopy(&si, &ifr->ifr_addr, sizeof(ifr->ifr_addr));
632 break;
633 case SIOCGIFPDSTADDR:
634 if (sc->g_dst.s_addr == INADDR_ANY) {
635 error = EADDRNOTAVAIL;
636 break;
637 }
638 memset(&si, 0, sizeof(si));
639 si.sin_family = AF_INET;
640 si.sin_len = sizeof(struct sockaddr_in);
641 si.sin_addr.s_addr = sc->g_dst.s_addr;
642 bcopy(&si, &ifr->ifr_addr, sizeof(ifr->ifr_addr));
643 break;
644 default:
645 error = EINVAL;
646 break;
647 }
648
649 splx(s);
650 return (error);
651}
652
653/*
654 * computes a route to our destination that is not the one
655 * which would be taken by ip_output(), as this one will loop back to
656 * us. If the interface is p2p as a--->b, then a routing entry exists
657 * If we now send a packet to b (e.g. ping b), this will come down here
658 * gets src=a, dst=b tacked on and would from ip_ouput() sent back to
659 * if_gre.
660 * Goal here is to compute a route to b that is less specific than
661 * a-->b. We know that this one exists as in normal operation we have
662 * at least a default route which matches.
663 */
664static int
665gre_compute_route(struct gre_softc *sc)
666{
667 struct route *ro;
668 u_int32_t a, b, c;
669
670 ro = &sc->route;
671
672 memset(ro, 0, sizeof(struct route));
673 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sc->g_dst;
674 ro->ro_dst.sa_family = AF_INET;
675 ro->ro_dst.sa_len = sizeof(ro->ro_dst);
676
677 /*
678 * toggle last bit, so our interface is not found, but a less
679 * specific route. I'd rather like to specify a shorter mask,
680 * but this is not possible. Should work though. XXX
681 * there is a simpler way ...
682 */
683 if ((sc->sc_if.if_flags & IFF_LINK1) == 0) {
684 a = ntohl(sc->g_dst.s_addr);
685 b = a & 0x01;
686 c = a & 0xfffffffe;
687 b = b ^ 0x01;
688 a = b | c;
689 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr.s_addr
690 = htonl(a);
691 }
692
693#ifdef DIAGNOSTIC
694 printf("%s: searching a route to %s", if_name(&sc->sc_if),
695 inet_ntoa(((struct sockaddr_in *)&ro->ro_dst)->sin_addr));
696#endif
697
698 rtalloc(ro);
699
700 /*
701 * check if this returned a route at all and this route is no
702 * recursion to ourself
703 */
704 if (ro->ro_rt == NULL || ro->ro_rt->rt_ifp->if_softc == sc) {
705#ifdef DIAGNOSTIC
706 if (ro->ro_rt == NULL)
707 printf(" - no route found!\n");
708 else
709 printf(" - route loops back to ourself!\n");
710#endif
711 return EADDRNOTAVAIL;
712 }
713
714 /*
715 * now change it back - else ip_output will just drop
716 * the route and search one to this interface ...
717 */
718 if ((sc->sc_if.if_flags & IFF_LINK1) == 0)
719 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sc->g_dst;
720
721#ifdef DIAGNOSTIC
722 printf(", choosing %s with gateway %s", if_name(ro->ro_rt->rt_ifp),
723 inet_ntoa(((struct sockaddr_in *)(ro->ro_rt->rt_gateway))->sin_addr));
724 printf("\n");
725#endif
726
727 return 0;
728}
729
730/*
731 * do a checksum of a buffer - much like in_cksum, which operates on
732 * mbufs.
