xref: /openbsd/sys/net/if_mpw.c (revision 0f9e9ec2)
1 /*	$OpenBSD: if_mpw.c,v 1.66 2024/05/13 01:15:53 jsg Exp $ */
2 
3 /*
4  * Copyright (c) 2015 Rafael Zalamena <rzalamena@openbsd.org>
5  *
6  * Permission to use, copy, modify, and distribute this software for any
7  * purpose with or without fee is hereby granted, provided that the above
8  * copyright notice and this permission notice appear in all copies.
9  *
10  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
11  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
12  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
13  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
14  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
15  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
16  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
17  */
18 
19 #include "bpfilter.h"
20 
21 #include <sys/param.h>
22 #include <sys/systm.h>
23 #include <sys/mbuf.h>
24 #include <sys/socket.h>
25 #include <sys/ioctl.h>
26 #include <sys/errno.h>
27 
28 #include <net/if.h>
29 #include <net/if_dl.h>
30 #include <net/if_types.h>
31 #include <net/route.h>
32 
33 #include <netinet/in.h>
34 
35 #include <netinet/if_ether.h>
36 #include <netmpls/mpls.h>
37 
38 #if NBPFILTER > 0
39 #include <net/bpf.h>
40 #endif /* NBPFILTER */
41 
42 struct mpw_neighbor {
43 	struct shim_hdr		n_rshim;
44 	struct sockaddr_storage	n_nexthop;
45 };
46 
47 struct mpw_softc {
48 	struct arpcom		sc_ac;
49 #define sc_if			sc_ac.ac_if
50 
51 	int			sc_txhprio;
52 	int			sc_rxhprio;
53 	unsigned int		sc_rdomain;
54 	struct ifaddr		sc_ifa;
55 	struct sockaddr_mpls	sc_smpls; /* Local label */
56 
57 	unsigned int		sc_cword;
58 	unsigned int		sc_fword;
59 	uint32_t		sc_flow;
60 	uint32_t		sc_type;
61 	struct mpw_neighbor	*sc_neighbor;
62 
63 	unsigned int		sc_dead;
64 };
65 
66 void	mpwattach(int);
67 int	mpw_clone_create(struct if_clone *, int);
68 int	mpw_clone_destroy(struct ifnet *);
69 int	mpw_ioctl(struct ifnet *, u_long, caddr_t);
70 int	mpw_output(struct ifnet *, struct mbuf *, struct sockaddr *,
71     struct rtentry *);
72 void	mpw_start(struct ifnet *);
73 
74 struct if_clone mpw_cloner =
75     IF_CLONE_INITIALIZER("mpw", mpw_clone_create, mpw_clone_destroy);
76 
77 void
mpwattach(int n)78 mpwattach(int n)
79 {
80 	if_clone_attach(&mpw_cloner);
81 }
82 
83 int
mpw_clone_create(struct if_clone * ifc,int unit)84 mpw_clone_create(struct if_clone *ifc, int unit)
85 {
86 	struct mpw_softc *sc;
87 	struct ifnet *ifp;
88 
89 	sc = malloc(sizeof(*sc), M_DEVBUF, M_WAITOK|M_CANFAIL|M_ZERO);
90 	if (sc == NULL)
91 		return (ENOMEM);
92 
93 	sc->sc_flow = arc4random();
94 	sc->sc_neighbor = NULL;
95 
96 	ifp = &sc->sc_if;
97 	snprintf(ifp->if_xname, sizeof(ifp->if_xname), "%s%d",
98 	    ifc->ifc_name, unit);
99 	ifp->if_softc = sc;
