Home | History | Annotate | Line # | Download | only in net
if_tun.c revision 1.154
      1 /*	$NetBSD: if_tun.c,v 1.154 2019/03/25 09:07:24 pgoyette 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 /*
     18  * tun - tunnel software network interface.
     19  */
     20 
     21 #include <sys/cdefs.h>
     22 __KERNEL_RCSID(0, "$NetBSD: if_tun.c,v 1.154 2019/03/25 09:07:24 pgoyette Exp $");
     23 
     24 #ifdef _KERNEL_OPT
     25 #include "opt_inet.h"
     26 #endif
     27 
     28 #include <sys/param.h>
     29 
     30 #include <sys/buf.h>
     31 #include <sys/conf.h>
     32 #include <sys/cpu.h>
     33 #include <sys/device.h>
     34 #include <sys/file.h>
     35 #include <sys/ioctl.h>
     36 #include <sys/kauth.h>
     37 #include <sys/kmem.h>
     38 #include <sys/lwp.h>
     39 #include <sys/mbuf.h>
     40 #include <sys/module.h>
     41 #include <sys/mutex.h>
     42 #include <sys/poll.h>
     43 #include <sys/select.h>
     44 #include <sys/signalvar.h>
     45 #include <sys/socket.h>
     46 
     47 #include <net/bpf.h>
     48 #include <net/if.h>
     49 #include <net/if_types.h>
     50 #include <net/route.h>
     51 
     52 #ifdef INET
     53 #include <netinet/in.h>
     54 #include <netinet/in_systm.h>
     55 #include <netinet/in_var.h>
     56 #include <netinet/ip.h>
     57 #include <netinet/if_inarp.h>
     58 #endif
     59 
     60 #include <net/if_tun.h>
     61 
     62 #include "ioconf.h"
     63 
     64 #define TUNDEBUG	if (tundebug) printf
     65 int	tundebug = 0;
     66 
     67 extern int ifqmaxlen;
     68 
     69 static LIST_HEAD(, tun_softc) tun_softc_list;
     70 static LIST_HEAD(, tun_softc) tunz_softc_list;
     71 static kmutex_t tun_softc_lock;
     72 
     73 static int	tun_ioctl(struct ifnet *, u_long, void *);
     74 static int	tun_output(struct ifnet *, struct mbuf *,
     75 			const struct sockaddr *, const struct rtentry *rt);
     76 static int	tun_clone_create(struct if_clone *, int);
     77 static int	tun_clone_destroy(struct ifnet *);
     78 
     79 static struct if_clone tun_cloner =
     80     IF_CLONE_INITIALIZER("tun", tun_clone_create, tun_clone_destroy);
     81 
     82 static void tunattach0(struct tun_softc *);
     83 static void tun_enable(struct tun_softc *, const struct ifaddr *);
     84 static void tun_i_softintr(void *);
     85 static void tun_o_softintr(void *);
     86 #ifdef ALTQ
     87 static void tunstart(struct ifnet *);
     88 #endif
     89 static struct tun_softc *tun_find_unit(dev_t);
     90 static struct tun_softc *tun_find_zunit(int);
     91 
     92 static dev_type_open(tunopen);
     93 static dev_type_close(tunclose);
     94 static dev_type_read(tunread);
     95 static dev_type_write(tunwrite);
     96 static dev_type_ioctl(tunioctl);
     97 static dev_type_poll(tunpoll);
     98 static dev_type_kqfilter(tunkqfilter);
     99 
    100 const struct cdevsw tun_cdevsw = {
    101 	.d_open = tunopen,
    102 	.d_close = tunclose,
    103 	.d_read = tunread,
    104 	.d_write = tunwrite,
    105 	.d_ioctl = tunioctl,
    106 	.d_stop = nostop,
    107 	.d_tty = notty,
    108 	.d_poll = tunpoll,
    109 	.d_mmap = nommap,
    110 	.d_kqfilter = tunkqfilter,
    111 	.d_discard = nodiscard,
    112 	.d_flag = D_OTHER | D_MPSAFE
    113 };
    114 
    115 #ifdef _MODULE
    116 devmajor_t tun_bmajor = -1, tun_cmajor = -1;
    117 #endif
    118 
    119 void
    120 tunattach(int unused)
    121 {
    122 
    123 	/*
    124 	 * Nothing to do here, initialization is handled by the
    125 	 * module initialization code in tuninit() below).
