Home | History | Annotate | Line # | Download | only in net
if_gif.c revision 1.130
      1 /*	$NetBSD: if_gif.c,v 1.130 2017/09/21 09:46:14 knakahara Exp $	*/
      2 /*	$KAME: if_gif.c,v 1.76 2001/08/20 02:01:02 kjc Exp $	*/
      3 
      4 /*
      5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
      6  * All rights reserved.
      7  *
      8  * Redistribution and use in source and binary forms, with or without
      9  * modification, are permitted provided that the following conditions
     10  * are met:
     11  * 1. Redistributions of source code must retain the above copyright
     12  *    notice, this list of conditions and the following disclaimer.
     13  * 2. Redistributions in binary form must reproduce the above copyright
     14  *    notice, this list of conditions and the following disclaimer in the
     15  *    documentation and/or other materials provided with the distribution.
     16  * 3. Neither the name of the project nor the names of its contributors
     17  *    may be used to endorse or promote products derived from this software
     18  *    without specific prior written permission.
     19  *
     20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
     21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
     24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     30  * SUCH DAMAGE.
     31  */
     32 
     33 #include <sys/cdefs.h>
     34 __KERNEL_RCSID(0, "$NetBSD: if_gif.c,v 1.130 2017/09/21 09:46:14 knakahara Exp $");
     35 
     36 #ifdef _KERNEL_OPT
     37 #include "opt_inet.h"
     38 #include "opt_net_mpsafe.h"
     39 #endif
     40 
     41 #include <sys/param.h>
     42 #include <sys/systm.h>
     43 #include <sys/kernel.h>
     44 #include <sys/mbuf.h>
     45 #include <sys/socket.h>
     46 #include <sys/sockio.h>
     47 #include <sys/errno.h>
     48 #include <sys/ioctl.h>
     49 #include <sys/time.h>
     50 #include <sys/socketvar.h>
     51 #include <sys/syslog.h>
     52 #include <sys/proc.h>
     53 #include <sys/cpu.h>
     54 #include <sys/intr.h>
     55 #include <sys/kmem.h>
     56 #include <sys/sysctl.h>
     57 #include <sys/xcall.h>
     58 #include <sys/device.h>
     59 #include <sys/module.h>
     60 
     61 #include <net/if.h>
     62 #include <net/if_types.h>
     63 #include <net/netisr.h>
     64 #include <net/route.h>
     65 #include <net/bpf.h>
     66 
     67 #include <netinet/in.h>
     68 #include <netinet/in_systm.h>
     69 #include <netinet/ip.h>
     70 #ifdef	INET
     71 #include <netinet/in_var.h>
     72 #endif	/* INET */
     73 #include <netinet/in_gif.h>
     74 
     75 #ifdef INET6
     76 #ifndef INET
     77 #include <netinet/in.h>
     78 #endif
     79 #include <netinet6/in6_var.h>
     80 #include <netinet/ip6.h>
     81 #include <netinet6/ip6_var.h>
     82 #include <netinet6/in6_gif.h>
     83 #endif /* INET6 */
     84 
     85 #include <netinet/ip_encap.h>
     86 #include <net/if_gif.h>
     87 
     88 #include <net/net_osdep.h>
     89 
     90 #include "ioconf.h"
     91 
     92 #ifdef NET_MPSAFE
     93 #define GIF_MPSAFE	1
     94 #endif
     95 
     96 /*
     97  * gif global variable definitions
     98  */
     99 LIST_HEAD(gif_sclist, gif_softc);
    100 static struct {
    101 	struct gif_sclist list;
    102 	kmutex_t lock;
    103 } gif_softcs __cacheline_aligned;
    104 
    105 static void	gif_ro_init_pc(void *, void *, struct cpu_info *);
    106 static void	gif_ro_fini_pc(void *, void *, struct cpu_info *);
    107 
    108 static void	gifattach0(struct gif_softc *);
    109 static int	gif_output(struct ifnet *, struct mbuf *,
    110 			   const struct sockaddr *, const struct rtentry *);
    111 static void	gif_start(struct ifnet *);
    112 static int	gif_transmit(struct ifnet *, struct mbuf *);
    113 static int	gif_ioctl(struct ifnet *, u_long, void *);
    114 static int	gif_set_tunnel(struct ifnet *, struct sockaddr *,
    115 			       struct sockaddr *);
    116 static void	gif_delete_tunnel(struct ifnet *);
    117 
    118 static int	gif_clone_create(struct if_clone *, int);
    119 static int	gif_clone_destroy(struct ifnet *);
    120 static int	gif_check_nesting(struct ifnet *, struct mbuf *);
    121 
    122 static int	gif_encap_attach(struct gif_softc *);
    123 static int	gif_encap_detach(struct gif_softc *);
    124 static void	gif_encap_pause(struct gif_softc *);
    125 
    126 static struct if_clone gif_cloner =
    127     IF_CLONE_INITIALIZER("gif", gif_clone_create, gif_clone_destroy);
    128 
    129 #ifndef MAX_GIF_NEST
    130 /*
    131  * This macro controls the upper limitation on nesting of gif tunnels.