100 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
101 	ifp->if_xflags = IFXF_CLONED;
102 	ifp->if_ioctl = mpw_ioctl;
103 	ifp->if_output = mpw_output;
104 	ifp->if_start = mpw_start;
105 	ifp->if_hardmtu = ETHER_MAX_HARDMTU_LEN;
106 	ether_fakeaddr(ifp);
107 
108 	sc->sc_dead = 0;
109 
110 	if_counters_alloc(ifp);
111 	if_attach(ifp);
112 	ether_ifattach(ifp);
113 
114 	sc->sc_txhprio = 0;
115 	sc->sc_rxhprio = IF_HDRPRIO_PACKET;
116 	sc->sc_rdomain = 0;
117 	refcnt_init_trace(&sc->sc_ifa.ifa_refcnt, DT_REFCNT_IDX_IFADDR);
118 	sc->sc_ifa.ifa_ifp = ifp;
119 	sc->sc_ifa.ifa_addr = sdltosa(ifp->if_sadl);
120 	sc->sc_smpls.smpls_len = sizeof(sc->sc_smpls);
121 	sc->sc_smpls.smpls_family = AF_MPLS;
122 
123 	return (0);
124 }
125 
126 int
mpw_clone_destroy(struct ifnet * ifp)127 mpw_clone_destroy(struct ifnet *ifp)
128 {
129 	struct mpw_softc *sc = ifp->if_softc;
130 
131 	NET_LOCK();
132 	ifp->if_flags &= ~IFF_RUNNING;
133 	sc->sc_dead = 1;
134 
135 	if (sc->sc_smpls.smpls_label) {
136 		rt_ifa_del(&sc->sc_ifa, RTF_MPLS|RTF_LOCAL,
137 		    smplstosa(&sc->sc_smpls), sc->sc_rdomain);
138 	}
139 	NET_UNLOCK();
140 
141 	ifq_barrier(&ifp->if_snd);
142 
143 	ether_ifdetach(ifp);
144 	if_detach(ifp);
145 	if (refcnt_rele(&sc->sc_ifa.ifa_refcnt) == 0) {
146 		panic("%s: ifa refcnt has %u refs", __func__,
147 		    sc->sc_ifa.ifa_refcnt.r_refs);
148 	}
149 	free(sc->sc_neighbor, M_DEVBUF, sizeof(*sc->sc_neighbor));
150 	free(sc, M_DEVBUF, sizeof(*sc));
151 
152 	return (0);
153 }
154 
155 int
mpw_set_route(struct mpw_softc * sc,uint32_t label,unsigned int rdomain)156 mpw_set_route(struct mpw_softc *sc, uint32_t label, unsigned int rdomain)
157 {
158 	int error;
159 
160 	if (sc->sc_dead)
161 		return (ENXIO);
162 
163 	if (sc->sc_smpls.smpls_label) {
164 		rt_ifa_del(&sc->sc_ifa, RTF_MPLS|RTF_LOCAL,
165 		    smplstosa(&sc->sc_smpls), sc->sc_rdomain);
166 	}
167 
168 	sc->sc_smpls.smpls_label = label;
169 	sc->sc_rdomain = rdomain;
170 
171 	/* only install with a label or mpw_clone_destroy() will ignore it */
172 	if (sc->sc_smpls.smpls_label == MPLS_LABEL2SHIM(0))
173 		return 0;
174 
175 	error = rt_ifa_add(&sc->sc_ifa, RTF_MPLS|RTF_LOCAL,
176 	    smplstosa(&sc->sc_smpls), sc->sc_rdomain);
177 	if (error != 0)
178 		sc->sc_smpls.smpls_label = 0;
179 
180 	return (error);
181 }
182 
183 static int
mpw_set_neighbor(struct mpw_softc * sc,const struct if_laddrreq * req)184 mpw_set_neighbor(struct mpw_softc *sc, const struct if_laddrreq *req)
185 {
186 	struct mpw_neighbor *n, *o;
187 	const struct sockaddr_storage *ss;
188 	const struct sockaddr_mpls *smpls;
189 	uint32_t label;
190 
191 	smpls = (const struct sockaddr_mpls *)&req->dstaddr;
192 
193 	if (smpls->smpls_family != AF_MPLS)
194 		return (EINVAL);
195 	label = smpls->smpls_label;