    126 	 */
    127 }
    128 
    129 static void
    130 tuninit(void)
    131 {
    132 
    133 	mutex_init(&tun_softc_lock, MUTEX_DEFAULT, IPL_NET);
    134 	LIST_INIT(&tun_softc_list);
    135 	LIST_INIT(&tunz_softc_list);
    136 	if_clone_attach(&tun_cloner);
    137 #ifdef _MODULE
    138 	devsw_attach("tun", NULL, &tun_bmajor, &tun_cdevsw, &tun_cmajor);
    139 #endif
    140 }
    141 
    142 static int
    143 tundetach(void)
    144 {
    145 	int error = 0;
    146 
    147 	if_clone_detach(&tun_cloner);
    148 #ifdef _MODULE
    149 	error = devsw_detach(NULL, &tun_cdevsw);
    150 	if (error != 0) {
    151 		if_clone_attach(&tun_cloner);
    152 		return error;
    153 	}
    154 #endif
    155 
    156 	if (!LIST_EMPTY(&tun_softc_list) || !LIST_EMPTY(&tunz_softc_list)) {
    157 #ifdef _MODULE
    158 		devsw_attach("tun", NULL, &tun_bmajor, &tun_cdevsw, &tun_cmajor);
    159 #endif
    160 		if_clone_attach(&tun_cloner);
    161 		return EBUSY;
    162 }
    163 
    164 	mutex_destroy(&tun_softc_lock);
    165 
    166 	return 0;
    167 }
    168 
    169 /*
    170  * Find driver instance from dev_t.
    171  * Returns with tp locked (if found).
    172  */
    173 static struct tun_softc *
    174 tun_find_unit(dev_t dev)
    175 {
    176 	struct tun_softc *tp;
    177 	int unit = minor(dev);
    178 
    179 	mutex_enter(&tun_softc_lock);
    180 	LIST_FOREACH(tp, &tun_softc_list, tun_list)
    181 		if (unit == tp->tun_unit)
    182 			break;
    183 	if (tp)
    184 		mutex_enter(&tp->tun_lock);
    185 	mutex_exit(&tun_softc_lock);
    186 
    187 	return tp;
    188 }
    189 
    190 /*
    191  * Find zombie driver instance by unit number.
    192  * Remove tp from list and return it unlocked (if found).
    193  */
    194 static struct tun_softc *
    195 tun_find_zunit(int unit)
    196 {
    197 	struct tun_softc *tp;
    198 
    199 	mutex_enter(&tun_softc_lock);
    200 	LIST_FOREACH(tp, &tunz_softc_list, tun_list)
    201 		if (unit == tp->tun_unit)
    202 			break;
    203 	if (tp)
    204 		LIST_REMOVE(tp, tun_list);
    205 	mutex_exit(&tun_softc_lock);
    206 	KASSERTMSG(!tp || (tp->tun_flags & (TUN_INITED|TUN_OPEN)) == TUN_OPEN,
    207 	    "tun%d: inconsistent flags: %x", unit, tp->tun_flags);
    208 
    209 	return tp;
    210 }
    211 
    212 static int
    213 tun_clone_create(struct if_clone *ifc, int unit)
    214 {
    215 	struct tun_softc *tp;
    216 
    217 	if ((tp = tun_find_zunit(unit)) == NULL) {
    218 		tp = kmem_zalloc(sizeof(*tp), KM_SLEEP);
    219 
    220 		tp->tun_unit = unit;
    221 		mutex_init(&tp->tun_lock, MUTEX_DEFAULT, IPL_NET);
    222 		cv_init(&tp->tun_cv, "tunread");
    223 		selinit(&tp->tun_rsel);
    224 		selinit(&tp->tun_wsel);
    225 	} else {
    226 		/* Revive tunnel instance; clear ifp part */
    227 		(void)memset(&tp->tun_if, 0, sizeof(struct ifnet));
    228 	}
    229 
    230 	if_initname(&tp->tun_if, ifc->ifc_name, unit);
    231 	tunattach0(tp);
    232 	tp->tun_flags |= TUN_INITED;
    233 	tp->tun_osih = softint_establish(SOFTINT_CLOCK, tun_o_softintr, tp);
    234 	tp->tun_isih = softint_establish(SOFTINT_CLOCK, tun_i_softintr, tp);
    235 
    236 	mutex_enter(&tun_softc_lock);
    237 	LIST_INSERT_HEAD(&tun_softc_list, tp, tun_list);
    238 	mutex_exit(&tun_softc_lock);
    239 
    240 	return 0;
    241 }
    242 
    243 static void