    132  * Since, setting a large value to this macro with a careless configuration
    133  * may introduce system crash, we don't allow any nestings by default.
    134  * If you need to configure nested gif tunnels, you can define this macro
    135  * in your kernel configuration file.  However, if you do so, please be
    136  * careful to configure the tunnels so that it won't make a loop.
    137  */
    138 #define MAX_GIF_NEST 1
    139 #endif
    140 static int max_gif_nesting = MAX_GIF_NEST;
    141 
    142 static struct sysctllog *gif_sysctl;
    143 
    144 static void
    145 gif_sysctl_setup(void)
    146 {
    147 	gif_sysctl = NULL;
    148 
    149 #ifdef INET
    150 	/*
    151 	 * Previously create "net.inet.ip" entry to avoid sysctl_createv error.
    152 	 */
    153 	sysctl_createv(NULL, 0, NULL, NULL,
    154 		       CTLFLAG_PERMANENT,
    155 		       CTLTYPE_NODE, "inet",
    156 		       SYSCTL_DESCR("PF_INET related settings"),
    157 		       NULL, 0, NULL, 0,
    158 		       CTL_NET, PF_INET, CTL_EOL);
    159 	sysctl_createv(NULL, 0, NULL, NULL,
    160 		       CTLFLAG_PERMANENT,
    161 		       CTLTYPE_NODE, "ip",
    162 		       SYSCTL_DESCR("IPv4 related settings"),
    163 		       NULL, 0, NULL, 0,
    164 		       CTL_NET, PF_INET, IPPROTO_IP, CTL_EOL);
    165 
    166 	sysctl_createv(&gif_sysctl, 0, NULL, NULL,
    167 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    168 		       CTLTYPE_INT, "gifttl",
    169 		       SYSCTL_DESCR("Default TTL for a gif tunnel datagram"),
    170 		       NULL, 0, &ip_gif_ttl, 0,
    171 		       CTL_NET, PF_INET, IPPROTO_IP,
    172 		       IPCTL_GIF_TTL, CTL_EOL);
    173 #endif
    174 #ifdef INET6
    175 	/*
    176 	 * Previously create "net.inet6.ip6" entry to avoid sysctl_createv error.
    177 	 */
    178 	sysctl_createv(NULL, 0, NULL, NULL,
    179 		       CTLFLAG_PERMANENT,
    180 		       CTLTYPE_NODE, "inet6",
    181 		       SYSCTL_DESCR("PF_INET6 related settings"),
    182 		       NULL, 0, NULL, 0,
    183 		       CTL_NET, PF_INET6, CTL_EOL);
    184 	sysctl_createv(NULL, 0, NULL, NULL,
    185 		       CTLFLAG_PERMANENT,
    186 		       CTLTYPE_NODE, "ip6",
    187 		       SYSCTL_DESCR("IPv6 related settings"),
    188 		       NULL, 0, NULL, 0,
    189 		       CTL_NET, PF_INET6, IPPROTO_IPV6, CTL_EOL);
    190 
    191 	sysctl_createv(&gif_sysctl, 0, NULL, NULL,
    192 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
    193 		       CTLTYPE_INT, "gifhlim",
    194 		       SYSCTL_DESCR("Default hop limit for a gif tunnel datagram"),
    195 		       NULL, 0, &ip6_gif_hlim, 0,
    196 		       CTL_NET, PF_INET6, IPPROTO_IPV6,
    197 		       IPV6CTL_GIF_HLIM, CTL_EOL);
    198 #endif
    199 }
    200 
    201 /* ARGSUSED */
    202 void
    203 gifattach(int count)
    204 {
    205 	/*
    206 	 * Nothing to do here, initialization is handled by the
    207 	 * module initialization code in gifinit() below).
