Home | History | Annotate | Line # | Download | only in net
if_tun.c revision 1.112
      1 /*	$NetBSD: if_tun.c,v 1.112 2010/01/19 22:08:01 pooka Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1988, Julian Onions <jpo (at) cs.nott.ac.uk>
      5  * Nottingham University 1987.
      6  *
      7  * This source may be freely distributed, however I would be interested
      8  * in any changes that are made.
      9  *
     10  * This driver takes packets off the IP i/f and hands them up to a
     11  * user process to have its wicked way with. This driver has its
     12  * roots in a similar driver written by Phil Cockcroft (formerly) at
     13  * UCL. This driver is based much more on read/write/poll mode of
     14  * operation though.
     15  */
     16 
     17 #include <sys/cdefs.h>
     18 __KERNEL_RCSID(0, "$NetBSD: if_tun.c,v 1.112 2010/01/19 22:08:01 pooka Exp $");
     19 
     20 #include "opt_inet.h"
     21 
     22 #include <sys/param.h>
     23 #include <sys/proc.h>
     24 #include <sys/systm.h>
     25 #include <sys/mbuf.h>
     26 #include <sys/buf.h>
     27 #include <sys/protosw.h>
     28 #include <sys/socket.h>
     29 #include <sys/ioctl.h>
     30 #include <sys/errno.h>
     31 #include <sys/syslog.h>
     32 #include <sys/select.h>
     33 #include <sys/poll.h>
     34 #include <sys/file.h>
     35 #include <sys/signalvar.h>
     36 #include <sys/conf.h>
     37 #include <sys/kauth.h>
     38 #include <sys/simplelock.h>
     39 #include <sys/cpu.h>
     40 
     41 #include <net/if.h>
     42 #include <net/if_types.h>
     43 #include <net/netisr.h>
     44 #include <net/route.h>
     45 
     46 
     47 #ifdef INET
     48 #include <netinet/in.h>
     49 #include <netinet/in_systm.h>
     50 #include <netinet/in_var.h>
     51 #include <netinet/ip.h>
     52 #include <netinet/if_inarp.h>
     53 #endif
     54 
     55 
     56 #include <sys/time.h>
     57 #include <net/bpf.h>
     58 
     59 #include <net/if_tun.h>
     60 
     61 #define TUNDEBUG	if (tundebug) printf
     62 int	tundebug = 0;
     63 
     64 extern int ifqmaxlen;
     65 void	tunattach(int);
     66 
     67 static LIST_HEAD(, tun_softc) tun_softc_list;
     68 static LIST_HEAD(, tun_softc) tunz_softc_list;
     69 static struct simplelock tun_softc_lock;
     70 
     71 static int	tun_ioctl(struct ifnet *, u_long, void *);
     72 static int	tun_output(struct ifnet *, struct mbuf *,
     73 			const struct sockaddr *, struct rtentry *rt);
     74 static int	tun_clone_create(struct if_clone *, int);
     75 static int	tun_clone_destroy(struct ifnet *);
     76 
     77 static struct if_clone tun_cloner =
     78     IF_CLONE_INITIALIZER("tun", tun_clone_create, tun_clone_destroy);
     79 
     80 static void tunattach0(struct tun_softc *);
     81 static void tuninit(struct tun_softc *);
     82 static void tun_i_softintr(void *);
     83 static void tun_o_softintr(void *);
     84 #ifdef ALTQ
     85 static void tunstart(struct ifnet *);
     86 #endif
     87 static struct tun_softc *tun_find_unit(dev_t);
     88 static struct tun_softc *tun_find_zunit(int);
     89 
     90 static dev_type_open(tunopen);
     91 static dev_type_close(tunclose);
     92 static dev_type_read(tunread);
     93 static dev_type_write(tunwrite);
     94 static dev_type_ioctl(tunioctl);
     95 static dev_type_poll(tunpoll);
     96 static dev_type_kqfilter(tunkqfilter);
     97 
     98 const struct cdevsw tun_cdevsw = {
     99 	tunopen, tunclose, tunread, tunwrite, tunioctl,
    100 	nostop, notty, tunpoll, nommap, tunkqfilter, D_OTHER,
    101 };
    102 
    103 void
    104 tunattach(int unused)
    105 {
    106 
    107 	simple_lock_init(&tun_softc_lock);
    108 	LIST_INIT(&tun_softc_list);
    109 	LIST_INIT(&tunz_softc_list);
    110 	if_clone_attach(&tun_cloner);
    111 }
    112 
    113 /*
    114  * Find driver instance from dev_t.
    115  * Call at splnet().
    116  * Returns with tp locked (if found).
    117  */
    118 static struct tun_softc *
    119 tun_find_unit(dev_t dev)
    120 {
    121 	struct tun_softc *tp;
    122 	int unit = minor(dev);
    123 
    124 	simple_lock(&tun_softc_lock);
    125 	LIST_FOREACH(tp, &tun_softc_list, tun_list)
    126 		if (unit == tp->tun_unit)
    127 			break;
    128 	if (tp)
    129 		simple_lock(&tp->tun_lock);
    130 	simple_unlock(&tun_softc_lock);
    131 
    132 	return (tp);
    133 }
    134 
    135 /*
    136  * Find zombie driver instance by unit number.