733 */
734u_short
735gre_in_cksum(u_short *p, u_int len)
736{
737 u_int sum = 0;
738 int nwords = len >> 1;
739
740 while (nwords-- != 0)
741 sum += *p++;
742
743 if (len & 1) {
744 union {
745 u_short w;
746 u_char c[2];
747 } u;
748 u.c[0] = *(u_char *)p;
749 u.c[1] = 0;
750 sum += u.w;
751 }
752
753 /* end-around-carry */
754 sum = (sum >> 16) + (sum & 0xffff);
755 sum += (sum >> 16);
756 return (~sum);
757}
758
759static int
760gremodevent(module_t mod, int type, void *data)
761{
762
763 switch (type) {
764 case MOD_LOAD:
765 greattach();
766 break;
767 case MOD_UNLOAD:
768 if_clone_detach(&gre_cloner);
769
770 while (!LIST_EMPTY(&gre_softc_list))
771 gre_clone_destroy(&LIST_FIRST(&gre_softc_list)->sc_if);
772 break;
773 }
774 return 0;
775}
776
777static moduledata_t gre_mod = {
778 "if_gre",
779 gremodevent,
780 0
781};
782
783DECLARE_MODULE(if_gre, gre_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
784MODULE_VERSION(if_gre, 1);
| 85#include <net/bpf.h>
86#endif
87
88#include <net/net_osdep.h>
89#include <net/if_gre.h>
90
91/*
92 * It is not easy to calculate the right value for a GRE MTU.
93 * We leave this task to the admin and use the same default that
94 * other vendors use.
95 */
96#define GREMTU 1476
97
98#define GRENAME "gre"
99
100static MALLOC_DEFINE(M_GRE, GRENAME, "Generic Routing Encapsulation");
101
102struct gre_softc_head gre_softc_list;
103
104static int gre_clone_create __P((struct if_clone *, int));
105static void gre_clone_destroy __P((struct ifnet *));
106static int gre_ioctl(struct ifnet *, u_long, caddr_t);
107static int gre_output(struct ifnet *, struct mbuf *, struct sockaddr *,
108 struct rtentry *rt);
109
110static struct if_clone gre_cloner =
111 IF_CLONE_INITIALIZER("gre", gre_clone_create, gre_clone_destroy, 0, IF_MAXUNIT);
112
113static int gre_compute_route(struct gre_softc *sc);
114
115static void greattach __P((void));
116
117#ifdef INET
118extern struct domain inetdomain;
119static const struct protosw in_gre_protosw =
120{ SOCK_RAW, &inetdomain, IPPROTO_GRE, PR_ATOMIC|PR_ADDR,
121 (pr_input_t*)gre_input, (pr_output_t*)rip_output, rip_ctlinput, rip_ctloutput,
122 0,
123 0, 0, 0, 0,
124 &rip_usrreqs
125};
126static const struct protosw in_mobile_protosw =
127{ SOCK_RAW, &inetdomain, IPPROTO_MOBILE, PR_ATOMIC|PR_ADDR,
128 (pr_input_t*)gre_mobile_input, (pr_output_t*)rip_output, rip_ctlinput, rip_ctloutput,
129 0,
130 0, 0, 0, 0,
131 &rip_usrreqs
132};
133#endif
134
135SYSCTL_DECL(_net_link);
136SYSCTL_NODE(_net_link, IFT_OTHER, gre, CTLFLAG_RW, 0,
137 "Generic Routing Encapsulation");
138#ifndef MAX_GRE_NEST
139/*
140 * This macro controls the default upper limitation on nesting of gre tunnels.
141 * Since, setting a large value to this macro with a careless configuration
142 * may introduce system crash, we don't allow any nestings by default.
143 * If you need to configure nested gre tunnels, you can define this macro
144 * in your kernel configuration file. However, if you do so, please be
145 * careful to configure the tunnels so that it won't make a loop.