    208 	 */
    209 }
    210 
    211 static void
    212 gifinit(void)
    213 {
    214 
    215 	mutex_init(&gif_softcs.lock, MUTEX_DEFAULT, IPL_NONE);
    216 	LIST_INIT(&gif_softcs.list);
    217 	if_clone_attach(&gif_cloner);
    218 
    219 	gif_sysctl_setup();
    220 }
    221 
    222 static int
    223 gifdetach(void)
    224 {
    225 	int error = 0;
    226 
    227 	mutex_enter(&gif_softcs.lock);
    228 	if (!LIST_EMPTY(&gif_softcs.list)) {
    229 		mutex_exit(&gif_softcs.lock);
    230 		error = EBUSY;
    231 	}
    232 
    233 	if (error == 0) {
    234 		if_clone_detach(&gif_cloner);
    235 		sysctl_teardown(&gif_sysctl);
    236 	}
    237 
    238 	return error;
    239 }
    240 
    241 static int
    242 gif_clone_create(struct if_clone *ifc, int unit)
    243 {
    244 	struct gif_softc *sc;
    245 
    246 	sc = kmem_zalloc(sizeof(struct gif_softc), KM_SLEEP);
    247 
    248 	if_initname(&sc->gif_if, ifc->ifc_name, unit);
    249 
    250 	gifattach0(sc);
    251 
    252 	sc->gif_ro_percpu = percpu_alloc(sizeof(struct gif_ro));
    253 	percpu_foreach(sc->gif_ro_percpu, gif_ro_init_pc, NULL);
    254 
    255 	mutex_enter(&gif_softcs.lock);
    256 	LIST_INSERT_HEAD(&gif_softcs.list, sc, gif_list);
    257 	mutex_exit(&gif_softcs.lock);
    258 	return (0);
    259 }
    260 
    261 static void
    262 gifattach0(struct gif_softc *sc)
    263 {
    264 
    265 	sc->encap_cookie4 = sc->encap_cookie6 = NULL;
    266 
    267 	sc->gif_if.if_addrlen = 0;
    268 	sc->gif_if.if_mtu    = GIF_MTU;
    269 	sc->gif_if.if_flags  = IFF_POINTOPOINT | IFF_MULTICAST;
    270 	sc->gif_if.if_extflags  = IFEF_NO_LINK_STATE_CHANGE;
    271 #ifdef GIF_MPSAFE
    272 	sc->gif_if.if_extflags  |= IFEF_OUTPUT_MPSAFE;
    273 #endif
    274 	sc->gif_if.if_ioctl  = gif_ioctl;
    275 	sc->gif_if.if_output = gif_output;
    276 	sc->gif_if.if_start = gif_start;
    277 	sc->gif_if.if_transmit = gif_transmit;
    278 	sc->gif_if.if_type   = IFT_GIF;
    279 	sc->gif_if.if_dlt    = DLT_NULL;
    280 	sc->gif_if.if_softc  = sc;
    281 	IFQ_SET_READY(&sc->gif_if.if_snd);
    282 	if_initialize(&sc->gif_if);
    283 	if_register(&sc->gif_if);
    284 	if_alloc_sadl(&sc->gif_if);
    285 	bpf_attach(&sc->gif_if, DLT_NULL, sizeof(u_int));
    286 }
    287 
    288 static void
    289 gif_ro_init_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
    290 {
    291 	struct gif_ro *gro = p;
    292 
    293 	mutex_init(&gro->gr_lock, MUTEX_DEFAULT, IPL_NONE);
    294 }
    295 
    296 static void
    297 gif_ro_fini_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
    298 {
    299 	struct gif_ro *gro = p;
    300 
    301 	rtcache_free(&gro->gr_ro);
    302 
    303 	mutex_destroy(&gro->gr_lock);
    304 }
    305 
    306 void
    307 gif_rtcache_free_pc(void *p, void *arg __unused, struct cpu_info *ci __unused)
    308 {
    309 	struct gif_ro *gro = p;
    310 
    311 	rtcache_free(&gro->gr_ro);
    312 }
    313 
    314 static int
    315 gif_clone_destroy(struct ifnet *ifp)
    316 {
    317 	struct gif_softc *sc = (void *) ifp;
    318 
    319 	LIST_REMOVE(sc, gif_list);
    320 
    321 	gif_delete_tunnel(&sc->gif_if);
    322 	bpf_detach(ifp);
    323 	if_detach(ifp);
    324 
    325 	percpu_foreach(sc->gif_ro_percpu, gif_ro_fini_pc, NULL);
    326 	percpu_free(sc->gif_ro_percpu, sizeof(struct gif_ro));
    327 
    328 	kmem_free(sc, sizeof(struct gif_softc));
    329 
    330 	return (0);
    331 }
    332 
    333 #ifdef GIF_ENCAPCHECK
    334 int
    335 gif_encapcheck(struct mbuf *m, int off, int proto, void *arg)
    336 {
    337 	struct ip ip;
    338 	struct gif_softc *sc;
    339 
    340 	sc = arg;
    341 	if (sc == NULL)
    342 		return 0;
    343 
    344 	if ((sc->gif_if.if_flags & (IFF_UP|IFF_RUNNING))
    345 	    != (IFF_UP|IFF_RUNNING))
    346 		return 0;
    347 
    348 	/* no physical address */
    349 	if (!sc->gif_psrc || !sc->gif_pdst)
    350 		return 0;
    351 
    352 	switch (proto) {
    353 #ifdef INET
    354 	case IPPROTO_IPV4:
    355 		break;
    356 #endif
    357 #ifdef INET6
    358 	case IPPROTO_IPV6:
    359 		break;
    360 #endif
    361 	default:
    362 		return 0;
    363 	}
    364 
    365 	/* Bail on short packets */
    366 	KASSERT(m->m_flags & M_PKTHDR);
    367 	if (m->m_pkthdr.len < sizeof(ip))
    368 		return 0;
    369 
    370 	m_copydata(m, 0, sizeof(ip), &ip);
    371 
    372 	switch (ip.ip_v) {
    373 #ifdef INET
    374 	case 4:
    375 		if (sc->gif_psrc->sa_family != AF_INET ||
    376 		    sc->gif_pdst->sa_family != AF_INET)
    377 			return 0;
    378 		return gif_encapcheck4(m, off, proto, arg);
    379 #endif
    380 #ifdef INET6
    381 	case 6:
    382 		if (m->m_pkthdr.len < sizeof(struct ip6_hdr))
    383 			return 0;
    384 		if (sc->gif_psrc->sa_family != AF_INET6 ||
    385 		    sc->gif_pdst->sa_family != AF_INET6)
    386 			return 0;
    387 		return gif_encapcheck6(m, off, proto, arg);
    388 #endif
    389 	default:
    390 		return 0;
    391 	}
    392 }
    393 #endif
    394 
    395 /*
    396  * gif may cause infinite recursion calls when misconfigured.