196 	if (label > MPLS_LABEL_MAX || label <= MPLS_LABEL_RESERVED_MAX)
197 		return (EINVAL);
198 
199 	ss = &req->addr;
200 	switch (ss->ss_family) {
201 	case AF_INET: {
202 		const struct sockaddr_in *sin =
203 		    (const struct sockaddr_in *)ss;
204 
205 		if (in_nullhost(sin->sin_addr) ||
206 		    IN_MULTICAST(sin->sin_addr.s_addr))
207 			return (EINVAL);
208 
209 		break;
210 	}
211 #ifdef INET6
212 	case AF_INET6: {
213 		const struct sockaddr_in6 *sin6 =
214 		    (const struct sockaddr_in6 *)ss;
215 
216 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr) ||
217 		    IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
218 			return (EINVAL);
219 
220 		/* check scope */
221 
222 		break;
223 	}
224 #endif
225 	default:
226 		return (EAFNOSUPPORT);
227 	}
228 
229 	if (sc->sc_dead)
230 		return (ENXIO);
231 
232 	n = malloc(sizeof(*n), M_DEVBUF, M_WAITOK|M_CANFAIL|M_ZERO);
233 	if (n == NULL)
234 		return (ENOMEM);
235 
236 	n->n_rshim.shim_label = MPLS_LABEL2SHIM(label);
237 	n->n_nexthop = *ss;
238 
239 	o = sc->sc_neighbor;
240 	sc->sc_neighbor = n;
241 
242 	NET_UNLOCK();
243 	ifq_barrier(&sc->sc_if.if_snd);
244 	NET_LOCK();
245 
246 	free(o, M_DEVBUF, sizeof(*o));
247 
248 	return (0);
249 }
250 
251 static int
mpw_get_neighbor(struct mpw_softc * sc,struct if_laddrreq * req)252 mpw_get_neighbor(struct mpw_softc *sc, struct if_laddrreq *req)
253 {
254 	struct sockaddr_mpls *smpls = (struct sockaddr_mpls *)&req->dstaddr;
255 	struct mpw_neighbor *n = sc->sc_neighbor;
256 
257 	if (n == NULL)
258 		return (EADDRNOTAVAIL);
259 
260 	smpls->smpls_len = sizeof(*smpls);
261 	smpls->smpls_family = AF_MPLS;
262 	smpls->smpls_label = MPLS_SHIM2LABEL(n->n_rshim.shim_label);
263 
264 	req->addr = n->n_nexthop;
265 
266 	return (0);
267 }
268 
269 static int
mpw_del_neighbor(struct mpw_softc * sc)270 mpw_del_neighbor(struct mpw_softc *sc)
271 {
272 	struct mpw_neighbor *o;
273 
274 	if (sc->sc_dead)
275 		return (ENXIO);
276 
277 	o = sc->sc_neighbor;
278 	sc->sc_neighbor = NULL;
279 
280 	NET_UNLOCK();
281 	ifq_barrier(&sc->sc_if.if_snd);
282 	NET_LOCK();
283 
284 	free(o, M_DEVBUF, sizeof(*o));
285 
286 	return (0);
287 }
288 
289 static int
mpw_set_label(struct mpw_softc * sc,const struct shim_hdr * label)290 mpw_set_label(struct mpw_softc *sc, const struct shim_hdr *label)
291 {
292 	uint32_t shim;
293 
294 	if (label->shim_label > MPLS_LABEL_MAX ||
295 	    label->shim_label <= MPLS_LABEL_RESERVED_MAX)
296 		return (EINVAL);
297 
298 	shim = MPLS_LABEL2SHIM(label->shim_label);
299 	if (sc->sc_smpls.smpls_label == shim)
300 		return (0);
301 
302 	return (mpw_set_route(sc, shim, sc->sc_rdomain));
303 }
304 
305 static int
mpw_get_label(struct mpw_softc * sc,struct ifreq * ifr)306 mpw_get_label(struct mpw_softc *sc, struct ifreq *ifr)
307 {
308 	struct shim_hdr label;
309 
