Home | History | Annotate | Line # | Download | only in net
if_gre.c revision 1.98.6.7
      1 /*	$NetBSD: if_gre.c,v 1.98.6.7 2007/11/27 19:38:56 joerg Exp $ */
      2 
      3 /*
      4  * Copyright (c) 1998 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Heiko W.Rupp <hwr (at) pilhuhn.de>
      9  *
     10  * IPv6-over-GRE contributed by Gert Doering <gert (at) greenie.muc.de>
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *        This product includes software developed by the NetBSD
     23  *        Foundation, Inc. and its contributors.
     24  * 4. Neither the name of The NetBSD Foundation nor the names of its
     25  *    contributors may be used to endorse or promote products derived
     26  *    from this software without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     29  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     30  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     31  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     32  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     33  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     34  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     35  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     36  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     37  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     38  * POSSIBILITY OF SUCH DAMAGE.
     39  */
     40 
     41 /*
     42  * Encapsulate L3 protocols into IP
     43  * See RFC 1701 and 1702 for more details.
     44  * If_gre is compatible with Cisco GRE tunnels, so you can
     45  * have a NetBSD box as the other end of a tunnel interface of a Cisco
     46  * router. See gre(4) for more details.
     47  */
     48 
     49 #include <sys/cdefs.h>
     50 __KERNEL_RCSID(0, "$NetBSD: if_gre.c,v 1.98.6.7 2007/11/27 19:38:56 joerg Exp $");
     51 
     52 #include "opt_gre.h"
     53 #include "opt_inet.h"
     54 #include "bpfilter.h"
     55 
     56 #ifdef INET
     57 #include <sys/param.h>
     58 #include <sys/file.h>
     59 #include <sys/filedesc.h>
     60 #include <sys/malloc.h>
     61 #include <sys/mallocvar.h>
     62 #include <sys/mbuf.h>
     63 #include <sys/proc.h>
     64 #include <sys/domain.h>
     65 #include <sys/protosw.h>
     66 #include <sys/socket.h>
     67 #include <sys/socketvar.h>
     68 #include <sys/ioctl.h>
     69 #include <sys/queue.h>
     70 #include <sys/intr.h>
     71 #if __NetBSD__
     72 #include <sys/systm.h>
     73 #include <sys/sysctl.h>
     74 #include <sys/kauth.h>
     75 #endif
     76 
     77 #include <sys/kernel.h>
     78 #include <sys/mutex.h>
     79 #include <sys/condvar.h>
     80 #include <sys/kthread.h>
     81 
     82 #include <sys/cpu.h>
     83 
     84 #include <net/ethertypes.h>
     85 #include <net/if.h>
     86 #include <net/if_types.h>
     87 #include <net/netisr.h>
     88 #include <net/route.h>
     89 
     90 #ifdef INET
     91 #include <netinet/in.h>
     92 #include <netinet/in_systm.h>
     93 #include <netinet/in_var.h>
     94 #include <netinet/ip.h>
     95 #include <netinet/ip_var.h>
     96 #else
     97 #error "Huh? if_gre without inet?"
     98 #endif
     99 
    100 
    101 #ifdef NETATALK
    102 #include <netatalk/at.h>
    103 #include <netatalk/at_var.h>
    104 #include <netatalk/at_extern.h>
    105 #endif
    106 
    107 #if NBPFILTER > 0
    108 #include <sys/time.h>
    109 #include <net/bpf.h>
    110 #endif
    111 
    112 #include <net/if_gre.h>
    113 
    114 #include <compat/sys/socket.h>
    115 #include <compat/sys/sockio.h>
    116 /*
    117  * It is not easy to calculate the right value for a GRE MTU.
    118  * We leave this task to the admin and use the same default that
    119  * other vendors use.
    120  */
    121 #define GREMTU 1476
    122 
    123 #ifdef GRE_DEBUG
    124 int gre_debug = 0;
    125 #define	GRE_DPRINTF(__sc, __fmt, ...)				\
    126 	do {							\
    127 		if (__predict_false(gre_debug ||		\
    128 		    ((__sc)->sc_if.if_flags & IFF_DEBUG) != 0))	\
    129 			printf(__fmt, __VA_ARGS__);		\
    130 	} while (/*CONSTCOND*/0)
    131 #else
    132 #define	GRE_DPRINTF(__sc, __fmt, ...)	do { } while (/*CONSTCOND*/0)
    133 #endif /* GRE_DEBUG */
    134 
    135 int ip_gre_ttl = GRE_TTL;
    136 MALLOC_DEFINE(M_GRE_BUFQ, "gre_bufq", "gre mbuf queue");
    137 
    138 static int gre_clone_create(struct if_clone *, int);
    139 static int gre_clone_destroy(struct ifnet *);
    140 
    141 static struct if_clone gre_cloner =
    142     IF_CLONE_INITIALIZER("gre", gre_clone_create, gre_clone_destroy);
    143 
    144 static int gre_input(struct gre_softc *, struct mbuf *, int,
    145     const struct gre_h *);
    146 static bool gre_is_nullconf(const struct gre_soparm *);
    147 static int gre_output(struct ifnet *, struct mbuf *,
    148 			   const struct sockaddr *, struct rtentry *);
    149 static int gre_ioctl(struct ifnet *, u_long, void *);
    150 static void gre_closef(struct file **, struct lwp *);
    151 static int gre_getsockname(struct socket *, struct mbuf *, struct lwp *);
    152 static int gre_getpeername(struct socket *, struct mbuf *, struct lwp *);
    153 static int gre_getnames(struct socket *, struct lwp *,
    154     struct sockaddr_storage *, struct sockaddr_storage *);
    155 static void gre_clearconf(struct gre_soparm *, bool);
    156 static int gre_soreceive(struct socket *, struct mbuf **);
    157 static int gre_sosend(struct socket *, struct mbuf *, struct lwp *);
    158 static struct socket *gre_reconf(struct gre_softc *, struct socket *, lwp_t *,
    159     const struct gre_soparm *);
    160 
    161 static int
    162 nearest_pow2(size_t len0)
    163 {
    164 	size_t len, mid;
    165 
    166 	if (len0 == 0)
    167 		return 1;
    168 
    169 	for (len = len0; (len & (len - 1)) != 0; len &= len - 1)
    170 		;
    171 
    172 	mid = len | (len >> 1);
    173 
    174 	/* avoid overflow */
    175 	if ((len << 1) < len)
    176 		return len;
    177 	if (len0 >= mid)