    137  * Call at splnet().
    138  * Remove tp from list and return it unlocked (if found).
    139  */
    140 static struct tun_softc *
    141 tun_find_zunit(int unit)
    142 {
    143 	struct tun_softc *tp;
    144 
    145 	simple_lock(&tun_softc_lock);
    146 	LIST_FOREACH(tp, &tunz_softc_list, tun_list)
    147 		if (unit == tp->tun_unit)
    148 			break;
    149 	if (tp)
    150 		LIST_REMOVE(tp, tun_list);
    151 	simple_unlock(&tun_softc_lock);
    152 #ifdef DIAGNOSTIC
    153 	if (tp != NULL && (tp->tun_flags & (TUN_INITED|TUN_OPEN)) != TUN_OPEN)
    154 		printf("tun%d: inconsistent flags: %x\n", unit, tp->tun_flags);
    155 #endif
    156 
    157 	return (tp);
    158 }
    159 
    160 static int
    161 tun_clone_create(struct if_clone *ifc, int unit)
    162 {
    163 	struct tun_softc *tp;
    164 
    165 	if ((tp = tun_find_zunit(unit)) == NULL) {
    166 		/* Allocate a new instance */
    167 		tp = malloc(sizeof(*tp), M_DEVBUF, M_WAITOK|M_ZERO);
    168 
    169 		tp->tun_unit = unit;
    170 		simple_lock_init(&tp->tun_lock);
    171 		selinit(&tp->tun_rsel);
    172 		selinit(&tp->tun_wsel);
    173 	} else {
    174 		/* Revive tunnel instance; clear ifp part */
    175 		(void)memset(&tp->tun_if, 0, sizeof(struct ifnet));
    176 	}
    177 
    178 	if_initname(&tp->tun_if, ifc->ifc_name, unit);
    179 	tunattach0(tp);
    180 	tp->tun_flags |= TUN_INITED;
    181 	tp->tun_osih = softint_establish(SOFTINT_CLOCK, tun_o_softintr, tp);
    182 	tp->tun_isih = softint_establish(SOFTINT_CLOCK, tun_i_softintr, tp);
    183 
    184 	simple_lock(&tun_softc_lock);
    185 	LIST_INSERT_HEAD(&tun_softc_list, tp, tun_list);
    186 	simple_unlock(&tun_softc_lock);
    187 
    188 	return (0);
    189 }
    190 
    191 static void
    192 tunattach0(struct tun_softc *tp)
    193 {
    194 	struct ifnet *ifp;
    195 
    196 	ifp = &tp->tun_if;
    197 	ifp->if_softc = tp;
    198 	ifp->if_mtu = TUNMTU;
    199 	ifp->if_ioctl = tun_ioctl;
    200 	ifp->if_output = tun_output;
    201 #ifdef ALTQ
    202 	ifp->if_start = tunstart;
    203 #endif
    204 	ifp->if_flags = IFF_POINTOPOINT;
    205 	ifp->if_type = IFT_TUNNEL;
    206 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
    207 	ifp->if_collisions = 0;
    208 	ifp->if_ierrors = 0;
    209 	ifp->if_oerrors = 0;
    210 	ifp->if_ipackets = 0;
    211 	ifp->if_opackets = 0;
    212 	ifp->if_ibytes   = 0;
    213 	ifp->if_obytes   = 0;
    214 	ifp->if_dlt = DLT_NULL;
    215 	IFQ_SET_READY(&ifp->if_snd);
    216 	if_attach(ifp);
    217 	if_alloc_sadl(ifp);
    218 	bpf_ops->bpf_attach(ifp, DLT_NULL, sizeof(uint32_t), &ifp->if_bpf);
    219 }
    220 
    221 static int
    222 tun_clone_destroy(struct ifnet *ifp)
    223 {
    224 	struct tun_softc *tp = (void *)ifp;
    225 	int s, zombie = 0;
    226 
    227 	s = splnet();
    228 	simple_lock(&tun_softc_lock);
    229 	simple_lock(&tp->tun_lock);
    230 	LIST_REMOVE(tp, tun_list);
    231 	if (tp->tun_flags & TUN_OPEN) {
    232 		/* Hang on to storage until last close */
    233 		zombie = 1;
    234 		tp->tun_flags &= ~TUN_INITED;
    235 		LIST_INSERT_HEAD(&tunz_softc_list, tp, tun_list);
    236 	}
    237 	simple_unlock(&tun_softc_lock);
    238 
    239 	IF_PURGE(&ifp->if_snd);
    240 	ifp->if_flags &= ~IFF_RUNNING;
    241 
    242 	if (tp->tun_flags & TUN_RWAIT) {
    243 		tp->tun_flags &= ~TUN_RWAIT;
    244 		wakeup((void *)tp);
    245 	}
    246 	selnotify(&tp->tun_rsel, 0, 0);
    247 