146 */
147#define MAX_GRE_NEST 1
148#endif
149static int max_gre_nesting = MAX_GRE_NEST;
150SYSCTL_INT(_net_link_gre, OID_AUTO, max_nesting, CTLFLAG_RW,
151 &max_gre_nesting, 0, "Max nested tunnels");
152
153/* ARGSUSED */
154static void
155greattach(void)
156{
157
158 LIST_INIT(&gre_softc_list);
159 if_clone_attach(&gre_cloner);
160}
161
162static int
163gre_clone_create(ifc, unit)
164 struct if_clone *ifc;
165 int unit;
166{
167 struct gre_softc *sc;
168
169 sc = malloc(sizeof(struct gre_softc), M_GRE, M_WAITOK);
170 memset(sc, 0, sizeof(struct gre_softc));
171
172 sc->sc_if.if_name = GRENAME;
173 sc->sc_if.if_softc = sc;
174 sc->sc_if.if_unit = unit;
175 sc->sc_if.if_snd.ifq_maxlen = IFQ_MAXLEN;
176 sc->sc_if.if_type = IFT_OTHER;
177 sc->sc_if.if_addrlen = 0;
178 sc->sc_if.if_hdrlen = 24; /* IP + GRE */
179 sc->sc_if.if_mtu = GREMTU;
180 sc->sc_if.if_flags = IFF_POINTOPOINT|IFF_MULTICAST;
181 sc->sc_if.if_output = gre_output;
182 sc->sc_if.if_ioctl = gre_ioctl;
183 sc->g_dst.s_addr = sc->g_src.s_addr = INADDR_ANY;
184 sc->g_proto = IPPROTO_GRE;
185 sc->sc_if.if_flags |= IFF_LINK0;
186 sc->encap = NULL;
187 sc->called = 0;
188 if_attach(&sc->sc_if);
189#if NBPF
190 bpfattach(&sc->sc_if, DLT_NULL, sizeof(u_int32_t));
191#endif
192 LIST_INSERT_HEAD(&gre_softc_list, sc, sc_list);
193 return (0);
194}
195
196static void
197gre_clone_destroy(ifp)
198 struct ifnet *ifp;
199{
200 struct gre_softc *sc = ifp->if_softc;
201
202#ifdef INET
203 if (sc->encap != NULL)
204 encap_detach(sc->encap);
205#endif
206 LIST_REMOVE(sc, sc_list);
207#if NBPF
208 bpfdetach(ifp);
209#endif
210 if_detach(ifp);
211 free(sc, M_GRE);
212}
213
214/*
215 * The output routine. Takes a packet and encapsulates it in the protocol
216 * given by sc->g_proto. See also RFC 1701 and RFC 2004
217 */
218static int
219gre_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
220 struct rtentry *rt)
221{
222 int error = 0;
223 struct gre_softc *sc = ifp->if_softc;
224 struct greip *gh;
225 struct ip *ip;
226 u_char osrc;
227 u_short etype = 0;
228 struct mobile_h mob_h;
229
230 /*
231 * gre may cause infinite recursion calls when misconfigured.
232 * We'll prevent this by introducing upper limit.
233 */
234 if (++(sc->called) > max_gre_nesting) {
235 printf("%s: gre_output: recursively called too many "
236 "times(%d)\n", if_name(&sc->sc_if), sc->called);
237 m_freem(m);
238 error = EIO; /* is there better errno? */
239 goto end;
240 }
241
242 if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) == 0 ||
243 sc->g_src.s_addr == INADDR_ANY || sc->g_dst.s_addr == INADDR_ANY) {
244 m_freem(m);
245 error = ENETDOWN;
246 goto end;
247 }
248
249 gh = NULL;
250 ip = NULL;
251 osrc = 0;
252
253#if NBPF
254 if (ifp->if_bpf) {
255 /* see comment of other if_foo.c files */
256 struct mbuf m0;
257 u_int32_t af = dst->sa_family;
258
259 m0.m_next = m;
260 m0.m_len = 4;
261 m0.m_data = (char *)⁡
262
263 bpf_mtap(ifp, &m0);
264 }
265#endif
266
267 m->m_flags &= ~(M_BCAST|M_MCAST);
268
269 if (sc->g_proto == IPPROTO_MOBILE) {
270 if (dst->sa_family == AF_INET) {
271 struct mbuf *m0;
272 int msiz;
273
274 ip = mtod(m, struct ip *);
275
276 /*
277 * RFC2004 specifies that fragmented diagrams shouldn't
278 * be encapsulated.
279 */
280 if ((ip->ip_off & IP_MF) != 0) {
281 _IF_DROP(&ifp->if_snd);
282 m_freem(m);
283 error = EINVAL; /* is there better errno? */
284 goto end;
285 }
286 memset(&mob_h, 0, MOB_H_SIZ_L);
287 mob_h.proto = (ip->ip_p) << 8;
288 mob_h.odst = ip->ip_dst.s_addr;
289 ip->ip_dst.s_addr = sc->g_dst.s_addr;
290
291 /*
292 * If the packet comes from our host, we only change
293 * the destination address in the IP header.