    178 		return len << 1;
    179 	return len;
    180 }
    181 
    182 static struct gre_bufq *
    183 gre_bufq_init(struct gre_bufq *bq, size_t len0)
    184 {
    185 	size_t len;
    186 
    187 	len = nearest_pow2(len0);
    188 
    189 	memset(bq, 0, sizeof(*bq));
    190 	bq->bq_buf = malloc(len * sizeof(struct mbuf *), M_GRE_BUFQ, M_WAITOK);
    191 	bq->bq_len = len;
    192 	bq->bq_lenmask = len - 1;
    193 
    194 	return bq;
    195 }
    196 
    197 static bool
    198 gre_bufq_empty(struct gre_bufq *bq)
    199 {
    200 	return bq->bq_prodidx == bq->bq_considx;
    201 }
    202 
    203 static struct mbuf *
    204 gre_bufq_dequeue(struct gre_bufq *bq)
    205 {
    206 	struct mbuf *m;
    207 
    208 	if (gre_bufq_empty(bq))
    209 		return NULL;
    210 
    211 	m = bq->bq_buf[bq->bq_considx];
    212 	bq->bq_considx = (bq->bq_considx + 1) & bq->bq_lenmask;
    213 
    214 	return m;
    215 }
    216 
    217 static void
    218 gre_bufq_purge(struct gre_bufq *bq)
    219 {
    220 	struct mbuf *m;
    221 
    222 	while ((m = gre_bufq_dequeue(bq)) != NULL)
    223 		m_freem(m);
    224 }
    225 
    226 static int
    227 gre_bufq_enqueue(struct gre_bufq *bq, struct mbuf *m)
    228 {
    229 	int next;
    230 
    231 	next = (bq->bq_prodidx + 1) & bq->bq_lenmask;
    232 
    233 	if (next == bq->bq_considx) {
    234 		bq->bq_drops++;
    235 		return ENOBUFS;
    236 	}
    237 
    238 	bq->bq_buf[bq->bq_prodidx] = m;
    239 	bq->bq_prodidx = next;
    240 	return 0;
    241 }
    242 
    243 static void
    244 greintr(void *arg)
    245 {
    246 	struct gre_softc *sc = (struct gre_softc *)arg;
    247 	struct socket *so = sc->sc_so;
    248 	lwp_t *l = curlwp;
    249 	int rc;
    250 	struct mbuf *m;
    251 
    252 	KASSERT(sc->sc_so != NULL);
    253 
    254 	sc->sc_send_ev.ev_count++;
    255 	GRE_DPRINTF(sc, "%s: enter\n", __func__);
    256 	while ((m = gre_bufq_dequeue(&sc->sc_snd)) != NULL) {
    257 		/* XXX handle ENOBUFS? */
    258 		if ((rc = gre_sosend(so, m, l)) != 0) {
    259 			GRE_DPRINTF(sc, "%s: gre_sosend failed %d\n", __func__,
    260 			    rc);
    261 		}
    262 	}
    263 }
    264 
    265 /* Caller must hold sc->sc_mtx. */
    266 static void
    267 gre_wait(struct gre_softc *sc)
    268 {
    269 	sc->sc_waiters++;
    270 	cv_wait(&sc->sc_condvar, &sc->sc_mtx);
    271 	sc->sc_waiters--;
    272 }
    273 
    274 /* Caller must hold sc->sc_mtx. */
    275 static void
    276 gre_join(struct gre_softc *sc)
    277 {
    278 	while (sc->sc_waiters > 0)
    279 		cv_wait(&sc->sc_condvar, &sc->sc_mtx);
    280 }
    281 
    282 static void
    283 gre_evcnt_detach(struct gre_softc *sc)
    284 {
    285 	evcnt_detach(&sc->sc_unsupp_ev);
    286 	evcnt_detach(&sc->sc_pullup_ev);
    287 	evcnt_detach(&sc->sc_error_ev);
    288 	evcnt_detach(&sc->sc_block_ev);
    289 	evcnt_detach(&sc->sc_recv_ev);
    290 
    291 	evcnt_detach(&sc->sc_oflow_ev);
    292 	evcnt_detach(&sc->sc_send_ev);
    293 }
    294 
    295 static void
    296 gre_evcnt_attach(struct gre_softc *sc)
    297 {
    298 	evcnt_attach_dynamic(&sc->sc_recv_ev, EVCNT_TYPE_MISC,
    299 	    NULL, sc->sc_if.if_xname, "recv");
    300 	evcnt_attach_dynamic(&sc->sc_block_ev, EVCNT_TYPE_MISC,
    301 	    &sc->sc_recv_ev, sc->sc_if.if_xname, "would block");
    302 	evcnt_attach_dynamic(&sc->sc_error_ev, EVCNT_TYPE_MISC,
    303 	    &sc->sc_recv_ev, sc->sc_if.if_xname, "error");
    304 	evcnt_attach_dynamic(&sc->sc_pullup_ev, EVCNT_TYPE_MISC,
    305 	    &sc->sc_recv_ev, sc->sc_if.if_xname, "pullup failed");
    306 	evcnt_attach_dynamic(&sc->sc_unsupp_ev, EVCNT_TYPE_MISC,
    307 	    &sc->sc_recv_ev, sc->sc_if.if_xname, "unsupported");
    308 
    309 	evcnt_attach_dynamic(&sc->sc_send_ev, EVCNT_TYPE_MISC,
    310 	    NULL, sc->sc_if.if_xname, "send");
    311 	evcnt_attach_dynamic(&sc->sc_oflow_ev, EVCNT_TYPE_MISC,
    312 	    &sc->sc_send_ev, sc->sc_if.if_xname, "overflow");
    313 }
    314 
    315 static int
    316 gre_clone_create(struct if_clone *ifc, int unit)
    317 {
    318 	struct gre_soparm sp;
    319 	struct gre_softc *sc;
    320 
    321 	sc = malloc(sizeof(struct gre_softc), M_DEVBUF, M_WAITOK|M_ZERO);
    322 	mutex_init(&sc->sc_mtx, MUTEX_DRIVER, IPL_SOFTNET);
    323 	cv_init(&sc->sc_condvar, "gre wait");
    324 
    325 	snprintf(sc->sc_if.if_xname, sizeof(sc->sc_if.if_xname), "%s%d",
    326 	    ifc->ifc_name, unit);
    327 	sc->sc_if.if_softc = sc;
    328 	sc->sc_if.if_type = IFT_TUNNEL;
    329 	sc->sc_if.if_addrlen = 0;
    330 	sc->sc_if.if_hdrlen = sizeof(struct ip) + sizeof(struct gre_h);
    331 	sc->sc_if.if_dlt = DLT_NULL;
    332 	sc->sc_if.if_mtu = GREMTU;
    333 	sc->sc_if.if_flags = IFF_POINTOPOINT|IFF_MULTICAST;
    334 	sc->sc_if.if_output = gre_output;
    335 	sc->sc_if.if_ioctl = gre_ioctl;
    336 	sockaddr_copy(sstosa(&sp.sp_dst), sizeof(sp.sp_dst), sintocsa(&in_any));
    337 	sockaddr_copy(sstosa(&sp.sp_src), sizeof(sp.sp_src), sintocsa(&in_any));
    338 	sp.sp_proto = IPPROTO_GRE;
    339 	sp.sp_type = SOCK_RAW;
    340 	sp.sp_bysock = 0;
    341 	sp.sp_fd = -1;
    342 	sc->sc_soparm = sp;
    343 
    344 	gre_evcnt_attach(sc);
    345 
    346 	gre_bufq_init(&sc->sc_snd, 17);
    347 	sc->sc_if.if_flags |= IFF_LINK0;
    348 	if_attach(&sc->sc_if);
    349 	if_alloc_sadl(&sc->sc_if);
    350 #if NBPFILTER > 0
    351 	bpfattach(&sc->sc_if, DLT_NULL, sizeof(u_int32_t));
    352 #endif
    353 	sc->sc_lwp = &lwp0;
    354 	sc->sc_state = GRE_S_IDLE;
    355 	return 0;
    356 }
    357 
    358 static int
    359 gre_clone_destroy(struct ifnet *ifp)
    360 {
    361 	int s;
    362 	struct gre_softc *sc = ifp->if_softc;
    363 
    364 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    365 
    366 #if NBPFILTER > 0
    367 	bpfdetach(ifp);
    368 #endif
    369 	if_detach(ifp);
    370 
    371 	/* Some LWPs may still wait in gre_ioctl_lock(), however,
    372 	 * no new LWP will enter gre_ioctl_lock(), because ifunit()
    373 	 * cannot locate the interface any longer.