    244 tunattach0(struct tun_softc *tp)
    245 {
    246 	struct ifnet *ifp;
    247 
    248 	ifp = &tp->tun_if;
    249 	ifp->if_softc = tp;
    250 	ifp->if_mtu = TUNMTU;
    251 	ifp->if_ioctl = tun_ioctl;
    252 	ifp->if_output = tun_output;
    253 #ifdef ALTQ
    254 	ifp->if_start = tunstart;
    255 #endif
    256 	ifp->if_flags = IFF_POINTOPOINT;
    257 	ifp->if_extflags = IFEF_NO_LINK_STATE_CHANGE;
    258 	ifp->if_type = IFT_TUNNEL;
    259 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
    260 	ifp->if_collisions = 0;
    261 	ifp->if_ierrors = 0;
    262 	ifp->if_oerrors = 0;
    263 	ifp->if_ipackets = 0;
    264 	ifp->if_opackets = 0;
    265 	ifp->if_ibytes   = 0;
    266 	ifp->if_obytes   = 0;
    267 	ifp->if_dlt = DLT_NULL;
    268 	IFQ_SET_READY(&ifp->if_snd);
    269 	if_attach(ifp);
    270 	if_alloc_sadl(ifp);
    271 	bpf_attach(ifp, DLT_NULL, sizeof(uint32_t));
    272 }
    273 
    274 static int
    275 tun_clone_destroy(struct ifnet *ifp)
    276 {
    277 	struct tun_softc *tp = (void *)ifp;
    278 	bool zombie = false;
    279 
    280 	IF_PURGE(&ifp->if_snd);
    281 	ifp->if_flags &= ~IFF_RUNNING;
    282 
    283 	mutex_enter(&tun_softc_lock);
    284 	mutex_enter(&tp->tun_lock);
    285 	LIST_REMOVE(tp, tun_list);
    286 	if (tp->tun_flags & TUN_OPEN) {
    287 		/* Hang on to storage until last close. */
    288 		tp->tun_flags &= ~TUN_INITED;
    289 		LIST_INSERT_HEAD(&tunz_softc_list, tp, tun_list);
    290 		zombie = true;
    291 	}
    292 	mutex_exit(&tun_softc_lock);
    293 
    294 	if (tp->tun_flags & TUN_RWAIT) {
    295 		tp->tun_flags &= ~TUN_RWAIT;
    296 		cv_broadcast(&tp->tun_cv);
    297 	}
    298 	selnotify(&tp->tun_rsel, 0, NOTE_SUBMIT);
    299 
    300 	mutex_exit(&tp->tun_lock);
    301 
    302 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    303 		fownsignal(tp->tun_pgid, SIGIO, POLL_HUP, 0, NULL);
    304 
    305 	bpf_detach(ifp);
    306 	if_detach(ifp);
    307 
    308 	if (!zombie) {
    309 		seldestroy(&tp->tun_rsel);
    310 		seldestroy(&tp->tun_wsel);
    311 		softint_disestablish(tp->tun_osih);
    312 		softint_disestablish(tp->tun_isih);
    313 		mutex_destroy(&tp->tun_lock);
    314 		cv_destroy(&tp->tun_cv);
    315 		kmem_free(tp, sizeof(*tp));
    316 	}
    317 
    318 	return 0;
    319 }
    320 
    321 /*
    322  * tunnel open - must be superuser & the device must be
    323  * configured in
    324  */
    325 static int
    326 tunopen(dev_t dev, int flag, int mode, struct lwp *l)
    327 {
    328 	struct ifnet	*ifp;
    329 	struct tun_softc *tp;
    330 	int	error;
    331 
    332 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE_TUN,
    333 	    KAUTH_REQ_NETWORK_INTERFACE_TUN_ADD, NULL, NULL, NULL);
    334 	if (error)
    335 		return error;
    336 
    337 	tp = tun_find_unit(dev);
    338 
    339 	if (tp == NULL) {
    340 		(void)tun_clone_create(&tun_cloner, minor(dev));
    341 		tp = tun_find_unit(dev);
    342 		if (tp == NULL) {
    343 			return ENXIO;
    344 		}
    345 	}
    346 
    347 	if (tp->tun_flags & TUN_OPEN) {
    348 		mutex_exit(&tp->tun_lock);
    349 		return EBUSY;
    350 	}
    351 
    352 	ifp = &tp->tun_if;
    353 	tp->tun_flags |= TUN_OPEN;
    354 	TUNDEBUG("%s: open\n", ifp->if_xname);
    355 
    356 	mutex_exit(&tp->tun_lock);
    357 
    358 	return error;
    359 }