294 * Else we also need to save and change the source
295 */
296 if (in_hosteq(ip->ip_src, sc->g_src)) {
297 msiz = MOB_H_SIZ_S;
298 } else {
299 mob_h.proto |= MOB_H_SBIT;
300 mob_h.osrc = ip->ip_src.s_addr;
301 ip->ip_src.s_addr = sc->g_src.s_addr;
302 msiz = MOB_H_SIZ_L;
303 }
304 mob_h.proto = htons(mob_h.proto);
305 mob_h.hcrc = gre_in_cksum((u_short *)&mob_h, msiz);
306
307 if ((m->m_data - msiz) < m->m_pktdat) {
308 /* need new mbuf */
309 MGETHDR(m0, M_DONTWAIT, MT_HEADER);
310 if (m0 == NULL) {
311 _IF_DROP(&ifp->if_snd);
312 m_freem(m);
313 error = ENOBUFS;
314 goto end;
315 }
316 m0->m_next = m;
317 m->m_data += sizeof(struct ip);
318 m->m_len -= sizeof(struct ip);
319 m0->m_pkthdr.len = m->m_pkthdr.len + msiz;
320 m0->m_len = msiz + sizeof(struct ip);
321 m0->m_data += max_linkhdr;
322 memcpy(mtod(m0, caddr_t), (caddr_t)ip,
323 sizeof(struct ip));
324 m = m0;
325 } else { /* we have some space left in the old one */
326 m->m_data -= msiz;
327 m->m_len += msiz;
328 m->m_pkthdr.len += msiz;
329 bcopy(ip, mtod(m, caddr_t),
330 sizeof(struct ip));
331 }
332 ip = mtod(m, struct ip *);
333 memcpy((caddr_t)(ip + 1), &mob_h, (unsigned)msiz);
334 ip->ip_len = ntohs(ip->ip_len) + msiz;
335 } else { /* AF_INET */
336 _IF_DROP(&ifp->if_snd);
337 m_freem(m);
338 error = EINVAL;
339 goto end;
340 }
341 } else if (sc->g_proto == IPPROTO_GRE) {
342 switch (dst->sa_family) {
343 case AF_INET:
344 ip = mtod(m, struct ip *);
345 etype = ETHERTYPE_IP;
346 break;
347#ifdef NETATALK
348 case AF_APPLETALK:
349 etype = ETHERTYPE_ATALK;
350 break;
351#endif
352#ifdef NS
353 case AF_NS:
354 etype = ETHERTYPE_NS;
355 break;
356#endif
357 default:
358 _IF_DROP(&ifp->if_snd);
359 m_freem(m);
360 error = EAFNOSUPPORT;
361 goto end;
362 }
363 M_PREPEND(m, sizeof(struct greip), M_DONTWAIT);
364 } else {
365 _IF_DROP(&ifp->if_snd);
366 m_freem(m);
367 error = EINVAL;
368 goto end;
369 }
370
371 if (m == NULL) { /* impossible */
372 _IF_DROP(&ifp->if_snd);
373 error = ENOBUFS;
374 goto end;
375 }
376
377 gh = mtod(m, struct greip *);
378 if (sc->g_proto == IPPROTO_GRE) {
379 /* we don't have any GRE flags for now */
380
381 memset((void *)&gh->gi_g, 0, sizeof(struct gre_h));
382 gh->gi_ptype = htons(etype);
383 }
384
385 gh->gi_pr = sc->g_proto;
386 if (sc->g_proto != IPPROTO_MOBILE) {
387 gh->gi_src = sc->g_src;
388 gh->gi_dst = sc->g_dst;
389 ((struct ip*)gh)->ip_hl = (sizeof(struct ip)) >> 2;
390 ((struct ip*)gh)->ip_ttl = GRE_TTL;
391 ((struct ip*)gh)->ip_tos = ip->ip_tos;
392 ((struct ip*)gh)->ip_id = ip->ip_id;
393 gh->gi_len = m->m_pkthdr.len;
394 }
395
396 ifp->if_opackets++;
397 ifp->if_obytes += m->m_pkthdr.len;
398 /* send it off */
399 error = ip_output(m, NULL, &sc->route, 0, NULL);
400 end:
401 sc->called = 0;
402 if (error)
403 ifp->if_oerrors++;
404 return (error);
405}
406
407static int