    374 	 */
    375 	mutex_enter(&sc->sc_mtx);
    376 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    377 	while (sc->sc_state != GRE_S_IDLE)
    378 		gre_wait(sc);
    379 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    380 	sc->sc_state = GRE_S_DIE;
    381 	cv_broadcast(&sc->sc_condvar);
    382 	gre_join(sc);
    383 	/* At this point, no other LWP will access the gre_softc, so
    384 	 * we can release the mutex.
    385 	 */
    386 	mutex_exit(&sc->sc_mtx);
    387 	s = splnet();
    388 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    389 	/* Note that we must not hold the mutex while we call gre_reconf(). */
    390 	sc->sc_so = gre_reconf(sc, sc->sc_so, sc->sc_lwp, NULL);
    391 	splx(s);
    392 
    393 	cv_destroy(&sc->sc_condvar);
    394 	mutex_destroy(&sc->sc_mtx);
    395 	gre_evcnt_detach(sc);
    396 	free(sc, M_DEVBUF);
    397 
    398 	return 0;
    399 }
    400 
    401 static void
    402 gre_receive(struct socket *so, void *arg, int waitflag)
    403 {
    404 	struct gre_softc *sc = (struct gre_softc *)arg;
    405 	int rc;
    406 	const struct gre_h *gh;
    407 	struct mbuf *m;
    408 
    409 	GRE_DPRINTF(sc, "%s: enter\n", __func__);
    410 
    411 	sc->sc_recv_ev.ev_count++;
    412 
    413 	rc = gre_soreceive(so, &m);
    414 	/* TBD Back off if ECONNREFUSED (indicates
    415 	 * ICMP Port Unreachable)?
    416 	 */
    417 	if (rc == EWOULDBLOCK) {
    418 		GRE_DPRINTF(sc, "%s: EWOULDBLOCK\n", __func__);
    419 		sc->sc_block_ev.ev_count++;
    420 		return;
    421 	} else if (rc != 0 || m == NULL) {
    422 		GRE_DPRINTF(sc, "%s: rc %d m %p\n",
    423 		    sc->sc_if.if_xname, rc, (void *)m);
    424 		sc->sc_error_ev.ev_count++;
    425 		return;
    426 	}
    427 	if (m->m_len < sizeof(*gh) &&
    428 	    (m = m_pullup(m, sizeof(*gh))) == NULL) {
    429 		GRE_DPRINTF(sc, "%s: m_pullup failed\n", __func__);
    430 		sc->sc_pullup_ev.ev_count++;
    431 		return;
    432 	}
    433 	gh = mtod(m, const struct gre_h *);
    434 
    435 	if (gre_input(sc, m, 0, gh) == 0) {
    436 		sc->sc_unsupp_ev.ev_count++;
    437 		GRE_DPRINTF(sc, "%s: dropping unsupported\n", __func__);
    438 		m_freem(m);
    439 	}
    440 }
    441 
    442 static void
    443 gre_upcall_add(struct socket *so, void *arg)
    444 {
    445 	/* XXX What if the kernel already set an upcall? */
    446 	KASSERT((so->so_rcv.sb_flags & SB_UPCALL) == 0);
    447 	so->so_upcallarg = arg;
    448 	so->so_upcall = gre_receive;
    449 	so->so_rcv.sb_flags |= SB_UPCALL;
    450 }
    451 
    452 static void
    453 gre_upcall_remove(struct socket *so)
    454 {
    455 	so->so_rcv.sb_flags &= ~SB_UPCALL;
    456 	so->so_upcallarg = NULL;
    457 	so->so_upcall = NULL;
    458 }
    459 
    460 static int
    461 gre_socreate(struct gre_softc *sc, struct lwp *l,
    462     struct gre_soparm *sp, int *fdout)
    463 {
    464 	const struct protosw *pr;
    465 	int fd, rc;
    466 	struct mbuf *m;
    467 	struct sockaddr *sa;
    468 	sa_family_t af;
    469 	struct socket *so;
    470 
    471 	GRE_DPRINTF(sc, "%s: enter\n", __func__);
    472 
    473 	af = sp->sp_src.ss_family;
    474 	rc = fsocreate(af, &so, sp->sp_type, sp->sp_proto, l, &fd);
    475 	if (rc != 0) {
    476 		GRE_DPRINTF(sc, "%s: fsocreate failed\n", __func__);
    477 		return rc;
    478 	}
    479 
    480 	if ((m = getsombuf(so, MT_SONAME)) == NULL) {
    481 		rc = ENOBUFS;
    482 		goto out;
    483 	}
    484 	sa = mtod(m, struct sockaddr *);
    485 	sockaddr_copy(sa, MIN(MLEN, sizeof(sp->sp_src)), sstosa(&sp->sp_src));
    486 	m->m_len = sp->sp_src.ss_len;
    487 
    488 #if 0
    489 	/* XXX */
    490 	GRE_DPRINTF(sc, "%s: bind 0x%08" PRIx32 " port %d\n", __func__,
    491 	    sin->sin_addr.s_addr, ntohs(sin->sin_port));
    492 #endif
    493 	if ((rc = sobind(so, m, l)) != 0) {
    494 		GRE_DPRINTF(sc, "%s: sobind failed\n", __func__);
    495 		goto out;
    496 	}
    497 
    498 	if ((rc = gre_getsockname(so, m, l)) != 0) {
    499 		GRE_DPRINTF(sc, "%s: gre_getsockname\n", __func__);
    500 		goto out;
    501 	}
    502 	sockaddr_copy(sstosa(&sp->sp_src), sizeof(sp->sp_src), sa);
    503 
    504 	sockaddr_copy(sa, MIN(MLEN, sizeof(sp->sp_dst)), sstosa(&sp->sp_dst));
    505 	m->m_len = sp->sp_dst.ss_len;
    506 
    507 	if ((rc = soconnect(so, m, l)) != 0) {
    508 		GRE_DPRINTF(sc, "%s: soconnect failed\n", __func__);
    509 		goto out;
    510 	}
    511 
    512 	/* XXX convert to a (new) SOL_SOCKET call */
    513 	*mtod(m, int *) = ip_gre_ttl;
    514 	m->m_len = sizeof(int);
    515 	pr = so->so_proto;
    516 	KASSERT(pr != NULL);
    517 	rc = sosetopt(so, IPPROTO_IP, IP_TTL, m);
    518 	m = NULL;
    519 	if (rc != 0) {
    520 		GRE_DPRINTF(sc, "%s: sosetopt ttl failed\n", __func__);
    521 		rc = 0;
    522 	}
    523 	rc = sosetopt(so, SOL_SOCKET, SO_NOHEADER, m_intopt(so, 1));
    524 	if (rc != 0) {
    525 		GRE_DPRINTF(sc, "%s: sosetopt SO_NOHEADER failed\n", __func__);
    526 		rc = 0;
    527 	}
    528 out:
    529 	m_freem(m);
    530 
    531 	if (rc != 0)
    532 		fdrelease(l, fd);
    533 	else
    534 		*fdout = fd;
    535 
    536 	return rc;
    537 }
    538 
    539 static int
    540 gre_sosend(struct socket *so, struct mbuf *top, struct lwp *l)
    541 {
    542 	struct mbuf	**mp;
    543 	struct proc	*p;
    544 	long		space, resid;
    545 	int		error, s;
    546 
    547 	p = l->l_proc;
    548 
    549 	resid = top->m_pkthdr.len;
    550 	if (p)
    551 		p->p_stats->p_ru.ru_msgsnd++;
    552 #define	snderr(errno)	{ error = errno; splx(s); goto release; }
    553 
    554 	if ((error = sblock(&so->so_snd, M_NOWAIT)) != 0)
    555 		goto out;
    556 	s = splsoftnet();
    557 	if (so->so_state & SS_CANTSENDMORE)
    558 		snderr(EPIPE);
    559 	if (so->so_error) {
    560 		error = so->so_error;
    561 		so->so_error = 0;
    562 		splx(s);
    563 		goto release;
    564 	}
    565 	if ((so->so_state & SS_ISCONNECTED) == 0) {
    566 		if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
    567 			if ((so->so_state & SS_ISCONFIRMING) == 0)
    568 				snderr(ENOTCONN);
    569 		} else
    570 			snderr(EDESTADDRREQ);
    571 	}
    572 	space = sbspace(&so->so_snd);
    573 	if (resid > so->so_snd.sb_hiwat)
    574 		snderr(EMSGSIZE);
    575 	if (space < resid)
    576 		snderr(EWOULDBLOCK);
    577 	splx(s);
    578 	mp = &top;
    579 	/*
    580 	 * Data is prepackaged in "top".