    397  * We'll prevent this by introducing upper limit.
    398  */
    399 static int
    400 gif_check_nesting(struct ifnet *ifp, struct mbuf *m)
    401 {
    402 	struct m_tag *mtag;
    403 	int *count;
    404 
    405 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO, NULL);
    406 	if (mtag != NULL) {
    407 		count = (int *)(mtag + 1);
    408 		if (++(*count) > max_gif_nesting) {
    409 			log(LOG_NOTICE,
    410 			    "%s: recursively called too many times(%d)\n",
    411 			    if_name(ifp),
    412 			    *count);
    413 			return EIO;
    414 		}
    415 	} else {
    416 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
    417 		    M_NOWAIT);
    418 		if (mtag != NULL) {
    419 			m_tag_prepend(m, mtag);
    420 			count = (int *)(mtag + 1);
    421 			*count = 0;
    422 		} else {
    423 			log(LOG_DEBUG,
    424 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
    425 			    if_name(ifp));
    426 		}
    427 	}
    428 
    429 	return 0;
    430 }
    431 
    432 static int
    433 gif_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
    434     const struct rtentry *rt)
    435 {
    436 	struct gif_softc *sc = ifp->if_softc;
    437 	int error = 0;
    438 
    439 	IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family);
    440 
    441 	if ((error = gif_check_nesting(ifp, m)) != 0) {
    442 		m_free(m);
    443 		goto end;
    444 	}
    445 
    446 	m->m_flags &= ~(M_BCAST|M_MCAST);
    447 	if (((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING)) ||
    448 	    sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
    449 		m_freem(m);
    450 		error = ENETDOWN;
    451 		goto end;
    452 	}
    453 
    454 	/* XXX should we check if our outer source is legal? */
    455 
    456 	/* use DLT_NULL encapsulation here to pass inner af type */
    457 	M_PREPEND(m, sizeof(int), M_DONTWAIT);
    458 	if (!m) {
    459 		error = ENOBUFS;
    460 		goto end;
    461 	}
    462 	*mtod(m, int *) = dst->sa_family;
    463 
    464 	/* Clear checksum-offload flags. */
    465 	m->m_pkthdr.csum_flags = 0;
    466 	m->m_pkthdr.csum_data = 0;
    467 
    468 	error = if_transmit_lock(ifp, m);
    469   end:
    470 	if (error)
    471 		ifp->if_oerrors++;
    472 	return error;
    473 }
    474 
    475 static void
    476 gif_start(struct ifnet *ifp)
    477 {
    478 	struct gif_softc *sc;
    479 	struct mbuf *m;
    480 	int family;
    481 	int len;
    482 	int error;
    483 
    484 	sc = ifp->if_softc;
    485 
    486 	/* output processing */
    487 	while (1) {
    488 		IFQ_DEQUEUE(&sc->gif_if.if_snd, m);
    489 		if (m == NULL)
    490 			break;
    491 
    492 		/* grab and chop off inner af type */
    493 		if (sizeof(int) > m->m_len) {
    494 			m = m_pullup(m, sizeof(int));
    495 			if (!m) {
    496 				ifp->if_oerrors++;
    497 				continue;
    498 			}
    499 		}
    500 		family = *mtod(m, int *);
    501 		bpf_mtap(ifp, m);
    502 		m_adj(m, sizeof(int));
    503 
    504 		len = m->m_pkthdr.len;
    505 
    506 		/* dispatch to output logic based on outer AF */
    507 		switch (sc->gif_psrc->sa_family) {
    508 #ifdef INET
    509 		case AF_INET:
    510 			/* XXX
    511 			 * To add mutex_enter(softnet_lock) or
    512 			 * KASSERT(mutex_owned(softnet_lock)) here, we shold
    513 			 * coordinate softnet_lock between in6_if_up() and
    514 			 * in6_purgeif().
    515 			 */
    516 			error = in_gif_output(ifp, family, m);
    517 			break;
    518 #endif
    519 #ifdef INET6
    520 		case AF_INET6:
    521 			/* XXX
    522 			 * the same as in_gif_output()
    523 			 */
    524 			error = in6_gif_output(ifp, family, m);
    525 			break;
    526 #endif
    527 		default:
    528 			m_freem(m);
    529 			error = ENETDOWN;
    530 			break;
    531 		}
    532 
    533 		if (error)
    534 			ifp->if_oerrors++;
    535 		else {
    536 			ifp->if_opackets++;
    537 			ifp->if_obytes += len;
    538 		}
    539 	}
    540 }
    541 
    542 static int
    543 gif_transmit(struct ifnet *ifp, struct mbuf *m)
    544 {
    545 	struct gif_softc *sc;
    546 	int family;
    547 	int len;
    548 	int error;
    549 
    550 	sc = ifp->if_softc;
    551 
    552 	/* output processing */
    553 	if (m == NULL)
    554 		return EINVAL;
    555 
    556 	/* grab and chop off inner af type */
    557 	if (sizeof(int) > m->m_len) {
    558 		m = m_pullup(m, sizeof(int));
    559 		if (!m) {
    560 			ifp->if_oerrors++;
    561 			return ENOBUFS;
    562 		}
    563 	}
    564 	family = *mtod(m, int *);
    565 	bpf_mtap(ifp, m);
    566 	m_adj(m, sizeof(int));
    567 
    568 	len = m->m_pkthdr.len;
    569 
    570 	/* dispatch to output logic based on outer AF */
    571 	switch (sc->gif_psrc->sa_family) {
    572 #ifdef INET
    573 	case AF_INET:
    574 		/* XXX
    575 		 * To add mutex_enter(softnet_lock) or
    576 		 * KASSERT(mutex_owned(softnet_lock)) here, we shold
    577 		 * coordinate softnet_lock between in6_if_up() and
    578 		 * in6_purgeif().