310 	label.shim_label = MPLS_SHIM2LABEL(sc->sc_smpls.smpls_label);
311 
312 	if (label.shim_label == 0)
313 		return (EADDRNOTAVAIL);
314 
315 	return (copyout(&label, ifr->ifr_data, sizeof(label)));
316 }
317 
318 static int
mpw_del_label(struct mpw_softc * sc)319 mpw_del_label(struct mpw_softc *sc)
320 {
321 	if (sc->sc_dead)
322 		return (ENXIO);
323 
324 	if (sc->sc_smpls.smpls_label != MPLS_LABEL2SHIM(0)) {
325 		rt_ifa_del(&sc->sc_ifa, RTF_MPLS | RTF_LOCAL,
326 		    smplstosa(&sc->sc_smpls), sc->sc_rdomain);
327 	}
328 
329 	sc->sc_smpls.smpls_label = MPLS_LABEL2SHIM(0);
330 
331 	return (0);
332 }
333 
334 static int
mpw_set_config(struct mpw_softc * sc,const struct ifreq * ifr)335 mpw_set_config(struct mpw_softc *sc, const struct ifreq *ifr)
336 {
337 	struct ifmpwreq imr;
338 	struct if_laddrreq req;
339 	struct sockaddr_mpls *smpls;
340 	struct sockaddr_in *sin;
341 	int error;
342 
343 	error = copyin(ifr->ifr_data, &imr, sizeof(imr));
344 	if (error != 0)
345 		return (error);
346 
347 	/* Teardown all configuration if got no nexthop */
348 	sin = (struct sockaddr_in *)&imr.imr_nexthop;
349 	if (sin->sin_addr.s_addr == 0) {
350 		mpw_del_label(sc);
351 		mpw_del_neighbor(sc);
352 		sc->sc_cword = 0;
353 		sc->sc_type = 0;
354 		return (0);
355 	}
356 
357 	error = mpw_set_label(sc, &imr.imr_lshim);
358 	if (error != 0)
359 		return (error);
360 
361 	smpls = (struct sockaddr_mpls *)&req.dstaddr;
362 	smpls->smpls_family = AF_MPLS;
363 	smpls->smpls_label = imr.imr_rshim.shim_label;
364 	req.addr = imr.imr_nexthop;
365 
366 	error = mpw_set_neighbor(sc, &req);
367 	if (error != 0)
368 		return (error);
369 
370 	sc->sc_cword = ISSET(imr.imr_flags, IMR_FLAG_CONTROLWORD);
371 	sc->sc_type = imr.imr_type;
372 
373 	return (0);
374 }
375 
376 static int
mpw_get_config(struct mpw_softc * sc,const struct ifreq * ifr)377 mpw_get_config(struct mpw_softc *sc, const struct ifreq *ifr)
378 {
379 	struct ifmpwreq imr;
380 
381 	memset(&imr, 0, sizeof(imr));
382 	imr.imr_flags = sc->sc_cword ? IMR_FLAG_CONTROLWORD : 0;
383 	imr.imr_type = sc->sc_type;
384 
385 	imr.imr_lshim.shim_label = MPLS_SHIM2LABEL(sc->sc_smpls.smpls_label);
386 	if (sc->sc_neighbor) {
387 		imr.imr_rshim.shim_label =
388 		    MPLS_SHIM2LABEL(sc->sc_neighbor->n_rshim.shim_label);
389 		imr.imr_nexthop = sc->sc_neighbor->n_nexthop;
390 	}
391 
392 	return (copyout(&imr, ifr->ifr_data, sizeof(imr)));
393 }
394 
395 int
mpw_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data)396 mpw_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
397 {
398 	struct ifreq *ifr = (struct ifreq *) data;
399 	struct mpw_softc *sc = ifp->if_softc;
400 	struct shim_hdr shim;
401 	int error = 0;
402 
403 	switch (cmd) {
404 	case SIOCSIFFLAGS:
405 		if ((ifp->if_flags & IFF_UP))
406 			ifp->if_flags |= IFF_RUNNING;