    581 	 */
    582 	s = splsoftnet();
    583 
    584 	if (so->so_state & SS_CANTSENDMORE)
    585 		snderr(EPIPE);
    586 
    587 	error = (*so->so_proto->pr_usrreq)(so, PRU_SEND, top, NULL, NULL, l);
    588 	splx(s);
    589 
    590 	top = NULL;
    591 	mp = &top;
    592 	if (error != 0)
    593 		goto release;
    594 
    595  release:
    596 	sbunlock(&so->so_snd);
    597  out:
    598 	if (top != NULL)
    599 		m_freem(top);
    600 	return error;
    601 }
    602 
    603 /* This is a stripped-down version of soreceive() that will never
    604  * block.  It will support SOCK_DGRAM sockets.  It may also support
    605  * SOCK_SEQPACKET sockets.
    606  */
    607 static int
    608 gre_soreceive(struct socket *so, struct mbuf **mp0)
    609 {
    610 	struct lwp *l = curlwp;
    611 	struct mbuf *m, **mp;
    612 	int flags, len, error, s, type;
    613 	const struct protosw	*pr;
    614 	struct mbuf *nextrecord;
    615 
    616 	KASSERT(mp0 != NULL);
    617 
    618 	flags = MSG_DONTWAIT;
    619 	pr = so->so_proto;
    620 	mp = mp0;
    621 	type = 0;
    622 
    623 	*mp = NULL;
    624 
    625 	KASSERT(pr->pr_flags & PR_ATOMIC);
    626 
    627 	if (so->so_state & SS_ISCONFIRMING)
    628 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL, NULL, NULL, l);
    629 
    630  restart:
    631 	if ((error = sblock(&so->so_rcv, M_NOWAIT)) != 0)
    632 		return error;
    633 	s = splsoftnet();
    634 
    635 	m = so->so_rcv.sb_mb;
    636 	/*
    637 	 * If we have less data than requested, do not block awaiting more.
    638 	 */
    639 	if (m == NULL) {
    640 #ifdef DIAGNOSTIC
    641 		if (so->so_rcv.sb_cc)
    642 			panic("receive 1");
    643 #endif
    644 		if (so->so_error) {
    645 			error = so->so_error;
    646 			so->so_error = 0;
    647 		} else if (so->so_state & SS_CANTRCVMORE)
    648 			;
    649 		else if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0
    650 		      && (so->so_proto->pr_flags & PR_CONNREQUIRED))
    651 			error = ENOTCONN;
    652 		else
    653 			error = EWOULDBLOCK;
    654 		goto release;
    655 	}
    656 	/*
    657 	 * On entry here, m points to the first record of the socket buffer.
    658 	 * While we process the initial mbufs containing address and control
    659 	 * info, we save a copy of m->m_nextpkt into nextrecord.
    660 	 */
    661 	if (l != NULL)
    662 		l->l_proc->p_stats->p_ru.ru_msgrcv++;
    663 	KASSERT(m == so->so_rcv.sb_mb);
    664 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 1");
    665 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 1");
    666 	nextrecord = m->m_nextpkt;
    667 	if (pr->pr_flags & PR_ADDR) {
    668 #ifdef DIAGNOSTIC
    669 		if (m->m_type != MT_SONAME)
    670 			panic("receive 1a");
    671 #endif
    672 		sbfree(&so->so_rcv, m);
    673 		MFREE(m, so->so_rcv.sb_mb);
    674 		m = so->so_rcv.sb_mb;
    675 	}
    676 	while (m != NULL && m->m_type == MT_CONTROL && error == 0) {
    677 		sbfree(&so->so_rcv, m);
    678 		/*
    679 		 * Dispose of any SCM_RIGHTS message that went
    680 		 * through the read path rather than recv.
    681 		 */
    682 		if (pr->pr_domain->dom_dispose &&
    683 		    mtod(m, struct cmsghdr *)->cmsg_type == SCM_RIGHTS)
    684 			(*pr->pr_domain->dom_dispose)(m);
    685 		MFREE(m, so->so_rcv.sb_mb);
    686 		m = so->so_rcv.sb_mb;
    687 	}
    688 
    689 	/*
    690 	 * If m is non-NULL, we have some data to read.  From now on,
    691 	 * make sure to keep sb_lastrecord consistent when working on
    692 	 * the last packet on the chain (nextrecord == NULL) and we
    693 	 * change m->m_nextpkt.
    694 	 */
    695 	if (m != NULL) {
    696 		m->m_nextpkt = nextrecord;
    697 		/*
    698 		 * If nextrecord == NULL (this is a single chain),
    699 		 * then sb_lastrecord may not be valid here if m
    700 		 * was changed earlier.
    701 		 */
    702 		if (nextrecord == NULL) {
    703 			KASSERT(so->so_rcv.sb_mb == m);
    704 			so->so_rcv.sb_lastrecord = m;
    705 		}
    706 		type = m->m_type;
    707 		if (type == MT_OOBDATA)
    708 			flags |= MSG_OOB;
    709 	} else {
    710 		KASSERT(so->so_rcv.sb_mb == m);
    711 		so->so_rcv.sb_mb = nextrecord;
    712 		SB_EMPTY_FIXUP(&so->so_rcv);
    713 	}
    714 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 2");
    715 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 2");
    716 
    717 	while (m != NULL) {
    718 		if (m->m_type == MT_OOBDATA) {
    719 			if (type != MT_OOBDATA)
    720 				break;
    721 		} else if (type == MT_OOBDATA)
    722 			break;
    723 #ifdef DIAGNOSTIC
    724 		else if (m->m_type != MT_DATA && m->m_type != MT_HEADER)
    725 			panic("receive 3");
    726 #endif
    727 		so->so_state &= ~SS_RCVATMARK;
    728 		if (so->so_oobmark != 0 && so->so_oobmark < m->m_len)
    729 			break;
    730 		len = m->m_len;
    731 		/*
    732 		 * mp is set, just pass back the mbufs.
    733 		 * Sockbuf must be consistent here (points to current mbuf,
    734 		 * it points to next record) when we drop priority;
    735 		 * we must note any additions to the sockbuf when we
    736 		 * block interrupts again.
    737 		 */
    738 		if (m->m_flags & M_EOR)
    739 			flags |= MSG_EOR;
    740 		nextrecord = m->m_nextpkt;
    741 		sbfree(&so->so_rcv, m);
    742 		*mp = m;
    743 		mp = &m->m_next;
    744 		so->so_rcv.sb_mb = m = m->m_next;
    745 		*mp = NULL;
    746 		/*
    747 		 * If m != NULL, we also know that
    748 		 * so->so_rcv.sb_mb != NULL.