    579 		 */
    580 		error = in_gif_output(ifp, family, m);
    581 		break;
    582 #endif
    583 #ifdef INET6
    584 	case AF_INET6:
    585 		/* XXX
    586 		 * the same as in_gif_output()
    587 		 */
    588 		error = in6_gif_output(ifp, family, m);
    589 		break;
    590 #endif
    591 	default:
    592 		m_freem(m);
    593 		error = ENETDOWN;
    594 		break;
    595 	}
    596 
    597 	if (error)
    598 		ifp->if_oerrors++;
    599 	else {
    600 		ifp->if_opackets++;
    601 		ifp->if_obytes += len;
    602 	}
    603 
    604 	return error;
    605 }
    606 
    607 void
    608 gif_input(struct mbuf *m, int af, struct ifnet *ifp)
    609 {
    610 	pktqueue_t *pktq;
    611 	size_t pktlen;
    612 
    613 	if (ifp == NULL) {
    614 		/* just in case */
    615 		m_freem(m);
    616 		return;
    617 	}
    618 
    619 	m_set_rcvif(m, ifp);
    620 	pktlen = m->m_pkthdr.len;
    621 
    622 	bpf_mtap_af(ifp, af, m);
    623 
    624 	/*
    625 	 * Put the packet to the network layer input queue according to the
    626 	 * specified address family.  Note: we avoid direct call to the
    627 	 * input function of the network layer in order to avoid recursion.
    628 	 * This may be revisited in the future.
    629 	 */
    630 	switch (af) {
    631 #ifdef INET
    632 	case AF_INET:
    633 		pktq = ip_pktq;
    634 		break;
    635 #endif
    636 #ifdef INET6
    637 	case AF_INET6:
    638 		pktq = ip6_pktq;
    639 		break;
    640 #endif
    641 	default:
    642 		m_freem(m);
    643 		return;
    644 	}
    645 
    646 #ifdef GIF_MPSAFE
    647 	const u_int h = curcpu()->ci_index;
    648 #else
    649 	const uint32_t h = pktq_rps_hash(m);
    650 #endif
    651 	if (__predict_true(pktq_enqueue(pktq, m, h))) {
    652 		ifp->if_ibytes += pktlen;
    653 		ifp->if_ipackets++;
    654 	} else {
    655 		m_freem(m);
    656 	}
    657 }
    658 
    659 /* XXX how should we handle IPv6 scope on SIOC[GS]IFPHYADDR? */
    660 static int
    661 gif_ioctl(struct ifnet *ifp, u_long cmd, void *data)
    662 {
    663 	struct gif_softc *sc  = ifp->if_softc;
    664 	struct ifreq     *ifr = (struct ifreq*)data;
    665 	struct ifaddr    *ifa = (struct ifaddr*)data;
    666 	int error = 0, size;
    667 	struct sockaddr *dst, *src;
    668 
    669 	switch (cmd) {
    670 	case SIOCINITIFADDR:
    671 		ifp->if_flags |= IFF_UP;
    672 		ifa->ifa_rtrequest = p2p_rtrequest;
    673 		break;
    674 
    675 	case SIOCADDMULTI:
    676 	case SIOCDELMULTI:
    677 		switch (ifr->ifr_addr.sa_family) {
    678 #ifdef INET
    679 		case AF_INET:	/* IP supports Multicast */
    680 			break;
    681 #endif /* INET */
    682 #ifdef INET6
    683 		case AF_INET6:	/* IP6 supports Multicast */
    684 			break;
    685 #endif /* INET6 */
    686 		default:  /* Other protocols doesn't support Multicast */
    687 			error = EAFNOSUPPORT;
    688 			break;
    689 		}
    690 		break;
    691 
    692 	case SIOCSIFMTU:
    693 		if (ifr->ifr_mtu < GIF_MTU_MIN || ifr->ifr_mtu > GIF_MTU_MAX)
    694 			return EINVAL;
    695 		else if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
    696 			error = 0;
    697 		break;
    698 
    699 #ifdef INET
    700 	case SIOCSIFPHYADDR:
    701 #endif
    702 #ifdef INET6
    703 	case SIOCSIFPHYADDR_IN6:
    704 #endif /* INET6 */
    705 	case SIOCSLIFPHYADDR:
    706 		switch (cmd) {
    707 #ifdef INET
    708 		case SIOCSIFPHYADDR:
    709 			src = (struct sockaddr *)
    710 				&(((struct in_aliasreq *)data)->ifra_addr);
    711 			dst = (struct sockaddr *)
    712 				&(((struct in_aliasreq *)data)->ifra_dstaddr);