407 		else
408 			ifp->if_flags &= ~IFF_RUNNING;
409 		break;
410 
411 	case SIOCGPWE3:
412 		ifr->ifr_pwe3 = IF_PWE3_ETHERNET;
413 		break;
414 	case SIOCSPWE3CTRLWORD:
415 		sc->sc_cword = ifr->ifr_pwe3 ? 1 : 0;
416 		break;
417 	case SIOCGPWE3CTRLWORD:
418 		ifr->ifr_pwe3 = sc->sc_cword;
419 		break;
420 	case SIOCSPWE3FAT:
421 		sc->sc_fword = ifr->ifr_pwe3 ? 1 : 0;
422 		break;
423 	case SIOCGPWE3FAT:
424 		ifr->ifr_pwe3 = sc->sc_fword;
425 		break;
426 
427 	case SIOCSPWE3NEIGHBOR:
428 		error = mpw_set_neighbor(sc, (struct if_laddrreq *)data);
429 		break;
430 	case SIOCGPWE3NEIGHBOR:
431 		error = mpw_get_neighbor(sc, (struct if_laddrreq *)data);
432 		break;
433 	case SIOCDPWE3NEIGHBOR:
434 		error = mpw_del_neighbor(sc);
435 		break;
436 
437 	case SIOCGETLABEL:
438 		error = mpw_get_label(sc, ifr);
439 		break;
440 	case SIOCSETLABEL:
441 		error = copyin(ifr->ifr_data, &shim, sizeof(shim));
442 		if (error != 0)
443 			break;
444 		error = mpw_set_label(sc, &shim);
445 		break;
446 	case SIOCDELLABEL:
447 		error = mpw_del_label(sc);
448 		break;
449 
450 	case SIOCSETMPWCFG:
451 		error = mpw_set_config(sc, ifr);
452 		break;
453 
454 	case SIOCGETMPWCFG:
455 		error = mpw_get_config(sc, ifr);
456 		break;
457 
458 	case SIOCSLIFPHYRTABLE:
459 		if (ifr->ifr_rdomainid < 0 ||
460 		    ifr->ifr_rdomainid > RT_TABLEID_MAX ||
461 		    !rtable_exists(ifr->ifr_rdomainid) ||
462 		    ifr->ifr_rdomainid != rtable_l2(ifr->ifr_rdomainid)) {
463 			error = EINVAL;
464 			break;
465 		}
466 		if (sc->sc_rdomain != ifr->ifr_rdomainid) {
467 			error = mpw_set_route(sc, sc->sc_smpls.smpls_label,
468 			    ifr->ifr_rdomainid);
469 		}
470 		break;
471 	case SIOCGLIFPHYRTABLE:
472 		ifr->ifr_rdomainid = sc->sc_rdomain;
473 		break;
474 
475 	case SIOCSTXHPRIO:
476 		error = if_txhprio_l2_check(ifr->ifr_hdrprio);
477 		if (error != 0)
478 			break;
479 
480 		sc->sc_txhprio = ifr->ifr_hdrprio;
481 		break;
482 	case SIOCGTXHPRIO:
483 		ifr->ifr_hdrprio = sc->sc_txhprio;
484 		break;
485 
486 	case SIOCSRXHPRIO:
487 		error = if_rxhprio_l2_check(ifr->ifr_hdrprio);
488 		if (error != 0)
489 			break;
490 
491 		sc->sc_rxhprio = ifr->ifr_hdrprio;
492 		break;
493 	case SIOCGRXHPRIO:
494 		ifr->ifr_hdrprio = sc->sc_rxhprio;
495 		break;
496 
497 	case SIOCADDMULTI:
498 	case SIOCDELMULTI:
499 		break;
500 
501 	default:
502 		error = ether_ioctl(ifp, &sc->sc_ac, cmd, data);
503 		break;
504 	}
505 
506 	return (error);
507 }
508 
509 static void
mpw_input(struct mpw_softc * sc,struct mbuf * m)510 mpw_input(struct mpw_softc *sc, struct mbuf *m)
511 {
512 	struct ifnet *ifp = &sc->sc_if;
513 	struct shim_hdr *shim;
514 	struct mbuf *n;
515 	uint32_t exp;
516 	int rxprio;
517 	int off;
518 
519 	if (!ISSET(ifp->if_flags, IFF_RUNNING))
520 		goto drop;
521 