    749 		 */
    750 		KASSERT(so->so_rcv.sb_mb == m);
    751 		if (m) {
    752 			m->m_nextpkt = nextrecord;
    753 			if (nextrecord == NULL)
    754 				so->so_rcv.sb_lastrecord = m;
    755 		} else {
    756 			so->so_rcv.sb_mb = nextrecord;
    757 			SB_EMPTY_FIXUP(&so->so_rcv);
    758 		}
    759 		SBLASTRECORDCHK(&so->so_rcv, "soreceive 3");
    760 		SBLASTMBUFCHK(&so->so_rcv, "soreceive 3");
    761 		if (so->so_oobmark) {
    762 			so->so_oobmark -= len;
    763 			if (so->so_oobmark == 0) {
    764 				so->so_state |= SS_RCVATMARK;
    765 				break;
    766 			}
    767 		}
    768 		if (flags & MSG_EOR)
    769 			break;
    770 	}
    771 
    772 	if (m != NULL) {
    773 		m_freem(*mp);
    774 		*mp = NULL;
    775 		error = ENOMEM;
    776 		(void) sbdroprecord(&so->so_rcv);
    777 	} else {
    778 		/*
    779 		 * First part is an inline SB_EMPTY_FIXUP().  Second
    780 		 * part makes sure sb_lastrecord is up-to-date if
    781 		 * there is still data in the socket buffer.
    782 		 */
    783 		so->so_rcv.sb_mb = nextrecord;
    784 		if (so->so_rcv.sb_mb == NULL) {
    785 			so->so_rcv.sb_mbtail = NULL;
    786 			so->so_rcv.sb_lastrecord = NULL;
    787 		} else if (nextrecord->m_nextpkt == NULL)
    788 			so->so_rcv.sb_lastrecord = nextrecord;
    789 	}
    790 	SBLASTRECORDCHK(&so->so_rcv, "soreceive 4");
    791 	SBLASTMBUFCHK(&so->so_rcv, "soreceive 4");
    792 	if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
    793 		(*pr->pr_usrreq)(so, PRU_RCVD, NULL,
    794 		    (struct mbuf *)(long)flags, NULL, l);
    795 	if (*mp0 == NULL && (flags & MSG_EOR) == 0 &&
    796 	    (so->so_state & SS_CANTRCVMORE) == 0) {
    797 		sbunlock(&so->so_rcv);
    798 		splx(s);
    799 		goto restart;
    800 	}
    801 
    802  release:
    803 	sbunlock(&so->so_rcv);
    804 	splx(s);
    805 	return error;
    806 }
    807 
    808 static struct socket *
    809 gre_reconf(struct gre_softc *sc, struct socket *so, lwp_t *l,
    810     const struct gre_soparm *newsoparm)
    811 {
    812 	int rc;
    813 	struct file *fp;
    814 	struct ifnet *ifp = &sc->sc_if;
    815 
    816 	GRE_DPRINTF(sc, "%s: enter\n", __func__);
    817 
    818 shutdown:
    819 	if (sc->sc_soparm.sp_fd != -1) {
    820 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    821 		gre_upcall_remove(so);
    822 		softint_disestablish(sc->sc_si);
    823 		sc->sc_si = NULL;
    824 		fdrelease(l, sc->sc_soparm.sp_fd);
    825 		gre_clearconf(&sc->sc_soparm, false);
    826 		sc->sc_soparm.sp_fd = -1;
    827 		so = NULL;
    828 	}
    829 
    830 	if (newsoparm != NULL) {
    831 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    832 		sc->sc_soparm = *newsoparm;
    833 	}
    834 
    835 	if (sc->sc_soparm.sp_fd != -1) {
    836 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    837 		rc = getsock(l->l_proc->p_fd, sc->sc_soparm.sp_fd, &fp);
    838 		if (rc != 0)
    839 			goto shutdown;
    840 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    841 		FILE_UNUSE(fp, NULL);
    842 		so = (struct socket *)fp->f_data;
    843 		sc->sc_si = softint_establish(SOFTINT_NET, greintr, sc);
    844 		gre_upcall_add(so, sc);
    845 		if ((ifp->if_flags & IFF_UP) == 0) {
    846 			GRE_DPRINTF(sc, "%s: down\n", __func__);
    847 			goto shutdown;
    848 		}
    849 	}
    850 
    851 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
    852 	if (so != NULL)
    853 		sc->sc_if.if_flags |= IFF_RUNNING;
    854 	else {
    855 		gre_bufq_purge(&sc->sc_snd);
    856 		sc->sc_if.if_flags &= ~IFF_RUNNING;
    857 	}
    858 	return so;
    859 }
    860 
    861 static int
    862 gre_input(struct gre_softc *sc, struct mbuf *m, int hlen,
    863     const struct gre_h *gh)
    864 {
    865 	u_int16_t flags;
    866 	u_int32_t af;		/* af passed to BPF tap */
    867 	int isr, s;
    868 	struct ifqueue *ifq;
    869 
    870 	sc->sc_if.if_ipackets++;
    871 	sc->sc_if.if_ibytes += m->m_pkthdr.len;
    872 
    873 	hlen += sizeof(struct gre_h);
    874 
    875 	/* process GRE flags as packet can be of variable len */
    876 	flags = ntohs(gh->flags);
    877 
    878 	/* Checksum & Offset are present */
    879 	if ((flags & GRE_CP) | (flags & GRE_RP))
    880 		hlen += 4;
    881 	/* We don't support routing fields (variable length) */
    882 	if (flags & GRE_RP) {
    883 		sc->sc_if.if_ierrors++;
    884 		return 0;
    885 	}
    886 	if (flags & GRE_KP)
    887 		hlen += 4;
    888 	if (flags & GRE_SP)
    889 		hlen += 4;
    890 
    891 	switch (ntohs(gh->ptype)) { /* ethertypes */
    892 	case ETHERTYPE_IP:
    893 		ifq = &ipintrq;
    894 		isr = NETISR_IP;
    895 		af = AF_INET;
    896 		break;
    897 #ifdef NETATALK
    898 	case ETHERTYPE_ATALK:
    899 		ifq = &atintrq1;
    900 		isr = NETISR_ATALK;
    901 		af = AF_APPLETALK;
    902 		break;
    903 #endif
    904 #ifdef INET6
    905 	case ETHERTYPE_IPV6:
    906 		ifq = &ip6intrq;
    907 		isr = NETISR_IPV6;
    908 		af = AF_INET6;
    909 		break;
    910 #endif
    911 	default:	   /* others not yet supported */
    912 		GRE_DPRINTF(sc, "%s: unhandled ethertype 0x%04x\n", __func__,
    913 		    ntohs(gh->ptype));
    914 		sc->sc_if.if_noproto++;
    915 		return 0;
    916 	}
    917 
    918 	if (hlen > m->m_pkthdr.len) {
    919 		m_freem(m);
    920 		sc->sc_if.if_ierrors++;
    921 		return EINVAL;
    922 	}
    923 	m_adj(m, hlen);