    360 
    361 /*
    362  * tunclose - close the device - mark i/f down & delete
    363  * routing info
    364  */
    365 int
    366 tunclose(dev_t dev, int flag, int mode,
    367     struct lwp *l)
    368 {
    369 	struct tun_softc *tp;
    370 	struct ifnet	*ifp;
    371 
    372 	if ((tp = tun_find_zunit(minor(dev))) != NULL) {
    373 		/* interface was "destroyed" before the close */
    374 		seldestroy(&tp->tun_rsel);
    375 		seldestroy(&tp->tun_wsel);
    376 		softint_disestablish(tp->tun_osih);
    377 		softint_disestablish(tp->tun_isih);
    378 		mutex_destroy(&tp->tun_lock);
    379 		kmem_free(tp, sizeof(*tp));
    380 		return 0;
    381 	}
    382 
    383 	if ((tp = tun_find_unit(dev)) == NULL)
    384 		goto out_nolock;
    385 
    386 	ifp = &tp->tun_if;
    387 
    388 	tp->tun_flags &= ~TUN_OPEN;
    389 
    390 	tp->tun_pgid = 0;
    391 	selnotify(&tp->tun_rsel, 0, NOTE_SUBMIT);
    392 
    393 	TUNDEBUG ("%s: closed\n", ifp->if_xname);
    394 	mutex_exit(&tp->tun_lock);
    395 
    396 	/*
    397 	 * junk all pending output
    398 	 */
    399 	IFQ_PURGE(&ifp->if_snd);
    400 
    401 	if (ifp->if_flags & IFF_UP) {
    402 		if_down(ifp);
    403 		if (ifp->if_flags & IFF_RUNNING) {
    404 			/* find internet addresses and delete routes */
    405 			struct ifaddr *ifa;
    406 			IFADDR_READER_FOREACH(ifa, ifp) {
    407 #if defined(INET) || defined(INET6)
    408 				if (ifa->ifa_addr->sa_family == AF_INET ||
    409 				    ifa->ifa_addr->sa_family == AF_INET6) {
    410 					rtinit(ifa, (int)RTM_DELETE,
    411 					       tp->tun_flags & TUN_DSTADDR
    412 							? RTF_HOST
    413 							: 0);
    414 				}
    415 #endif
    416 			}
    417 		}
    418 	}
    419 out_nolock:
    420 	return 0;
    421 }
    422 
    423 static void
    424 tun_enable(struct tun_softc *tp, const struct ifaddr *ifa)
    425 {
    426 	struct ifnet	*ifp = &tp->tun_if;
    427 
    428 	TUNDEBUG("%s: %s\n", __func__, ifp->if_xname);
    429 
    430 	mutex_enter(&tp->tun_lock);
    431 	tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR);
    432 
    433 	switch (ifa->ifa_addr->sa_family) {
    434 #ifdef INET
    435 	case AF_INET: {
    436 		struct sockaddr_in *sin;
    437 
    438 		sin = satosin(ifa->ifa_addr);
    439 		if (sin && sin->sin_addr.s_addr)
    440 			tp->tun_flags |= TUN_IASET;
    441 
    442 		if (ifp->if_flags & IFF_POINTOPOINT) {
    443 			sin = satosin(ifa->ifa_dstaddr);
    444 			if (sin && sin->sin_addr.s_addr)
    445 				tp->tun_flags |= TUN_DSTADDR;
    446 		}
    447 		break;
    448 	    }
    449 #endif
    450 #ifdef INET6
    451 	case AF_INET6: {
    452 		struct sockaddr_in6 *sin;
    453 
    454 		sin = satosin6(ifa->ifa_addr);
    455 		if (!IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
    456 			tp->tun_flags |= TUN_IASET;
    457 
    458 		if (ifp->if_flags & IFF_POINTOPOINT) {
    459 			sin = satosin6(ifa->ifa_dstaddr);
    460 			if (sin && !IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
    461 				tp->tun_flags |= TUN_DSTADDR;
    462 		} else
    463 			tp->tun_flags &= ~TUN_DSTADDR;
    464 		break;
    465 	    }
    466 #endif /* INET6 */
    467 	default:
    468 		break;
    469 	}
    470 	ifp->if_flags |= IFF_UP | IFF_RUNNING;
    471 	mutex_exit(&tp->tun_lock);
    472 }
    473 
    474 /*
    475  * Process an ioctl request.