522 	shim = mtod(m, struct shim_hdr *);
523 	exp = ntohl(shim->shim_label & MPLS_EXP_MASK) >> MPLS_EXP_OFFSET;
524 	if (sc->sc_fword) {
525 		uint32_t flow;
526 
527 		if (MPLS_BOS_ISSET(shim->shim_label))
528 			goto drop;
529 		m_adj(m, sizeof(*shim));
530 
531 		if (m->m_len < sizeof(*shim)) {
532 			m = m_pullup(m, sizeof(*shim));
533 			if (m == NULL)
534 				return;
535 		}
536 		shim = mtod(m, struct shim_hdr *);
537 
538 		if (!MPLS_BOS_ISSET(shim->shim_label))
539 			goto drop;
540 
541 		flow = MPLS_SHIM2LABEL(shim->shim_label);
542 		flow ^= sc->sc_flow;
543 		SET(m->m_pkthdr.csum_flags, M_FLOWID);
544 		m->m_pkthdr.ph_flowid = flow;
545 	} else {
546 		if (!MPLS_BOS_ISSET(shim->shim_label))
547 			goto drop;
548 	}
549 	m_adj(m, sizeof(*shim));
550 
551 	if (sc->sc_cword) {
552 		if (m->m_len < sizeof(*shim)) {
553 			m = m_pullup(m, sizeof(*shim));
554 			if (m == NULL)
555 				return;
556 		}
557 		shim = mtod(m, struct shim_hdr *);
558 
559 		/*
560 		 * The first 4 bits identifies that this packet is a
561 		 * control word. If the control word is configured and
562 		 * we received an IP datagram we shall drop it.
563 		 */
564 		if (shim->shim_label & CW_ZERO_MASK) {
565 			ifp->if_ierrors++;
566 			goto drop;
567 		}
568 
569 		/* We don't support fragmentation just yet. */
570 		if (shim->shim_label & CW_FRAG_MASK) {
571 			ifp->if_ierrors++;
572 			goto drop;
573 		}
574 
575 		m_adj(m, MPLS_HDRLEN);
576 	}
577 
578 	if (m->m_len < sizeof(struct ether_header)) {
579 		m = m_pullup(m, sizeof(struct ether_header));
580 		if (m == NULL)
581 			return;
582 	}
583 
584 	n = m_getptr(m, sizeof(struct ether_header), &off);
585 	if (n == NULL) {
586 		ifp->if_ierrors++;
587 		goto drop;
588 	}
589 	if (!ALIGNED_POINTER(mtod(n, caddr_t) + off, uint32_t)) {
590 		n = m_dup_pkt(m, ETHER_ALIGN, M_NOWAIT);
591 		/* Dispose of the original mbuf chain */
592 		m_freem(m);
593 		if (n == NULL)
594 			return;
595 		m = n;
596 	}
597 
598 	rxprio = sc->sc_rxhprio;
599 	switch (rxprio) {
600 	case IF_HDRPRIO_PACKET:
601 		/* nop */
602 		break;
603 	case IF_HDRPRIO_OUTER:
604 		m->m_pkthdr.pf.prio = exp;
605 		break;
606 	default:
607 		m->m_pkthdr.pf.prio = rxprio;
608 		break;
609 	}
610 
611 	m->m_pkthdr.ph_ifidx = ifp->if_index;
612 	m->m_pkthdr.ph_rtableid = ifp->if_rdomain;
613 
614 	/* packet has not been processed by PF yet. */
615 	KASSERT(m->m_pkthdr.pf.statekey == NULL);
616 
617 	if_vinput(ifp, m);
618 	return;
619 drop:
620 	m_freem(m);
621 }
622 
623 int
mpw_output(struct ifnet * ifp,struct mbuf * m,struct sockaddr * dst,struct rtentry * rt)624 mpw_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
625     struct rtentry *rt)
626 {
627 	struct mpw_softc *sc = ifp->if_softc;
628 
629 	if (dst->sa_family == AF_LINK &&
630 	    rt != NULL && ISSET(rt->rt_flags, RTF_LOCAL)) {