    248 	simple_unlock(&tp->tun_lock);
    249 	splx(s);
    250 
    251 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    252 		fownsignal(tp->tun_pgid, SIGIO, POLL_HUP, 0, NULL);
    253 
    254 	bpf_ops->bpf_detach(ifp);
    255 	if_detach(ifp);
    256 
    257 	if (!zombie) {
    258 		seldestroy(&tp->tun_rsel);
    259 		seldestroy(&tp->tun_wsel);
    260 		softint_disestablish(tp->tun_osih);
    261 		softint_disestablish(tp->tun_isih);
    262 		free(tp, M_DEVBUF);
    263 	}
    264 
    265 	return (0);
    266 }
    267 
    268 /*
    269  * tunnel open - must be superuser & the device must be
    270  * configured in
    271  */
    272 static int
    273 tunopen(dev_t dev, int flag, int mode, struct lwp *l)
    274 {
    275 	struct ifnet	*ifp;
    276 	struct tun_softc *tp;
    277 	int	s, error;
    278 
    279 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE_TUN,
    280 	    KAUTH_REQ_NETWORK_INTERFACE_TUN_ADD, NULL, NULL, NULL);
    281 	if (error)
    282 		return (error);
    283 
    284 	s = splnet();
    285 	tp = tun_find_unit(dev);
    286 
    287 	if (tp == NULL) {
    288 		(void)tun_clone_create(&tun_cloner, minor(dev));
    289 		tp = tun_find_unit(dev);
    290 		if (tp == NULL) {
    291 			error = ENXIO;
    292 			goto out_nolock;
    293 		}
    294 	}
    295 
    296 	if (tp->tun_flags & TUN_OPEN) {
    297 		error = EBUSY;
    298 		goto out;
    299 	}
    300 
    301 	ifp = &tp->tun_if;
    302 	tp->tun_flags |= TUN_OPEN;
    303 	TUNDEBUG("%s: open\n", ifp->if_xname);
    304 out:
    305 	simple_unlock(&tp->tun_lock);
    306 out_nolock:
    307 	splx(s);
    308 	return (error);
    309 }
    310 
    311 /*
    312  * tunclose - close the device - mark i/f down & delete
    313  * routing info
    314  */
    315 int
    316 tunclose(dev_t dev, int flag, int mode,
    317     struct lwp *l)
    318 {
    319 	int	s;
    320 	struct tun_softc *tp;
    321 	struct ifnet	*ifp;
    322 
    323 	s = splnet();
    324 	if ((tp = tun_find_zunit(minor(dev))) != NULL) {
    325 		/* interface was "destroyed" before the close */
    326 		seldestroy(&tp->tun_rsel);
    327 		seldestroy(&tp->tun_wsel);
    328 		softint_disestablish(tp->tun_osih);
    329 		softint_disestablish(tp->tun_isih);
    330 		free(tp, M_DEVBUF);
    331 		goto out_nolock;
    332 	}
    333 
    334 	if ((tp = tun_find_unit(dev)) == NULL)
    335 		goto out_nolock;
    336 
    337 	ifp = &tp->tun_if;
    338 
    339 	tp->tun_flags &= ~TUN_OPEN;
    340 
    341 	/*
    342 	 * junk all pending output
    343 	 */
    344 	IFQ_PURGE(&ifp->if_snd);
    345 
    346 	if (ifp->if_flags & IFF_UP) {
    347 		if_down(ifp);
    348 		if (ifp->if_flags & IFF_RUNNING) {
    349 			/* find internet addresses and delete routes */
    350 			struct ifaddr *ifa;
    351 			IFADDR_FOREACH(ifa, ifp) {
    352 #if defined(INET) || defined(INET6)
    353 				if (ifa->ifa_addr->sa_family == AF_INET ||
    354 				    ifa->ifa_addr->sa_family == AF_INET6) {
    355 					rtinit(ifa, (int)RTM_DELETE,
    356 					       tp->tun_flags & TUN_DSTADDR
    357 							? RTF_HOST
    358 							: 0);
    359 				}
    360 #endif
    361 			}
    362 		}
    363 	}
    364 	tp->tun_pgid = 0;
    365 	selnotify(&tp->tun_rsel, 0, 0);
    366 
    367 	TUNDEBUG ("%s: closed\n", ifp->if_xname);
    368 	simple_unlock(&tp->tun_lock);
    369 out_nolock:
    370 	splx(s);
    371 	return (0);
    372 }
    373 
    374 /*
    375  * Call at splnet().