408gre_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
409{
410 struct ifreq *ifr = (struct ifreq *)data;
411 struct if_laddrreq *lifr = (struct if_laddrreq *)data;
412 struct in_aliasreq *aifr = (struct in_aliasreq *)data;
413 struct gre_softc *sc = ifp->if_softc;
414 int s;
415 struct sockaddr_in si;
416 struct sockaddr *sa = NULL;
417 int error;
418 struct sockaddr_in sp, sm, dp, dm;
419
420 error = 0;
421
422 s = splnet();
423 switch (cmd) {
424 case SIOCSIFADDR:
425 ifp->if_flags |= IFF_UP;
426 break;
427 case SIOCSIFDSTADDR:
428 break;
429 case SIOCSIFFLAGS:
430 if ((error = suser(curthread)) != 0)
431 break;
432 if ((ifr->ifr_flags & IFF_LINK0) != 0)
433 sc->g_proto = IPPROTO_GRE;
434 else
435 sc->g_proto = IPPROTO_MOBILE;
436 goto recompute;
437 case SIOCSIFMTU:
438 if ((error = suser(curthread)) != 0)
439 break;
440 if (ifr->ifr_mtu < 576) {
441 error = EINVAL;
442 break;
443 }
444 ifp->if_mtu = ifr->ifr_mtu;
445 break;
446 case SIOCGIFMTU:
447 ifr->ifr_mtu = sc->sc_if.if_mtu;
448 break;
449 case SIOCADDMULTI:
450 case SIOCDELMULTI:
451 if ((error = suser(curthread)) != 0)
452 break;
453 if (ifr == 0) {
454 error = EAFNOSUPPORT;
455 break;
456 }
457 switch (ifr->ifr_addr.sa_family) {
458#ifdef INET
459 case AF_INET:
460 break;
461#endif
462 default:
463 error = EAFNOSUPPORT;
464 break;
465 }
466 break;
467 case GRESPROTO:
468 if ((error = suser(curthread)) != 0)
469 break;
470 sc->g_proto = ifr->ifr_flags;
471 switch (sc->g_proto) {
472 case IPPROTO_GRE:
473 ifp->if_flags |= IFF_LINK0;
474 break;
475 case IPPROTO_MOBILE:
476 ifp->if_flags &= ~IFF_LINK0;
477 break;
478 default:
479 error = EPROTONOSUPPORT;
480 break;
481 }
482 goto recompute;
483 case GREGPROTO:
484 ifr->ifr_flags = sc->g_proto;
485 break;
486 case GRESADDRS:
487 case GRESADDRD:
488 if ((error = suser(curthread)) != 0)
489 break;
490 /*
491 * set tunnel endpoints, compute a less specific route
492 * to the remote end and mark if as up
493 */
494 sa = &ifr->ifr_addr;
495 if (cmd == GRESADDRS)
496 sc->g_src = (satosin(sa))->sin_addr;
497 if (cmd == GRESADDRD)
498 sc->g_dst = (satosin(sa))->sin_addr;
499 recompute:
500#ifdef INET
501 if (sc->encap != NULL) {
502 encap_detach(sc->encap);
503 sc->encap = NULL;
504 }
505#endif
506 if ((sc->g_src.s_addr != INADDR_ANY) &&
507 (sc->g_dst.s_addr != INADDR_ANY)) {
508 bzero(&sp, sizeof(sp));
509 bzero(&sm, sizeof(sm));
510 bzero(&dp, sizeof(dp));
511 bzero(&dm, sizeof(dm));
512 sp.sin_len = sm.sin_len = dp.sin_len = dm.sin_len =
513 sizeof(struct sockaddr_in);
514 sp.sin_family = sm.sin_family = dp.sin_family =
515 dm.sin_family = AF_INET;
516 sp.sin_addr = sc->g_src;
517 dp.sin_addr = sc->g_dst;
518 sm.sin_addr.s_addr = dm.sin_addr.s_addr =
519 INADDR_BROADCAST;
520#ifdef INET
521 sc->encap = encap_attach(AF_INET, sc->g_proto,
522 sintosa(&sp), sintosa(&sm), sintosa(&dp),
523 sintosa(&dm), (sc->g_proto == IPPROTO_GRE) ?