    713 			break;
    714 #endif
    715 #ifdef INET6
    716 		case SIOCSIFPHYADDR_IN6:
    717 			src = (struct sockaddr *)
    718 				&(((struct in6_aliasreq *)data)->ifra_addr);
    719 			dst = (struct sockaddr *)
    720 				&(((struct in6_aliasreq *)data)->ifra_dstaddr);
    721 			break;
    722 #endif
    723 		case SIOCSLIFPHYADDR:
    724 			src = (struct sockaddr *)
    725 				&(((struct if_laddrreq *)data)->addr);
    726 			dst = (struct sockaddr *)
    727 				&(((struct if_laddrreq *)data)->dstaddr);
    728 			break;
    729 		default:
    730 			return EINVAL;
    731 		}
    732 
    733 		/* sa_family must be equal */
    734 		if (src->sa_family != dst->sa_family)
    735 			return EINVAL;
    736 
    737 		/* validate sa_len */
    738 		switch (src->sa_family) {
    739 #ifdef INET
    740 		case AF_INET:
    741 			if (src->sa_len != sizeof(struct sockaddr_in))
    742 				return EINVAL;
    743 			break;
    744 #endif
    745 #ifdef INET6
    746 		case AF_INET6:
    747 			if (src->sa_len != sizeof(struct sockaddr_in6))
    748 				return EINVAL;
    749 			break;
    750 #endif
    751 		default:
    752 			return EAFNOSUPPORT;
    753 		}
    754 		switch (dst->sa_family) {
    755 #ifdef INET
    756 		case AF_INET:
    757 			if (dst->sa_len != sizeof(struct sockaddr_in))
    758 				return EINVAL;
    759 			break;
    760 #endif
    761 #ifdef INET6
    762 		case AF_INET6:
    763 			if (dst->sa_len != sizeof(struct sockaddr_in6))
    764 				return EINVAL;
    765 			break;
    766 #endif
    767 		default:
    768 			return EAFNOSUPPORT;
    769 		}
    770 
    771 		/* check sa_family looks sane for the cmd */
    772 		switch (cmd) {
    773 		case SIOCSIFPHYADDR:
    774 			if (src->sa_family == AF_INET)
    775 				break;
    776 			return EAFNOSUPPORT;
    777 #ifdef INET6
    778 		case SIOCSIFPHYADDR_IN6:
    779 			if (src->sa_family == AF_INET6)
    780 				break;
    781 			return EAFNOSUPPORT;
    782 #endif /* INET6 */
    783 		case SIOCSLIFPHYADDR:
    784 			/* checks done in the above */
    785 			break;
    786 		}
    787 
    788 		error = gif_set_tunnel(&sc->gif_if, src, dst);
    789 		break;
    790 
    791 #ifdef SIOCDIFPHYADDR
    792 	case SIOCDIFPHYADDR:
    793 		gif_delete_tunnel(&sc->gif_if);
    794 		break;
    795 #endif
    796 
    797 	case SIOCGIFPSRCADDR:
    798 #ifdef INET6
    799 	case SIOCGIFPSRCADDR_IN6:
    800 #endif /* INET6 */
    801 		if (sc->gif_psrc == NULL) {
    802 			error = EADDRNOTAVAIL;
    803 			goto bad;
    804 		}
    805 		src = sc->gif_psrc;
    806 		switch (cmd) {
    807 #ifdef INET
    808 		case SIOCGIFPSRCADDR:
    809 			dst = &ifr->ifr_addr;
    810 			size = sizeof(ifr->ifr_addr);
    811 			break;
    812 #endif /* INET */
    813 #ifdef INET6
    814 		case SIOCGIFPSRCADDR_IN6:
    815 			dst = (struct sockaddr *)
    816 				&(((struct in6_ifreq *)data)->ifr_addr);
    817 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
    818 			break;
    819 #endif /* INET6 */
    820 		default:
    821 			error = EADDRNOTAVAIL;
    822 			goto bad;
    823 		}
    824 		if (src->sa_len > size)
    825 			return EINVAL;
    826 		memcpy(dst, src, src->sa_len);
    827 		break;
    828 
    829 	case SIOCGIFPDSTADDR:
    830 #ifdef INET6
    831 	case SIOCGIFPDSTADDR_IN6:
    832 #endif /* INET6 */
    833 		if (sc->gif_pdst == NULL) {
    834 			error = EADDRNOTAVAIL;
    835 			goto bad;
    836 		}
    837 		src = sc->gif_pdst;
    838 		switch (cmd) {
    839 #ifdef INET
    840 		case SIOCGIFPDSTADDR:
    841 			dst = &ifr->ifr_addr;