    376  */
    377 static void
    378 tuninit(struct tun_softc *tp)
    379 {
    380 	struct ifnet	*ifp = &tp->tun_if;
    381 	struct ifaddr	*ifa;
    382 
    383 	TUNDEBUG("%s: tuninit\n", ifp->if_xname);
    384 
    385 	simple_lock(&tp->tun_lock);
    386 	ifp->if_flags |= IFF_UP | IFF_RUNNING;
    387 
    388 	tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR);
    389 	IFADDR_FOREACH(ifa, ifp) {
    390 #ifdef INET
    391 		if (ifa->ifa_addr->sa_family == AF_INET) {
    392 			struct sockaddr_in *sin;
    393 
    394 			sin = satosin(ifa->ifa_addr);
    395 			if (sin && sin->sin_addr.s_addr)
    396 				tp->tun_flags |= TUN_IASET;
    397 
    398 			if (ifp->if_flags & IFF_POINTOPOINT) {
    399 				sin = satosin(ifa->ifa_dstaddr);
    400 				if (sin && sin->sin_addr.s_addr)
    401 					tp->tun_flags |= TUN_DSTADDR;
    402 			}
    403 		}
    404 #endif
    405 #ifdef INET6
    406 		if (ifa->ifa_addr->sa_family == AF_INET6) {
    407 			struct sockaddr_in6 *sin;
    408 
    409 			sin = (struct sockaddr_in6 *)ifa->ifa_addr;
    410 			if (!IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
    411 				tp->tun_flags |= TUN_IASET;
    412 
    413 			if (ifp->if_flags & IFF_POINTOPOINT) {
    414 				sin = (struct sockaddr_in6 *)ifa->ifa_dstaddr;
    415 				if (sin &&
    416 				    !IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
    417 					tp->tun_flags |= TUN_DSTADDR;
    418 			} else
    419 				tp->tun_flags &= ~TUN_DSTADDR;
    420 		}
    421 #endif /* INET6 */
    422 	}
    423 
    424 	simple_unlock(&tp->tun_lock);
    425 	return;
    426 }
    427 
    428 /*
    429  * Process an ioctl request.
    430  */
    431 static int
    432 tun_ioctl(struct ifnet *ifp, u_long cmd, void *data)
    433 {
    434 	int		error = 0, s;
    435 	struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc);
    436 	struct ifreq *ifr = data;
    437 
    438 	s = splnet();
    439 
    440 	switch (cmd) {
    441 	case SIOCINITIFADDR:
    442 		tuninit(tp);
    443 		TUNDEBUG("%s: address set\n", ifp->if_xname);
    444 		break;
    445 	case SIOCSIFDSTADDR:
    446 		tuninit(tp);
    447 		TUNDEBUG("%s: destination address set\n", ifp->if_xname);
    448 		break;
    449 	case SIOCSIFBRDADDR:
    450 		TUNDEBUG("%s: broadcast address set\n", ifp->if_xname);
    451 		break;
    452 	case SIOCSIFMTU:
    453 		if (ifr->ifr_mtu > TUNMTU || ifr->ifr_mtu < 576) {
    454 			error = EINVAL;
    455 			break;
    456 		}
    457 		TUNDEBUG("%s: interface mtu set\n", ifp->if_xname);
    458 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
    459 			error = 0;
    460 		break;
    461 	case SIOCADDMULTI:
    462 	case SIOCDELMULTI:
    463 		if (ifr == NULL) {
    464 	        	error = EAFNOSUPPORT;           /* XXX */
    465 			break;
    466 		}
    467 		switch (ifreq_getaddr(cmd, ifr)->sa_family) {
    468 #ifdef INET
    469 		case AF_INET:
    470 			break;
    471 #endif
    472 #ifdef INET6
    473 		case AF_INET6:
    474 			break;
    475 #endif
    476 		default:
    477 			error = EAFNOSUPPORT;
    478 			break;
    479 		}
    480 		break;
    481 	default:
    482 		error = ifioctl_common(ifp, cmd, data);
    483 	}
    484 
    485 	splx(s);
    486 	return (error);
    487 }
    488 
    489 /*
    490  * tun_output - queue packets from higher level ready to put out.
    491  */
    492 static int
    493 tun_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
    494     struct rtentry *rt)
    495 {
    496 	struct tun_softc *tp = ifp->if_softc;
    497 	int		s;
    498 	int		error;
    499 #if defined(INET) || defined(INET6)
    500 	int		mlen;
    501 	uint32_t	*af;
    502 #endif
    503 	ALTQ_DECL(struct altq_pktattr pktattr;)
    504 
    505 	s = splnet();
    506 	simple_lock(&tp->tun_lock);
    507 	TUNDEBUG ("%s: tun_output\n", ifp->if_xname);
    508 
    509 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
    510 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname,
    511 			  tp->tun_flags);
    512 		m_freem (m0);
    513 		error = EHOSTDOWN;
    514 		goto out;
    515 	}
    516 
    517 	/*
    518 	 * if the queueing discipline needs packet classification,
    519 	 * do it before prepending link headers.