631 		mpw_input(sc, m);
632 		return (0);
633 	}
634 
635 	return (ether_output(ifp, m, dst, rt));
636 }
637 
638 void
mpw_start(struct ifnet * ifp)639 mpw_start(struct ifnet *ifp)
640 {
641 	struct mpw_softc *sc = ifp->if_softc;
642 	struct rtentry *rt;
643 	struct ifnet *ifp0;
644 	struct mbuf *m, *m0;
645 	struct shim_hdr *shim;
646 	struct mpw_neighbor *n;
647 	struct sockaddr_mpls smpls = {
648 		.smpls_len = sizeof(smpls),
649 		.smpls_family = AF_MPLS,
650 	};
651 	int txprio = sc->sc_txhprio;
652 	uint8_t prio;
653 	uint32_t exp, bos;
654 
655 	n = sc->sc_neighbor;
656 	if (!ISSET(ifp->if_flags, IFF_RUNNING) ||
657 	    n == NULL) {
658 		ifq_purge(&ifp->if_snd);
659 		return;
660 	}
661 
662 	rt = rtalloc(sstosa(&n->n_nexthop), RT_RESOLVE, sc->sc_rdomain);
663 	if (!rtisvalid(rt)) {
664 		ifq_purge(&ifp->if_snd);
665 		goto rtfree;
666 	}
667 
668 	ifp0 = if_get(rt->rt_ifidx);
669 	if (ifp0 == NULL) {
670 		ifq_purge(&ifp->if_snd);
671 		goto rtfree;
672 	}
673 
674 	while ((m = ifq_dequeue(&ifp->if_snd)) != NULL) {
675 #if NBPFILTER > 0
676 		if (sc->sc_if.if_bpf)
677 			bpf_mtap(sc->sc_if.if_bpf, m, BPF_DIRECTION_OUT);
678 #endif /* NBPFILTER */
679 
680 		m0 = m_get(M_DONTWAIT, m->m_type);
681 		if (m0 == NULL) {
682 			m_freem(m);
683 			continue;
684 		}
685 
686 		M_MOVE_PKTHDR(m0, m);
687 		m0->m_next = m;
688 		m_align(m0, 0);
689 		m0->m_len = 0;
690 
691 		if (sc->sc_cword) {
692 			m0 = m_prepend(m0, sizeof(*shim), M_NOWAIT);
693 			if (m0 == NULL)
694 				continue;
695 
696 			shim = mtod(m0, struct shim_hdr *);
697 			memset(shim, 0, sizeof(*shim));
698 		}
699 
700 		switch (txprio) {
701 		case IF_HDRPRIO_PACKET:
702 			prio = m->m_pkthdr.pf.prio;
703 			break;
704 		default:
705 			prio = txprio;
706 			break;
707 		}
708 		exp = htonl(prio << MPLS_EXP_OFFSET);
709 
710 		bos = MPLS_BOS_MASK;
711 		if (sc->sc_fword) {
712 			uint32_t flow = sc->sc_flow;
713 
714 			m0 = m_prepend(m0, sizeof(*shim), M_NOWAIT);
715 			if (m0 == NULL)
716 				continue;
717 
718 			if (ISSET(m->m_pkthdr.csum_flags, M_FLOWID))
719 				flow ^= m->m_pkthdr.ph_flowid;
720 
721 			shim = mtod(m0, struct shim_hdr *);
722 			shim->shim_label = htonl(1) & MPLS_TTL_MASK;
723 			shim->shim_label |= MPLS_LABEL2SHIM(flow) | exp | bos;
724 
725 			bos = 0;
726 		}
727 
728 		m0 = m_prepend(m0, sizeof(*shim), M_NOWAIT);
729 		if (m0 == NULL)
730 			continue;
731 
732 		shim = mtod(m0, struct shim_hdr *);
733 		shim->shim_label = htonl(mpls_defttl) & MPLS_TTL_MASK;
734 		shim->shim_label |= n->n_rshim.shim_label | exp | bos;
735 
736 		m0->m_pkthdr.ph_rtableid = sc->sc_rdomain;
737 		CLR(m0->m_flags, M_BCAST|M_MCAST);
738 
739 		mpls_output(ifp0, m0, (struct sockaddr *)&smpls, rt);
740 	}
741 
742 	if_put(ifp0);
743 rtfree:
744 	rtfree(rt);
745 }
746