    924 
    925 #if NBPFILTER > 0
    926 	if (sc->sc_if.if_bpf != NULL)
    927 		bpf_mtap_af(sc->sc_if.if_bpf, af, m);
    928 #endif /*NBPFILTER > 0*/
    929 
    930 	m->m_pkthdr.rcvif = &sc->sc_if;
    931 
    932 	s = splnet();
    933 	if (IF_QFULL(ifq)) {
    934 		IF_DROP(ifq);
    935 		m_freem(m);
    936 	} else {
    937 		IF_ENQUEUE(ifq, m);
    938 	}
    939 	/* we need schednetisr since the address family may change */
    940 	schednetisr(isr);
    941 	splx(s);
    942 
    943 	return 1;	/* packet is done, no further processing needed */
    944 }
    945 
    946 /*
    947  * The output routine. Takes a packet and encapsulates it in the protocol
    948  * given by sc->sc_soparm.sp_proto. See also RFC 1701 and RFC 2004
    949  */
    950 static int
    951 gre_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
    952 	   struct rtentry *rt)
    953 {
    954 	int error = 0;
    955 	struct gre_softc *sc = ifp->if_softc;
    956 	struct gre_h *gh;
    957 	struct ip *ip;
    958 	u_int8_t ip_tos = 0;
    959 	u_int16_t etype = 0;
    960 
    961 	if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) != (IFF_UP|IFF_RUNNING)) {
    962 		m_freem(m);
    963 		error = ENETDOWN;
    964 		goto end;
    965 	}
    966 
    967 #if NBPFILTER > 0
    968 	if (ifp->if_bpf != NULL)
    969 		bpf_mtap_af(ifp->if_bpf, dst->sa_family, m);
    970 #endif
    971 
    972 	m->m_flags &= ~(M_BCAST|M_MCAST);
    973 
    974 	GRE_DPRINTF(sc, "%s: dst->sa_family=%d\n", __func__, dst->sa_family);
    975 	switch (dst->sa_family) {
    976 	case AF_INET:
    977 		ip = mtod(m, struct ip *);
    978 		ip_tos = ip->ip_tos;
    979 		etype = htons(ETHERTYPE_IP);
    980 		break;
    981 #ifdef NETATALK
    982 	case AF_APPLETALK:
    983 		etype = htons(ETHERTYPE_ATALK);
    984 		break;
    985 #endif
    986 #ifdef INET6
    987 	case AF_INET6:
    988 		etype = htons(ETHERTYPE_IPV6);
    989 		break;
    990 #endif
    991 	default:
    992 		IF_DROP(&ifp->if_snd);
    993 		m_freem(m);
    994 		error = EAFNOSUPPORT;
    995 		goto end;
    996 	}
    997 
    998 	M_PREPEND(m, sizeof(*gh), M_DONTWAIT);
    999 
   1000 	if (m == NULL) {
   1001 		IF_DROP(&ifp->if_snd);
   1002 		error = ENOBUFS;
   1003 		goto end;
   1004 	}
   1005 
   1006 	gh = mtod(m, struct gre_h *);
   1007 	gh->flags = 0;
   1008 	gh->ptype = etype;
   1009 	/* XXX Need to handle IP ToS.  Look at how I handle IP TTL. */
   1010 
   1011 	ifp->if_opackets++;
   1012 	ifp->if_obytes += m->m_pkthdr.len;
   1013 
   1014 	/* send it off */
   1015 	if ((error = gre_bufq_enqueue(&sc->sc_snd, m)) != 0) {
   1016 		sc->sc_oflow_ev.ev_count++;
   1017 		m_freem(m);
   1018 	} else
   1019 		softint_schedule(sc->sc_si);
   1020   end:
   1021 	if (error)
   1022 		ifp->if_oerrors++;
   1023 	return error;
   1024 }
   1025 
   1026 static int
   1027 gre_getname(struct socket *so, int req, struct mbuf *nam, struct lwp *l)
   1028 {
   1029 	return (*so->so_proto->pr_usrreq)(so, req, NULL, nam, NULL, l);
   1030 }
   1031 
   1032 static int
   1033 gre_getsockname(struct socket *so, struct mbuf *nam, struct lwp *l)
   1034 {
   1035 	return gre_getname(so, PRU_SOCKADDR, nam, l);
   1036 }
   1037 
   1038 static int
   1039 gre_getpeername(struct socket *so, struct mbuf *nam, struct lwp *l)
   1040 {
   1041 	return gre_getname(so, PRU_PEERADDR, nam, l);
   1042 }
   1043 
   1044 static int
   1045 gre_getnames(struct socket *so, struct lwp *l, struct sockaddr_storage *src,
   1046     struct sockaddr_storage *dst)
   1047 {
   1048 	struct mbuf *m;
   1049 	struct sockaddr_storage *ss;
   1050 	int rc;
   1051 
   1052 	if ((m = getsombuf(so, MT_SONAME)) == NULL)
   1053 		return ENOBUFS;
   1054 
   1055 	ss = mtod(m, struct sockaddr_storage *);
   1056 
   1057 	if ((rc = gre_getsockname(so, m, l)) != 0)
   1058 		goto out;
   1059 	*src = *ss;
   1060 
   1061 	if ((rc = gre_getpeername(so, m, l)) != 0)
   1062 		goto out;
   1063 	*dst = *ss;
   1064 
   1065 out:
   1066 	m_freem(m);
   1067 	return rc;
   1068 }
   1069 
   1070 static void
   1071 gre_closef(struct file **fpp, struct lwp *l)
   1072 {
   1073 	struct file *fp = *fpp;
   1074 
   1075 	mutex_enter(&fp->f_lock);
   1076 	FILE_USE(fp);
   1077 	closef(fp, l);
   1078 	*fpp = NULL;
   1079 }
   1080 
   1081 static int
   1082 gre_ssock(struct ifnet *ifp, struct gre_soparm *sp, int fd)
   1083 {
   1084 	int error, kfd;
   1085 	const struct protosw *pr;
   1086 	struct file *fp;
   1087 	struct filedesc	*fdp;
   1088 	struct gre_softc *sc = ifp->if_softc;
   1089 	struct lwp *l = curlwp;
   1090 	struct proc *kp, *p = curproc;
   1091 	struct socket *so;
   1092 	struct sockaddr_storage dst, src;
   1093 
   1094 	/* getsock() will FILE_USE() and unlock the descriptor for us */
   1095 	if ((error = getsock(p->p_fd, fd, &fp)) != 0) {
   1096 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1097 		return EINVAL;
   1098 	}
   1099 
   1100 	/* Increase reference count.  Now that our reference to
   1101 	 * the file descriptor is counted, this thread can release
   1102 	 * our "use" of the descriptor, but it will not be destroyed
   1103 	 * by some other thread's action.  This thread needs to
   1104 	 * release its use, too, because one and only one thread
   1105 	 * can have use of the descriptor at once.  The kernel
   1106 	 * thread will pick up the use if it needs it.