    520 	 */
    521 	IFQ_CLASSIFY(&ifp->if_snd, m0, dst->sa_family, &pktattr);
    522 
    523 	if (ifp->if_bpf)
    524 		bpf_ops->bpf_mtap_af(ifp->if_bpf, dst->sa_family, m0);
    525 
    526 	switch(dst->sa_family) {
    527 #ifdef INET6
    528 	case AF_INET6:
    529 #endif
    530 #ifdef INET
    531 	case AF_INET:
    532 #endif
    533 #if defined(INET) || defined(INET6)
    534 		if (tp->tun_flags & TUN_PREPADDR) {
    535 			/* Simple link-layer header */
    536 			M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
    537 			if (m0 == NULL) {
    538 				IF_DROP(&ifp->if_snd);
    539 				error = ENOBUFS;
    540 				goto out;
    541 			}
    542 			bcopy(dst, mtod(m0, char *), dst->sa_len);
    543 		}
    544 
    545 		if (tp->tun_flags & TUN_IFHEAD) {
    546 			/* Prepend the address family */
    547 			M_PREPEND(m0, sizeof(*af), M_DONTWAIT);
    548 			if (m0 == NULL) {
    549 				IF_DROP(&ifp->if_snd);
    550 				error = ENOBUFS;
    551 				goto out;
    552 			}
    553 			af = mtod(m0,uint32_t *);
    554 			*af = htonl(dst->sa_family);
    555 		} else {
    556 #ifdef INET
    557 			if (dst->sa_family != AF_INET)
    558 #endif
    559 			{
    560 				m_freem(m0);
    561 				error = EAFNOSUPPORT;
    562 				goto out;
    563 			}
    564 		}
    565 		/* FALLTHROUGH */
    566 	case AF_UNSPEC:
    567 		IFQ_ENQUEUE(&ifp->if_snd, m0, &pktattr, error);
    568 		if (error) {
    569 			ifp->if_collisions++;
    570 			error = EAFNOSUPPORT;
    571 			goto out;
    572 		}
    573 		mlen = m0->m_pkthdr.len;
    574 		ifp->if_opackets++;
    575 		ifp->if_obytes += mlen;
    576 		break;
    577 #endif
    578 	default:
    579 		m_freem(m0);
    580 		error = EAFNOSUPPORT;
    581 		goto out;
    582 	}
    583 
    584 	if (tp->tun_flags & TUN_RWAIT) {
    585 		tp->tun_flags &= ~TUN_RWAIT;
    586 		wakeup((void *)tp);
    587 	}
    588 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    589 		softint_schedule(tp->tun_isih);
    590 
    591 	selnotify(&tp->tun_rsel, 0, 0);
    592 out:
    593 	simple_unlock(&tp->tun_lock);
    594 	splx(s);
    595 	return (0);
    596 }
    597 
    598 static void
    599 tun_i_softintr(void *cookie)
    600 {
    601 	struct tun_softc *tp = cookie;
    602 
    603 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    604 		fownsignal(tp->tun_pgid, SIGIO, POLL_IN, POLLIN|POLLRDNORM,
    605 		    NULL);
    606 }
    607 
    608 static void
    609 tun_o_softintr(void *cookie)
    610 {
    611 	struct tun_softc *tp = cookie;
    612 
    613 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    614 		fownsignal(tp->tun_pgid, SIGIO, POLL_OUT, POLLOUT|POLLWRNORM,
    615 		    NULL);
    616 }
    617 
    618 /*
    619  * the cdevsw interface is now pretty minimal.
    620  */
    621 int
    622 tunioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    623 {
    624 	struct tun_softc *tp;
    625 	int s, error = 0;
    626 
    627 	s = splnet();
    628 	tp = tun_find_unit(dev);
    629 
    630 	/* interface was "destroyed" already */
    631 	if (tp == NULL) {
    632 		error = ENXIO;
    633 		goto out_nolock;
    634 	}
    635 
    636 	switch (cmd) {
    637 	case TUNSDEBUG:
    638 		tundebug = *(int *)data;
    639 		break;
    640 
    641 	case TUNGDEBUG:
    642 		*(int *)data = tundebug;
    643 		break;
    644 
    645 	case TUNSIFMODE:
    646 		switch (*(int *)data & (IFF_POINTOPOINT|IFF_BROADCAST)) {
    647 		case IFF_POINTOPOINT:
    648 		case IFF_BROADCAST:
    649 			if (tp->tun_if.if_flags & IFF_UP) {
    650 				error = EBUSY;
    651 				goto out;
    652 			}
    653 			tp->tun_if.if_flags &=
    654 				~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
    655 			tp->tun_if.if_flags |= *(int *)data;
    656 			break;
    657 		default:
    658 			error = EINVAL;
    659 			goto out;
    660 		}
    661 		break;
    662 
    663 	case TUNSLMODE:
    664 		if (*(int *)data) {
    665 			tp->tun_flags |= TUN_PREPADDR;
    666 			tp->tun_flags &= ~TUN_IFHEAD;
    667 		} else