    476  */
    477 static int
    478 tun_ioctl(struct ifnet *ifp, u_long cmd, void *data)
    479 {
    480 	struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc);
    481 	struct ifreq *ifr = (struct ifreq *)data;
    482 	struct ifaddr *ifa = (struct ifaddr *)data;
    483 	int error = 0;
    484 
    485 	switch (cmd) {
    486 	case SIOCINITIFADDR:
    487 		tun_enable(tp, ifa);
    488 		ifa->ifa_rtrequest = p2p_rtrequest;
    489 		TUNDEBUG("%s: address set\n", ifp->if_xname);
    490 		break;
    491 	case SIOCSIFBRDADDR:
    492 		TUNDEBUG("%s: broadcast address set\n", ifp->if_xname);
    493 		break;
    494 	case SIOCSIFMTU:
    495 		if (ifr->ifr_mtu > TUNMTU || ifr->ifr_mtu < 576) {
    496 			error = EINVAL;
    497 			break;
    498 		}
    499 		TUNDEBUG("%s: interface mtu set\n", ifp->if_xname);
    500 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
    501 			error = 0;
    502 		break;
    503 	case SIOCADDMULTI:
    504 	case SIOCDELMULTI:
    505 		if (ifr == NULL) {
    506 	        	error = EAFNOSUPPORT;           /* XXX */
    507 			break;
    508 		}
    509 		switch (ifreq_getaddr(cmd, ifr)->sa_family) {
    510 #ifdef INET
    511 		case AF_INET:
    512 			break;
    513 #endif
    514 #ifdef INET6
    515 		case AF_INET6:
    516 			break;
    517 #endif
    518 		default:
    519 			error = EAFNOSUPPORT;
    520 			break;
    521 		}
    522 		break;
    523 	default:
    524 		error = ifioctl_common(ifp, cmd, data);
    525 	}
    526 
    527 	return error;
    528 }
    529 
    530 /*
    531  * tun_output - queue packets from higher level ready to put out.
    532  */
    533 static int
    534 tun_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
    535     const struct rtentry *rt)
    536 {
    537 	struct tun_softc *tp = ifp->if_softc;
    538 	int		error;
    539 #if defined(INET) || defined(INET6)
    540 	int		mlen;
    541 	uint32_t	*af;
    542 #endif
    543 
    544 	mutex_enter(&tp->tun_lock);
    545 	TUNDEBUG ("%s: tun_output\n", ifp->if_xname);
    546 
    547 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
    548 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname,
    549 			  tp->tun_flags);
    550 		error = EHOSTDOWN;
    551 		mutex_exit(&tp->tun_lock);
    552 		goto out;
    553 	}
    554 	// XXXrmind
    555 	mutex_exit(&tp->tun_lock);
    556 
    557 	/*
    558 	 * if the queueing discipline needs packet classification,
    559 	 * do it before prepending link headers.