    842 			size = sizeof(ifr->ifr_addr);
    843 			break;
    844 #endif /* INET */
    845 #ifdef INET6
    846 		case SIOCGIFPDSTADDR_IN6:
    847 			dst = (struct sockaddr *)
    848 				&(((struct in6_ifreq *)data)->ifr_addr);
    849 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
    850 			break;
    851 #endif /* INET6 */
    852 		default:
    853 			error = EADDRNOTAVAIL;
    854 			goto bad;
    855 		}
    856 		if (src->sa_len > size)
    857 			return EINVAL;
    858 		memcpy(dst, src, src->sa_len);
    859 		break;
    860 
    861 	case SIOCGLIFPHYADDR:
    862 		if (sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
    863 			error = EADDRNOTAVAIL;
    864 			goto bad;
    865 		}
    866 
    867 		/* copy src */
    868 		src = sc->gif_psrc;
    869 		dst = (struct sockaddr *)
    870 			&(((struct if_laddrreq *)data)->addr);
    871 		size = sizeof(((struct if_laddrreq *)data)->addr);
    872 		if (src->sa_len > size)
    873 			return EINVAL;
    874 		memcpy(dst, src, src->sa_len);
    875 
    876 		/* copy dst */
    877 		src = sc->gif_pdst;
    878 		dst = (struct sockaddr *)
    879 			&(((struct if_laddrreq *)data)->dstaddr);
    880 		size = sizeof(((struct if_laddrreq *)data)->dstaddr);
    881 		if (src->sa_len > size)
    882 			return EINVAL;
    883 		memcpy(dst, src, src->sa_len);
    884 		break;
    885 
    886 	default:
    887 		return ifioctl_common(ifp, cmd, data);
    888 	}
    889  bad:
    890 	return error;
    891 }
    892 
    893 static int
    894 gif_encap_attach(struct gif_softc *sc)
    895 {
    896 	int error;
    897 
    898 	if (sc == NULL || sc->gif_psrc == NULL)
    899 		return EINVAL;
    900 
    901 	switch (sc->gif_psrc->sa_family) {
    902 #ifdef INET
    903 	case AF_INET:
    904 		error = in_gif_attach(sc);
    905 		break;
    906 #endif
    907 #ifdef INET6
    908 	case AF_INET6:
    909 		error = in6_gif_attach(sc);
    910 		break;
    911 #endif
    912 	default:
    913 		error = EINVAL;
    914 		break;
    915 	}
    916 
    917 	return error;
    918 }
    919 
    920 static int
    921 gif_encap_detach(struct gif_softc *sc)
    922 {
    923 	int error;
    924 
    925 	if (sc == NULL || sc->gif_psrc == NULL)
    926 		return EINVAL;
    927 
    928 	switch (sc->gif_psrc->sa_family) {
    929 #ifdef INET
    930 	case AF_INET:
    931 		error = in_gif_detach(sc);
    932 		break;
    933 #endif
    934 #ifdef INET6
    935 	case AF_INET6:
    936 		error = in6_gif_detach(sc);
    937 		break;
    938 #endif
    939 	default:
    940 		error = EINVAL;
    941 		break;
    942 	}
    943 
    944 	return error;
    945 }
    946 
    947 static void
    948 gif_encap_pause(struct gif_softc *sc)
    949 {
    950 	struct ifnet *ifp;
    951 	uint64_t where;
    952 
    953 	if (sc == NULL || sc->gif_psrc == NULL)
    954 		return;
    955 
    956 	ifp = &sc->gif_if;
    957 	if ((ifp->if_flags & IFF_RUNNING) == 0)
    958 		return;
    959 
    960 	switch (sc->gif_psrc->sa_family) {
    961 #ifdef INET
    962 	case AF_INET:
    963 		(void)in_gif_pause(sc);
    964 		break;
    965 #endif
    966 #ifdef INET6
    967 	case AF_INET6:
    968 		(void)in6_gif_pause(sc);
    969 		break;
    970 #endif
    971 	}
    972 
    973 	ifp->if_flags &= ~IFF_RUNNING;
    974 	/* membar_sync() is done in xc_broadcast(). */
    975 
    976 	/*
    977 	 * Wait for softint_execute()(ipintr() or ip6intr())
    978 	 * completion done by other CPUs which already run over if_flags
    979 	 * check in in_gif_input() or in6_gif_input().
    980 	 * Furthermore, wait for gif_output() completion too.