    668 			tp->tun_flags &= ~TUN_PREPADDR;
    669 		break;
    670 
    671 	case TUNSIFHEAD:
    672 		if (*(int *)data) {
    673 			tp->tun_flags |= TUN_IFHEAD;
    674 			tp->tun_flags &= ~TUN_PREPADDR;
    675 		} else
    676 			tp->tun_flags &= ~TUN_IFHEAD;
    677 		break;
    678 
    679 	case TUNGIFHEAD:
    680 		*(int *)data = (tp->tun_flags & TUN_IFHEAD);
    681 		break;
    682 
    683 	case FIONBIO:
    684 		if (*(int *)data)
    685 			tp->tun_flags |= TUN_NBIO;
    686 		else
    687 			tp->tun_flags &= ~TUN_NBIO;
    688 		break;
    689 
    690 	case FIOASYNC:
    691 		if (*(int *)data)
    692 			tp->tun_flags |= TUN_ASYNC;
    693 		else
    694 			tp->tun_flags &= ~TUN_ASYNC;
    695 		break;
    696 
    697 	case FIONREAD:
    698 		if (tp->tun_if.if_snd.ifq_head)
    699 			*(int *)data = tp->tun_if.if_snd.ifq_head->m_pkthdr.len;
    700 		else
    701 			*(int *)data = 0;
    702 		break;
    703 
    704 	case TIOCSPGRP:
    705 	case FIOSETOWN:
    706 		error = fsetown(&tp->tun_pgid, cmd, data);
    707 		break;
    708 
    709 	case TIOCGPGRP:
    710 	case FIOGETOWN:
    711 		error = fgetown(tp->tun_pgid, cmd, data);
    712 		break;
    713 
    714 	default:
    715 		error = ENOTTY;
    716 	}
    717 
    718 out:
    719 	simple_unlock(&tp->tun_lock);
    720 out_nolock:
    721 	splx(s);
    722 	return (error);
    723 }
    724 
    725 /*
    726  * The cdevsw read interface - reads a packet at a time, or at
    727  * least as much of a packet as can be read.
    728  */
    729 int
    730 tunread(dev_t dev, struct uio *uio, int ioflag)
    731 {
    732 	struct tun_softc *tp;
    733 	struct ifnet	*ifp;
    734 	struct mbuf	*m, *m0;
    735 	int		error = 0, len, s, index;
    736 
    737 	s = splnet();
    738 	tp = tun_find_unit(dev);
    739 
    740 	/* interface was "destroyed" already */
    741 	if (tp == NULL) {
    742 		error = ENXIO;
    743 		goto out_nolock;
    744 	}
    745 
    746 	index = tp->tun_if.if_index;
    747 	ifp = &tp->tun_if;
    748 
    749 	TUNDEBUG ("%s: read\n", ifp->if_xname);
    750 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
    751 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags);
    752 		error = EHOSTDOWN;
    753 		goto out;
    754 	}
    755 
    756 	tp->tun_flags &= ~TUN_RWAIT;
    757 
    758 	do {
    759 		IFQ_DEQUEUE(&ifp->if_snd, m0);
    760 		if (m0 == 0) {
    761 			if (tp->tun_flags & TUN_NBIO) {
    762 				error = EWOULDBLOCK;
    763 				goto out;
    764 			}
    765 			tp->tun_flags |= TUN_RWAIT;
    766 			if (ltsleep((void *)tp, PZERO|PCATCH|PNORELOCK,
    767 					"tunread", 0, &tp->tun_lock) != 0) {
    768 				error = EINTR;
    769 				goto out_nolock;
    770 			} else {
    771 				/*
    772 				 * Maybe the interface was destroyed while
    773 				 * we were sleeping, so let's ensure that
    774 				 * we're looking at the same (valid) tun
    775 				 * interface before looping.
    776 				 */
    777 				tp = tun_find_unit(dev);
    778 				if (tp == NULL) {
    779 					error = ENXIO;
    780 					goto out_nolock;
    781 				}
    782 				if (tp->tun_if.if_index != index) {
    783 					error = ENXIO;
    784 					goto out;
    785 				}
    786 			}
    787 		}
    788 	} while (m0 == 0);
    789 
    790 	simple_unlock(&tp->tun_lock);
    791 	splx(s);
    792 
    793 	/* Copy the mbuf chain */
    794 	while (m0 && uio->uio_resid > 0 && error == 0) {
    795 		len = min(uio->uio_resid, m0->m_len);
    796 		if (len != 0)
    797 			error = uiomove(mtod(m0, void *), len, uio);
    798 		MFREE(m0, m);
    799 		m0 = m;
    800 	}
    801 
    802 	if (m0) {
    803 		TUNDEBUG("Dropping mbuf\n");
    804 		m_freem(m0);
    805 	}
    806 	if (error)
    807 		ifp->if_ierrors++;
    808 
    809 	return (error);
    810 
    811 out:
    812 	simple_unlock(&tp->tun_lock);
    813 out_nolock:
    814 	splx(s);
    815 	return (error);
    816 }
    817 
    818 /*
    819  * the cdevsw write interface - an atomic write is a packet - or else!