   1107 	 */
   1108 	fp->f_count++;
   1109 	GRE_DPRINTF(sc, "%s: l.%d f_count %d\n", __func__, __LINE__,
   1110 	    fp->f_count);
   1111 	FILE_UNUSE(fp, NULL);
   1112 
   1113 	kp = sc->sc_lwp->l_proc;
   1114 	while ((error = fdalloc(kp, 0, &kfd)) != 0 && error == ENOSPC)
   1115 		fdexpand(kp);
   1116 	if (error != 0)
   1117 		goto closef;
   1118 	fdp = kp->p_fd;
   1119 	rw_enter(&fdp->fd_lock, RW_WRITER);
   1120 	fdp->fd_ofiles[kfd] = fp;
   1121 	rw_exit(&fdp->fd_lock);
   1122 
   1123 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1124 
   1125 	so = (struct socket *)fp->f_data;
   1126 	pr = so->so_proto;
   1127 	if ((pr->pr_flags & PR_ATOMIC) == 0 ||
   1128 	    (sp->sp_type != 0 && pr->pr_type != sp->sp_type) ||
   1129 	    (sp->sp_proto != 0 && pr->pr_protocol != 0 &&
   1130 	     pr->pr_protocol != sp->sp_proto)) {
   1131 		GRE_DPRINTF(sc, "%s: l.%d, type %d, proto %d\n", __func__,
   1132 		    __LINE__, pr->pr_type, pr->pr_protocol);
   1133 		error = EINVAL;
   1134 		goto release;
   1135 	}
   1136 
   1137 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1138 
   1139 	/* check address */
   1140 	if ((error = gre_getnames(so, l, &src, &dst)) != 0)
   1141 		goto release;
   1142 
   1143 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1144 
   1145 	if (error != 0)
   1146 		goto release;
   1147 
   1148 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1149 
   1150 	sp->sp_src = src;
   1151 	sp->sp_dst = dst;
   1152 	/* fp does not any longer belong to this thread. */
   1153 	sp->sp_fd = kfd;
   1154 
   1155 	/* XXX print src & dst */
   1156 
   1157 	return 0;
   1158 release:
   1159 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1160 	fdrelease(sc->sc_lwp, kfd);
   1161 	return error;
   1162 closef:
   1163 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1164 	gre_closef(&fp, l);
   1165 	return error;
   1166 }
   1167 
   1168 static bool
   1169 sockaddr_is_anyaddr(const struct sockaddr *sa)
   1170 {
   1171 	socklen_t anylen, salen;
   1172 	const void *anyaddr, *addr;
   1173 
   1174 	if ((anyaddr = sockaddr_anyaddr(sa, &anylen)) == NULL ||
   1175 	    (addr = sockaddr_const_addr(sa, &salen)) == NULL)
   1176 		return false;
   1177 
   1178 	if (salen > anylen)
   1179 		return false;
   1180 
   1181 	return memcmp(anyaddr, addr, MIN(anylen, salen)) == 0;
   1182 }
   1183 
   1184 static bool
   1185 gre_is_nullconf(const struct gre_soparm *sp)
   1186 {
   1187 	return sockaddr_is_anyaddr(sstocsa(&sp->sp_src)) ||
   1188 	       sockaddr_is_anyaddr(sstocsa(&sp->sp_dst));
   1189 }
   1190 
   1191 static void
   1192 gre_clearconf(struct gre_soparm *sp, bool force)
   1193 {
   1194 	if (sp->sp_bysock || force) {
   1195 		sockaddr_copy(sstosa(&sp->sp_src), sizeof(sp->sp_src),
   1196 		    sockaddr_any(sstosa(&sp->sp_src)));
   1197 		sockaddr_copy(sstosa(&sp->sp_dst), sizeof(sp->sp_dst),
   1198 		    sockaddr_any(sstosa(&sp->sp_dst)));
   1199 		sp->sp_bysock = 0;
   1200 	}
   1201 	sp->sp_fd = -1;
   1202 }
   1203 
   1204 static int
   1205 gre_ioctl_lock(struct gre_softc *sc)
   1206 {
   1207 	mutex_enter(&sc->sc_mtx);
   1208 
   1209 	while (sc->sc_state == GRE_S_IOCTL)
   1210 		gre_wait(sc);
   1211 
   1212 	if (sc->sc_state != GRE_S_IDLE) {
   1213 		cv_signal(&sc->sc_condvar);
   1214 		mutex_exit(&sc->sc_mtx);
   1215 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1216 		return ENXIO;
   1217 	}
   1218 
   1219 	sc->sc_state = GRE_S_IOCTL;
   1220 
   1221 	mutex_exit(&sc->sc_mtx);
   1222 	return 0;
   1223 }
   1224 
   1225 static void
   1226 gre_ioctl_unlock(struct gre_softc *sc)
   1227 {
   1228 	mutex_enter(&sc->sc_mtx);
   1229 
   1230 	KASSERT(sc->sc_state == GRE_S_IOCTL);
   1231 	sc->sc_state = GRE_S_IDLE;
   1232 	cv_signal(&sc->sc_condvar);
   1233 
   1234 	mutex_exit(&sc->sc_mtx);
   1235 }
   1236 
   1237 static int
   1238 gre_ioctl(struct ifnet *ifp, const u_long cmd, void *data)
   1239 {
   1240 	struct lwp *l = curlwp;
   1241 	struct ifreq *ifr;
   1242 	struct if_laddrreq *lifr = (struct if_laddrreq *)data;
   1243 	struct gre_softc *sc = ifp->if_softc;
   1244 	struct gre_soparm *sp;
   1245 	int fd, error = 0, oproto, otype, s;
   1246 	struct gre_soparm sp0;
   1247 
   1248 	ifr = data;
   1249 
   1250 	GRE_DPRINTF(sc, "%s: l.%d, cmd %lu\n", __func__, __LINE__, cmd);
   1251 
   1252 	switch (cmd) {
   1253 	case SIOCSIFFLAGS:
   1254 	case SIOCSIFMTU:
   1255 	case GRESPROTO:
   1256 	case GRESADDRD:
   1257 	case GRESADDRS:
   1258 	case GRESSOCK:
   1259 	case GREDSOCK:
   1260 	case SIOCSLIFPHYADDR:
   1261 	case SIOCDIFPHYADDR:
   1262 		if (kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE,
   1263 		    KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd,
   1264 		    NULL) != 0)
   1265 			return EPERM;
   1266 		break;
   1267 	default:
   1268 		break;
   1269 	}
   1270 
   1271 	if ((error = gre_ioctl_lock(sc)) != 0) {
   1272 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1273 		return error;
   1274 	}
   1275 	s = splnet();
   1276 
   1277 	sp0 = sc->sc_soparm;
   1278 	sp0.sp_fd = -1;
   1279 	sp = &sp0;
   1280 
   1281 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1282 
   1283 	switch (cmd) {
   1284 	case SIOCSIFADDR:
   1285 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1286 		if ((ifp->if_flags & IFF_UP) != 0)
   1287 			break;
   1288 		gre_clearconf(sp, false);
   1289 		ifp->if_flags |= IFF_UP;
   1290 		goto mksocket;
   1291 	case SIOCSIFDSTADDR:
   1292 		break;
   1293 	case SIOCSIFFLAGS:
   1294 		oproto = sp->sp_proto;
   1295 		otype = sp->sp_type;
   1296 		switch (ifr->ifr_flags & (IFF_LINK0|IFF_LINK2)) {
   1297 		case IFF_LINK0|IFF_LINK2:
   1298 			sp->sp_proto = IPPROTO_UDP;
   1299 			sp->sp_type = SOCK_DGRAM;
   1300 			break;
   1301 		case IFF_LINK2:
   1302 			sp->sp_proto = 0;
   1303 			sp->sp_type = 0;
   1304 			break;
   1305 		case IFF_LINK0:
   1306 			sp->sp_proto = IPPROTO_GRE;
   1307 			sp->sp_type = SOCK_RAW;
   1308 			break;
   1309 		default:
   1310 			GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1311 			error = EINVAL;
   1312 			goto out;
   1313 		}
   1314 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1315 		gre_clearconf(sp, false);
   1316 		if ((ifp->if_flags & (IFF_UP|IFF_RUNNING)) ==
   1317 		    (IFF_UP|IFF_RUNNING) &&
   1318 		    (oproto == sp->sp_proto || sp->sp_proto == 0) &&
   1319 		    (otype == sp->sp_type || sp->sp_type == 0))
   1320 			break;
   1321 		switch (sp->sp_proto) {
   1322 		case IPPROTO_UDP:
   1323 		case IPPROTO_GRE:
   1324 			goto mksocket;
   1325 		default:
   1326 			break;
   1327 		}
   1328 		break;
   1329 	case SIOCSIFMTU:
   1330 		/* XXX determine MTU automatically by probing w/
   1331 		 * XXX do-not-fragment packets?