524 &in_gre_protosw : &in_mobile_protosw, sc);
525 if (sc->encap == NULL)
526 printf("%s: unable to attach encap\n",
527 if_name(&sc->sc_if));
528#endif
529 if (sc->route.ro_rt != 0) /* free old route */
530 RTFREE(sc->route.ro_rt);
531 if (gre_compute_route(sc) == 0)
532 ifp->if_flags |= IFF_RUNNING;
533 else
534 ifp->if_flags &= ~IFF_RUNNING;
535 }
536 break;
537 case GREGADDRS:
538 memset(&si, 0, sizeof(si));
539 si.sin_family = AF_INET;
540 si.sin_len = sizeof(struct sockaddr_in);
541 si.sin_addr.s_addr = sc->g_src.s_addr;
542 sa = sintosa(&si);
543 ifr->ifr_addr = *sa;
544 break;
545 case GREGADDRD:
546 memset(&si, 0, sizeof(si));
547 si.sin_family = AF_INET;
548 si.sin_len = sizeof(struct sockaddr_in);
549 si.sin_addr.s_addr = sc->g_dst.s_addr;
550 sa = sintosa(&si);
551 ifr->ifr_addr = *sa;
552 break;
553 case SIOCSIFPHYADDR:
554 if ((error = suser(curthread)) != 0)
555 break;
556 if (aifr->ifra_addr.sin_family != AF_INET ||
557 aifr->ifra_dstaddr.sin_family != AF_INET) {
558 error = EAFNOSUPPORT;
559 break;
560 }
561 if (aifr->ifra_addr.sin_len != sizeof(si) ||
562 aifr->ifra_dstaddr.sin_len != sizeof(si)) {
563 error = EINVAL;
564 break;
565 }
566 sc->g_src = aifr->ifra_addr.sin_addr;
567 sc->g_dst = aifr->ifra_dstaddr.sin_addr;
568 goto recompute;
569 case SIOCSLIFPHYADDR:
570 if ((error = suser(curthread)) != 0)
571 break;
572 if (lifr->addr.ss_family != AF_INET ||
573 lifr->dstaddr.ss_family != AF_INET) {
574 error = EAFNOSUPPORT;
575 break;
576 }
577 if (lifr->addr.ss_len != sizeof(si) ||
578 lifr->dstaddr.ss_len != sizeof(si)) {
579 error = EINVAL;
580 break;
581 }
582 sc->g_src = (satosin((struct sockadrr *)&lifr->addr))->sin_addr;
583 sc->g_dst =
584 (satosin((struct sockadrr *)&lifr->dstaddr))->sin_addr;
585 goto recompute;
586 case SIOCDIFPHYADDR:
587 if ((error = suser(curthread)) != 0)
588 break;
589 sc->g_src.s_addr = INADDR_ANY;
590 sc->g_dst.s_addr = INADDR_ANY;
591 goto recompute;
592 case SIOCGLIFPHYADDR:
593 if (sc->g_src.s_addr == INADDR_ANY ||
594 sc->g_dst.s_addr == INADDR_ANY) {
595 error = EADDRNOTAVAIL;
596 break;
597 }
598 memset(&si, 0, sizeof(si));
599 si.sin_family = AF_INET;
600 si.sin_len = sizeof(struct sockaddr_in);
601 si.sin_addr.s_addr = sc->g_src.s_addr;
602 memcpy(&lifr->addr, &si, sizeof(si));
603 si.sin_addr.s_addr = sc->g_dst.s_addr;
604 memcpy(&lifr->dstaddr, &si, sizeof(si));
605 break;
606 case SIOCGIFPSRCADDR:
607 if (sc->g_src.s_addr == INADDR_ANY) {
608 error = EADDRNOTAVAIL;
609 break;
610 }
611 memset(&si, 0, sizeof(si));
612 si.sin_family = AF_INET;
613 si.sin_len = sizeof(struct sockaddr_in);
614 si.sin_addr.s_addr = sc->g_src.s_addr;
615 bcopy(&si, &ifr->ifr_addr, sizeof(ifr->ifr_addr));
616 break;
617 case SIOCGIFPDSTADDR:
618 if (sc->g_dst.s_addr == INADDR_ANY) {
619 error = EADDRNOTAVAIL;
620 break;
621 }
622 memset(&si, 0, sizeof(si));
623 si.sin_family = AF_INET;
624 si.sin_len = sizeof(struct sockaddr_in);
625 si.sin_addr.s_addr = sc->g_dst.s_addr;
626 bcopy(&si, &ifr->ifr_addr, sizeof(ifr->ifr_addr));
627 break;
628 default:
629 error = EINVAL;
630 break;
631 }
632
633 splx(s);
634 return (error);
635}
636
637/*
638 * computes a route to our destination that is not the one
639 * which would be taken by ip_output(), as this one will loop back to
640 * us. If the interface is p2p as a--->b, then a routing entry exists
641 * If we now send a packet to b (e.g. ping b), this will come down here
642 * gets src=a, dst=b tacked on and would from ip_ouput() sent back to
643 * if_gre.