    820  */
    821 int
    822 tunwrite(dev_t dev, struct uio *uio, int ioflag)
    823 {
    824 	struct tun_softc *tp;
    825 	struct ifnet	*ifp;
    826 	struct mbuf	*top, **mp, *m;
    827 	struct ifqueue	*ifq;
    828 	struct sockaddr	dst;
    829 	int		isr, error = 0, s, tlen, mlen;
    830 	uint32_t	family;
    831 
    832 	s = splnet();
    833 	tp = tun_find_unit(dev);
    834 
    835 	/* interface was "destroyed" already */
    836 	if (tp == NULL) {
    837 		error = ENXIO;
    838 		goto out_nolock;
    839 	}
    840 
    841 	/* Unlock until we've got the data */
    842 	simple_unlock(&tp->tun_lock);
    843 	splx(s);
    844 
    845 	ifp = &tp->tun_if;
    846 
    847 	TUNDEBUG("%s: tunwrite\n", ifp->if_xname);
    848 
    849 	if (tp->tun_flags & TUN_PREPADDR) {
    850 		if (uio->uio_resid < sizeof(dst)) {
    851 			error = EIO;
    852 			goto out0;
    853 		}
    854 		error = uiomove((void *)&dst, sizeof(dst), uio);
    855 		if (dst.sa_len > sizeof(dst)) {
    856 			/* Duh.. */
    857 			char discard;
    858 			int n = dst.sa_len - sizeof(dst);
    859 			while (n--)
    860 				if ((error = uiomove(&discard, 1, uio)) != 0) {
    861 					goto out0;
    862 				}
    863 		}
    864 	} else if (tp->tun_flags & TUN_IFHEAD) {
    865 		if (uio->uio_resid < sizeof(family)){
    866 			error = EIO;
    867 			goto out0;
    868 		}
    869 		error = uiomove((void *)&family, sizeof(family), uio);
    870 		dst.sa_family = ntohl(family);
    871 	} else {
    872 #ifdef INET
    873 		dst.sa_family = AF_INET;
    874 #endif
    875 	}
    876 
    877 	if (uio->uio_resid > TUNMTU) {
    878 		TUNDEBUG("%s: len=%lu!\n", ifp->if_xname,
    879 		    (unsigned long)uio->uio_resid);
    880 		error = EIO;
    881 		goto out0;
    882 	}
    883 
    884 	switch (dst.sa_family) {
    885 #ifdef INET
    886 	case AF_INET:
    887 		ifq = &ipintrq;
    888 		isr = NETISR_IP;
    889 		break;
    890 #endif
    891 #ifdef INET6
    892 	case AF_INET6:
    893 		ifq = &ip6intrq;
    894 		isr = NETISR_IPV6;
    895 		break;
    896 #endif
    897 	default:
    898 		error = EAFNOSUPPORT;
    899 		goto out0;
    900 	}
    901 
    902 	tlen = uio->uio_resid;
    903 
    904 	/* get a header mbuf */
    905 	MGETHDR(m, M_DONTWAIT, MT_DATA);
    906 	if (m == NULL) {
    907 		error = ENOBUFS;
    908 		goto out0;
    909 	}
    910 	mlen = MHLEN;
    911 
    912 	top = NULL;
    913 	mp = &top;
    914 	while (error == 0 && uio->uio_resid > 0) {
    915 		m->m_len = min(mlen, uio->uio_resid);
    916 		error = uiomove(mtod(m, void *), m->m_len, uio);
    917 		*mp = m;
    918 		mp = &m->m_next;
    919 		if (error == 0 && uio->uio_resid > 0) {
    920 			MGET(m, M_DONTWAIT, MT_DATA);
    921 			if (m == NULL) {
    922 				error = ENOBUFS;
    923 				break;
    924 			}
    925 			mlen = MLEN;
    926 		}
    927 	}
    928 	if (error) {
    929 		if (top != NULL)
    930 			m_freem (top);
    931 		ifp->if_ierrors++;
    932 		goto out0;
    933 	}
    934 
    935 	top->m_pkthdr.len = tlen;
    936 	top->m_pkthdr.rcvif = ifp;
    937 
    938 	if (ifp->if_bpf)
    939 		bpf_ops->bpf_mtap_af(ifp->if_bpf, dst.sa_family, top);
    940 
    941 	s = splnet();
    942 	simple_lock(&tp->tun_lock);
    943 	if ((tp->tun_flags & TUN_INITED) == 0) {
    944 		/* Interface was destroyed */
    945 		error = ENXIO;
    946 		goto out;
    947 	}
    948 	if (IF_QFULL(ifq)) {
    949 		IF_DROP(ifq);
    950 		ifp->if_collisions++;
    951 		m_freem(top);
    952 		error = ENOBUFS;
    953 		goto out;
    954 	}
    955 
    956 	IF_ENQUEUE(ifq, top);
    957 	ifp->if_ipackets++;
    958 	ifp->if_ibytes += tlen;
    959 	schednetisr(isr);
    960 out:
    961 	simple_unlock(&tp->tun_lock);
    962 out_nolock:
    963 	splx(s);
    964 out0:
    965 	return (error);
    966 }
    967 
    968 #ifdef ALTQ
    969 /*
    970  * Start packet transmission on the interface.
    971  * when the interface queue is rate-limited by ALTQ or TBR,
    972  * if_start is needed to drain packets from the queue in order
    973  * to notify readers when outgoing packets become ready.