   1332 		 */
   1333 		if (ifr->ifr_mtu < 576) {
   1334 			error = EINVAL;
   1335 			break;
   1336 		}
   1337 		ifp->if_mtu = ifr->ifr_mtu;
   1338 		break;
   1339 	case SIOCGIFMTU:
   1340 		ifr->ifr_mtu = sc->sc_if.if_mtu;
   1341 		break;
   1342 	case SIOCADDMULTI:
   1343 	case SIOCDELMULTI:
   1344 		if (ifr == NULL) {
   1345 			error = EAFNOSUPPORT;
   1346 			break;
   1347 		}
   1348 		switch (ifreq_getaddr(cmd, ifr)->sa_family) {
   1349 #ifdef INET
   1350 		case AF_INET:
   1351 			break;
   1352 #endif
   1353 #ifdef INET6
   1354 		case AF_INET6:
   1355 			break;
   1356 #endif
   1357 		default:
   1358 			error = EAFNOSUPPORT;
   1359 			break;
   1360 		}
   1361 		break;
   1362 	case GRESPROTO:
   1363 		gre_clearconf(sp, false);
   1364 		oproto = sp->sp_proto;
   1365 		otype = sp->sp_type;
   1366 		sp->sp_proto = ifr->ifr_flags;
   1367 		switch (sp->sp_proto) {
   1368 		case IPPROTO_UDP:
   1369 			ifp->if_flags |= IFF_LINK0|IFF_LINK2;
   1370 			sp->sp_type = SOCK_DGRAM;
   1371 			break;
   1372 		case IPPROTO_GRE:
   1373 			ifp->if_flags |= IFF_LINK0;
   1374 			ifp->if_flags &= ~IFF_LINK2;
   1375 			sp->sp_type = SOCK_RAW;
   1376 			break;
   1377 		case 0:
   1378 			ifp->if_flags &= ~IFF_LINK0;
   1379 			ifp->if_flags |= IFF_LINK2;
   1380 			sp->sp_type = 0;
   1381 			break;
   1382 		default:
   1383 			error = EPROTONOSUPPORT;
   1384 			break;
   1385 		}
   1386 		if ((oproto == sp->sp_proto || sp->sp_proto == 0) &&
   1387 		    (otype == sp->sp_type || sp->sp_type == 0))
   1388 			break;
   1389 		switch (sp->sp_proto) {
   1390 		case IPPROTO_UDP:
   1391 		case IPPROTO_GRE:
   1392 			goto mksocket;
   1393 		default:
   1394 			break;
   1395 		}
   1396 		break;
   1397 	case GREGPROTO:
   1398 		ifr->ifr_flags = sp->sp_proto;
   1399 		break;
   1400 	case GRESADDRS:
   1401 	case GRESADDRD:
   1402 		gre_clearconf(sp, false);
   1403 		/*
   1404 		 * set tunnel endpoints, compute a less specific route
   1405 		 * to the remote end and mark if as up
   1406 		 */
   1407 		switch (cmd) {
   1408 		case GRESADDRS:
   1409 			sockaddr_copy(sstosa(&sp->sp_src),
   1410 			    sizeof(sp->sp_src), ifreq_getaddr(cmd, ifr));
   1411 			break;
   1412 		case GRESADDRD:
   1413 			sockaddr_copy(sstosa(&sp->sp_dst),
   1414 			    sizeof(sp->sp_dst), ifreq_getaddr(cmd, ifr));
   1415 			break;
   1416 		}
   1417 	checkaddr:
   1418 		if (sockaddr_any(sstosa(&sp->sp_src)) == NULL ||
   1419 		    sockaddr_any(sstosa(&sp->sp_dst)) == NULL) {
   1420 			error = EINVAL;
   1421 			break;
   1422 		}
   1423 		/* let gre_socreate() check the rest */
   1424 	mksocket:
   1425 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1426 		/* If we're administratively down, or the configuration
   1427 		 * is empty, there's no use creating a socket.
   1428 		 */
   1429 		if ((ifp->if_flags & IFF_UP) == 0 || gre_is_nullconf(sp))
   1430 			goto sendconf;
   1431 
   1432 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1433 		error = gre_socreate(sc, l, sp, &fd);
   1434 
   1435 		if (error != 0)
   1436 			break;
   1437 
   1438 	setsock:
   1439 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1440 
   1441 		error = gre_ssock(ifp, sp, fd);
   1442 
   1443 		if (cmd != GRESSOCK) {
   1444 			GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1445 			fdrelease(l, fd);
   1446 		}
   1447 
   1448 		if (error == 0) {
   1449 	sendconf:
   1450 			GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1451 			ifp->if_flags &= ~IFF_RUNNING;
   1452 			sc->sc_so = gre_reconf(sc, sc->sc_so, sc->sc_lwp, sp);
   1453 		}
   1454 
   1455 		break;
   1456 	case GREGADDRS:
   1457 		ifreq_setaddr(cmd, ifr, sstosa(&sp->sp_src));
   1458 		break;
   1459 	case GREGADDRD:
   1460 		ifreq_setaddr(cmd, ifr, sstosa(&sp->sp_dst));
   1461 		break;
   1462 	case GREDSOCK:
   1463 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1464 		if (sp->sp_bysock)
   1465 			ifp->if_flags &= ~IFF_UP;
   1466 		gre_clearconf(sp, false);
   1467 		goto mksocket;
   1468 	case GRESSOCK:
   1469 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1470 		gre_clearconf(sp, true);
   1471 		fd = (int)ifr->ifr_value;
   1472 		sp->sp_bysock = 1;
   1473 		ifp->if_flags |= IFF_UP;
   1474 		goto setsock;
   1475 	case SIOCSLIFPHYADDR:
   1476 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1477 		if (lifr->addr.ss_family != lifr->dstaddr.ss_family) {
   1478 			error = EAFNOSUPPORT;
   1479 			break;
   1480 		}
   1481 		sockaddr_copy(sstosa(&sp->sp_src), sizeof(sp->sp_src),
   1482 		    sstosa(&lifr->addr));
   1483 		sockaddr_copy(sstosa(&sp->sp_dst), sizeof(sp->sp_dst),
   1484 		    sstosa(&lifr->dstaddr));
   1485 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1486 		goto checkaddr;
   1487 	case SIOCDIFPHYADDR:
   1488 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1489 		gre_clearconf(sp, true);
   1490 		ifp->if_flags &= ~IFF_UP;
   1491 		goto mksocket;
   1492 	case SIOCGLIFPHYADDR:
   1493 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1494 		if (gre_is_nullconf(sp)) {
   1495 			error = EADDRNOTAVAIL;
   1496 			break;
   1497 		}
   1498 		sockaddr_copy(sstosa(&lifr->addr), sizeof(lifr->addr),
   1499 		    sstosa(&sp->sp_src));
   1500 		sockaddr_copy(sstosa(&lifr->dstaddr), sizeof(lifr->dstaddr),
   1501 		    sstosa(&sp->sp_dst));
   1502 		GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1503 		break;
   1504 	default:
   1505 		error = EINVAL;
   1506 		break;
   1507 	}
   1508 out:
   1509 	GRE_DPRINTF(sc, "%s: l.%d\n", __func__, __LINE__);
   1510 	splx(s);
   1511 	gre_ioctl_unlock(sc);
   1512 	return error;
   1513 }
   1514 
   1515 #endif
   1516 
   1517 void	greattach(int);
   1518 
   1519 /* ARGSUSED */
   1520 void
   1521 greattach(int count)
   1522 {
   1523 #ifdef INET
   1524 	if_clone_attach(&gre_cloner);
   1525 #endif
   1526 }
   1527