644 * Goal here is to compute a route to b that is less specific than
645 * a-->b. We know that this one exists as in normal operation we have
646 * at least a default route which matches.
647 */
648static int
649gre_compute_route(struct gre_softc *sc)
650{
651 struct route *ro;
652 u_int32_t a, b, c;
653
654 ro = &sc->route;
655
656 memset(ro, 0, sizeof(struct route));
657 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sc->g_dst;
658 ro->ro_dst.sa_family = AF_INET;
659 ro->ro_dst.sa_len = sizeof(ro->ro_dst);
660
661 /*
662 * toggle last bit, so our interface is not found, but a less
663 * specific route. I'd rather like to specify a shorter mask,
664 * but this is not possible. Should work though. XXX
665 * there is a simpler way ...
666 */
667 if ((sc->sc_if.if_flags & IFF_LINK1) == 0) {
668 a = ntohl(sc->g_dst.s_addr);
669 b = a & 0x01;
670 c = a & 0xfffffffe;
671 b = b ^ 0x01;
672 a = b | c;
673 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr.s_addr
674 = htonl(a);
675 }
676
677#ifdef DIAGNOSTIC
678 printf("%s: searching a route to %s", if_name(&sc->sc_if),
679 inet_ntoa(((struct sockaddr_in *)&ro->ro_dst)->sin_addr));
680#endif
681
682 rtalloc(ro);
683
684 /*
685 * check if this returned a route at all and this route is no
686 * recursion to ourself
687 */
688 if (ro->ro_rt == NULL || ro->ro_rt->rt_ifp->if_softc == sc) {
689#ifdef DIAGNOSTIC
690 if (ro->ro_rt == NULL)
691 printf(" - no route found!\n");
692 else
693 printf(" - route loops back to ourself!\n");
694#endif
695 return EADDRNOTAVAIL;
696 }
697
698 /*
699 * now change it back - else ip_output will just drop
700 * the route and search one to this interface ...
701 */
702 if ((sc->sc_if.if_flags & IFF_LINK1) == 0)
703 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr = sc->g_dst;
704
705#ifdef DIAGNOSTIC
706 printf(", choosing %s with gateway %s", if_name(ro->ro_rt->rt_ifp),
707 inet_ntoa(((struct sockaddr_in *)(ro->ro_rt->rt_gateway))->sin_addr));
708 printf("\n");
709#endif
710
711 return 0;
712}
713
714/*
715 * do a checksum of a buffer - much like in_cksum, which operates on
716 * mbufs.
717 */
718u_short
719gre_in_cksum(u_short *p, u_int len)
720{
721 u_int sum = 0;
722 int nwords = len >> 1;
723
724 while (nwords-- != 0)
725 sum += *p++;
726
727 if (len & 1) {
728 union {
729 u_short w;
730 u_char c[2];
731 } u;
732 u.c[0] = *(u_char *)p;
733 u.c[1] = 0;
734 sum += u.w;
735 }
736
737 /* end-around-carry */
738 sum = (sum >> 16) + (sum & 0xffff);
739 sum += (sum >> 16);
740 return (~sum);
741}
742
743static int
744gremodevent(module_t mod, int type, void *data)
745{
746
747 switch (type) {
748 case MOD_LOAD:
749 greattach();
750 break;
751 case MOD_UNLOAD:
752 if_clone_detach(&gre_cloner);
753
754 while (!LIST_EMPTY(&gre_softc_list))
755 gre_clone_destroy(&LIST_FIRST(&gre_softc_list)->sc_if);
756 break;
757 }
758 return 0;
759}
760
761static moduledata_t gre_mod = {
762 "if_gre",
763 gremodevent,
764 0
765};
766
767DECLARE_MODULE(if_gre, gre_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
768MODULE_VERSION(if_gre, 1);
|