    981 	 */
    982 	where = xc_broadcast(0, (xcfunc_t)nullop, NULL, NULL);
    983 	xc_wait(where);
    984 }
    985 
    986 static int
    987 gif_set_tunnel(struct ifnet *ifp, struct sockaddr *src, struct sockaddr *dst)
    988 {
    989 	struct gif_softc *sc = ifp->if_softc;
    990 	struct gif_softc *sc2;
    991 	struct sockaddr *osrc, *odst;
    992 	struct sockaddr *nsrc, *ndst;
    993 	int error;
    994 #ifndef GIF_MPSAFE
    995 	int s;
    996 
    997 	s = splsoftnet();
    998 #endif
    999 	error = encap_lock_enter();
   1000 	if (error) {
   1001 #ifndef GIF_MPSAFE
   1002 		splx(s);
   1003 #endif
   1004 		return error;
   1005 	}
   1006 
   1007 	mutex_enter(&gif_softcs.lock);
   1008 	LIST_FOREACH(sc2, &gif_softcs.list, gif_list) {
   1009 		if (sc2 == sc)
   1010 			continue;
   1011 		if (!sc2->gif_pdst || !sc2->gif_psrc)
   1012 			continue;
   1013 		/* can't configure same pair of address onto two gifs */
   1014 		if (sockaddr_cmp(sc2->gif_pdst, dst) == 0 &&
   1015 		    sockaddr_cmp(sc2->gif_psrc, src) == 0) {
   1016 			/* continue to use the old configureation. */
   1017 			mutex_exit(&gif_softcs.lock);
   1018 			error =  EADDRNOTAVAIL;
   1019 			goto out;
   1020 		}
   1021 
   1022 		/* XXX both end must be valid? (I mean, not 0.0.0.0) */
   1023 	}
   1024 	mutex_exit(&gif_softcs.lock);
   1025 
   1026 	nsrc = sockaddr_dup(src, M_WAITOK);
   1027 	ndst = sockaddr_dup(dst, M_WAITOK);
   1028 
   1029 	gif_encap_pause(sc);
   1030 
   1031 	/* Firstly, clear old configurations. */
   1032 	/* XXX we can detach from both, but be polite just in case */
   1033 	if (sc->gif_psrc)
   1034 		(void)gif_encap_detach(sc);
   1035 
   1036 	/*
   1037 	 * Secondly, try to set new configurations.
   1038 	 */
   1039 	osrc = sc->gif_psrc;
   1040 	odst = sc->gif_pdst;
   1041 	sc->gif_psrc = nsrc;
   1042 	sc->gif_pdst = ndst;
   1043 	error = gif_encap_attach(sc);
   1044 	if (error && osrc != NULL && odst != NULL) {
   1045 		/*
   1046 		 * Thirdly, when error occured, rollback to old configurations,
   1047 		 * if last setting is valid.
   1048 		 */
   1049 		sc->gif_psrc = osrc;
   1050 		sc->gif_pdst = odst;
   1051 		osrc = nsrc; /* to free */
   1052 		odst = ndst; /* to free */
   1053 		error = gif_encap_attach(sc);
   1054 	}
   1055 	if (error) {
   1056 		/*
   1057 		 * Fourthly, even rollback failed or last setting is not valid,
   1058 		 * clear configurations.
   1059 		 */
   1060 		osrc = sc->gif_psrc; /* to free */
   1061 		odst = sc->gif_pdst; /* to free */
   1062 		sc->gif_psrc = NULL;
   1063 		sc->gif_pdst = NULL;
   1064 		sockaddr_free(nsrc);
   1065 		sockaddr_free(ndst);
   1066 	}
   1067 
   1068 	if (osrc)
   1069 		sockaddr_free(osrc);
   1070 	if (odst)
   1071 		sockaddr_free(odst);
   1072 
   1073 	if (sc->gif_psrc && sc->gif_pdst)
   1074 		ifp->if_flags |= IFF_RUNNING;
   1075 	else
   1076 		ifp->if_flags &= ~IFF_RUNNING;
   1077 
   1078  out:
   1079 	encap_lock_exit();
   1080 #ifndef GIF_MPSAFE
   1081 	splx(s);
   1082 #endif
   1083 	return error;
   1084 }
   1085 
   1086 static void
   1087 gif_delete_tunnel(struct ifnet *ifp)
   1088 {
   1089 	struct gif_softc *sc = ifp->if_softc;
   1090 	int error;
   1091 #ifndef GIF_MPSAFE
   1092 	int s;
   1093 
   1094 	s = splsoftnet();
   1095 #endif
   1096 	error = encap_lock_enter();
   1097 	if (error) {
   1098 #ifndef GIF_MPSAFE
   1099 		splx(s);
   1100 #endif
   1101 		return;
   1102 	}
   1103 
   1104 	gif_encap_pause(sc);
   1105 	if (sc->gif_psrc) {
   1106 		sockaddr_free(sc->gif_psrc);
   1107 		sc->gif_psrc = NULL;
   1108 	}
   1109 	if (sc->gif_pdst) {
   1110 		sockaddr_free(sc->gif_pdst);
   1111 		sc->gif_pdst = NULL;
   1112 	}
   1113 	/* it is safe to detach from both */
   1114 #ifdef INET
   1115 	(void)in_gif_detach(sc);
   1116 #endif
   1117 #ifdef INET6
   1118 	(void)in6_gif_detach(sc);
   1119 #endif
   1120 
   1121 	if (sc->gif_psrc && sc->gif_pdst)
   1122 		ifp->if_flags |= IFF_RUNNING;
   1123 	else
   1124 		ifp->if_flags &= ~IFF_RUNNING;
   1125 
   1126 	encap_lock_exit();
   1127 #ifndef GIF_MPSAFE
   1128 	splx(s);
   1129 #endif
   1130 }
   1131 
   1132 /*
   1133  * Module infrastructure
   1134  */
   1135 #include "if_module.h"
   1136 
   1137 IF_MODULE(MODULE_CLASS_DRIVER, gif, "")
   1138