    974  *
    975  * Should be called at splnet.
    976  */
    977 static void
    978 tunstart(struct ifnet *ifp)
    979 {
    980 	struct tun_softc *tp = ifp->if_softc;
    981 
    982 	if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd))
    983 		return;
    984 
    985 	simple_lock(&tp->tun_lock);
    986 	if (!IF_IS_EMPTY(&ifp->if_snd)) {
    987 		if (tp->tun_flags & TUN_RWAIT) {
    988 			tp->tun_flags &= ~TUN_RWAIT;
    989 			wakeup((void *)tp);
    990 		}
    991 		if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    992 			softint_schedule(tp->tun_osih);
    993 
    994 		selnotify(&tp->tun_rsel, 0, 0);
    995 	}
    996 	simple_unlock(&tp->tun_lock);
    997 }
    998 #endif /* ALTQ */
    999 /*
   1000  * tunpoll - the poll interface, this is only useful on reads
   1001  * really. The write detect always returns true, write never blocks
   1002  * anyway, it either accepts the packet or drops it.
   1003  */
   1004 int
   1005 tunpoll(dev_t dev, int events, struct lwp *l)
   1006 {
   1007 	struct tun_softc *tp;
   1008 	struct ifnet	*ifp;
   1009 	int		s, revents = 0;
   1010 
   1011 	s = splnet();
   1012 	tp = tun_find_unit(dev);
   1013 
   1014 	/* interface was "destroyed" already */
   1015 	if (tp == NULL)
   1016 		goto out_nolock;
   1017 
   1018 	ifp = &tp->tun_if;
   1019 
   1020 	TUNDEBUG("%s: tunpoll\n", ifp->if_xname);
   1021 
   1022 	if (events & (POLLIN | POLLRDNORM)) {
   1023 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
   1024 			TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname,
   1025 			    ifp->if_snd.ifq_len);
   1026 			revents |= events & (POLLIN | POLLRDNORM);
   1027 		} else {
   1028 			TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname);
   1029 			selrecord(l, &tp->tun_rsel);
   1030 		}
   1031 	}
   1032 
   1033 	if (events & (POLLOUT | POLLWRNORM))
   1034 		revents |= events & (POLLOUT | POLLWRNORM);
   1035 
   1036 	simple_unlock(&tp->tun_lock);
   1037 out_nolock:
   1038 	splx(s);
   1039 	return (revents);
   1040 }
   1041 
   1042 static void
   1043 filt_tunrdetach(struct knote *kn)
   1044 {
   1045 	struct tun_softc *tp = kn->kn_hook;
   1046 	int s;
   1047 
   1048 	s = splnet();
   1049 	SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext);
   1050 	splx(s);
   1051 }
   1052 
   1053 static int
   1054 filt_tunread(struct knote *kn, long hint)
   1055 {
   1056 	struct tun_softc *tp = kn->kn_hook;
   1057 	struct ifnet *ifp = &tp->tun_if;
   1058 	struct mbuf *m;
   1059 	int s;
   1060 
   1061 	s = splnet();
   1062 	IF_POLL(&ifp->if_snd, m);
   1063 	if (m == NULL) {
   1064 		splx(s);
   1065 		return (0);
   1066 	}
   1067 
   1068 	for (kn->kn_data = 0; m != NULL; m = m->m_next)
   1069 		kn->kn_data += m->m_len;
   1070 
   1071 	splx(s);
   1072 	return (1);
   1073 }
   1074 
   1075 static const struct filterops tunread_filtops =
   1076 	{ 1, NULL, filt_tunrdetach, filt_tunread };
   1077 
   1078 static const struct filterops tun_seltrue_filtops =
   1079 	{ 1, NULL, filt_tunrdetach, filt_seltrue };
   1080 
   1081 int
   1082 tunkqfilter(dev_t dev, struct knote *kn)
   1083 {
   1084 	struct tun_softc *tp;
   1085 	struct klist *klist;
   1086 	int rv = 0, s;
   1087 
   1088 	s = splnet();
   1089 	tp = tun_find_unit(dev);
   1090 	if (tp == NULL)
   1091 		goto out_nolock;
   1092 
   1093 	switch (kn->kn_filter) {
   1094 	case EVFILT_READ:
   1095 		klist = &tp->tun_rsel.sel_klist;
   1096 		kn->kn_fop = &tunread_filtops;
   1097 		break;
   1098 
   1099 	case EVFILT_WRITE:
   1100 		klist = &tp->tun_rsel.sel_klist;
   1101 		kn->kn_fop = &tun_seltrue_filtops;
   1102 		break;
   1103 
   1104 	default:
   1105 		rv = EINVAL;
   1106 		goto out;
   1107 	}
   1108 
   1109 	kn->kn_hook = tp;
   1110 
   1111 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
   1112 
   1113 out:
   1114 	simple_unlock(&tp->tun_lock);
   1115 out_nolock:
   1116 	splx(s);
   1117 	return (rv);
   1118 }
   1119