    560 	 */
    561 	IFQ_CLASSIFY(&ifp->if_snd, m0, dst->sa_family);
    562 
    563 	bpf_mtap_af(ifp, dst->sa_family, m0, BPF_D_OUT);
    564 
    565 	if ((error = pfil_run_hooks(ifp->if_pfil, &m0, ifp, PFIL_OUT)) != 0)
    566 		goto out;
    567 	if (m0 == NULL)
    568 		goto out;
    569 
    570 	switch(dst->sa_family) {
    571 #ifdef INET6
    572 	case AF_INET6:
    573 #endif
    574 #ifdef INET
    575 	case AF_INET:
    576 #endif
    577 #if defined(INET) || defined(INET6)
    578 		if (tp->tun_flags & TUN_PREPADDR) {
    579 			/* Simple link-layer header */
    580 			M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
    581 			if (m0 == NULL) {
    582 				IF_DROP(&ifp->if_snd);
    583 				error = ENOBUFS;
    584 				goto out;
    585 			}
    586 			memcpy(mtod(m0, char *), dst, dst->sa_len);
    587 		}
    588 
    589 		if (tp->tun_flags & TUN_IFHEAD) {
    590 			/* Prepend the address family */
    591 			M_PREPEND(m0, sizeof(*af), M_DONTWAIT);
    592 			if (m0 == NULL) {
    593 				IF_DROP(&ifp->if_snd);
    594 				error = ENOBUFS;
    595 				goto out;
    596 			}
    597 			af = mtod(m0,uint32_t *);
    598 			*af = htonl(dst->sa_family);
    599 		} else {
    600 #ifdef INET
    601 			if (dst->sa_family != AF_INET)
    602 #endif
    603 			{
    604 				error = EAFNOSUPPORT;
    605 				goto out;
    606 			}
    607 		}
    608 		/* FALLTHROUGH */
    609 	case AF_UNSPEC:
    610 		IFQ_ENQUEUE(&ifp->if_snd, m0, error);
    611 		if (error) {
    612 			ifp->if_collisions++;
    613 			error = EAFNOSUPPORT;
    614 			m0 = NULL;
    615 			goto out;
    616 		}
    617 		mlen = m0->m_pkthdr.len;
    618 		ifp->if_opackets++;
    619 		ifp->if_obytes += mlen;
    620 		break;
    621 #endif
    622 	default:
    623 		error = EAFNOSUPPORT;
    624 		goto out;
    625 	}
    626 
    627 	mutex_enter(&tp->tun_lock);
    628 	if (tp->tun_flags & TUN_RWAIT) {
    629 		tp->tun_flags &= ~TUN_RWAIT;
    630 		cv_broadcast(&tp->tun_cv);
    631 	}
    632 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    633 		softint_schedule(tp->tun_isih);
    634 
    635 	selnotify(&tp->tun_rsel, 0, NOTE_SUBMIT);
    636 
    637 	mutex_exit(&tp->tun_lock);
    638 out:
    639 	if (error && m0)
    640 		m_freem(m0);
    641 
    642 	return error;
    643 }
    644 
    645 static void
    646 tun_i_softintr(void *cookie)
    647 {
    648 	struct tun_softc *tp = cookie;
    649 
    650 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    651 		fownsignal(tp->tun_pgid, SIGIO, POLL_IN, POLLIN|POLLRDNORM,
    652 		    NULL);
    653 }
    654 
    655 static void
    656 tun_o_softintr(void *cookie)
    657 {
    658 	struct tun_softc *tp = cookie;
    659 
    660 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    661 		fownsignal(tp->tun_pgid, SIGIO, POLL_OUT, POLLOUT|POLLWRNORM,
    662 		    NULL);
    663 }
    664 
    665 /*
    666  * the cdevsw interface is now pretty minimal.
    667  */
    668 int
    669 tunioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    670 {
    671 	struct tun_softc *tp;
    672 	int error = 0;
    673 
    674 	tp = tun_find_unit(dev);
    675 
    676 	/* interface was "destroyed" already */
    677 	if (tp == NULL) {
    678 		return ENXIO;
    679 	}
    680 
    681 	switch (cmd) {
    682 	case TUNSDEBUG:
    683 		tundebug = *(int *)data;
    684 		break;
    685 
    686 	case TUNGDEBUG:
    687 		*(int *)data = tundebug;
    688 		break;
    689 
    690 	case TUNSIFMODE:
    691 		switch (*(int *)data & (IFF_POINTOPOINT|IFF_BROADCAST)) {
    692 		case IFF_POINTOPOINT:
    693 		case IFF_BROADCAST:
    694 			if (tp->tun_if.if_flags & IFF_UP) {
    695 				error = EBUSY;
    696 				goto out;
    697 			}
    698 			tp->tun_if.if_flags &=
    699 				~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
    700 			tp->tun_if.if_flags |= *(int *)data;
    701 			break;
    702 		default:
    703 			error = EINVAL;
    704 			goto out;
    705 		}
    706 		break;
    707 
    708 	case TUNSLMODE:
    709 		if (*(int *)data) {
    710 			tp->tun_flags |= TUN_PREPADDR;
    711 			tp->tun_flags &= ~TUN_IFHEAD;
    712 		} else
    713 			tp->tun_flags &= ~TUN_PREPADDR;
    714 		break;
    715 
    716 	case TUNSIFHEAD:
    717 		if (*(int *)data) {
    718 			tp->tun_flags |= TUN_IFHEAD;
    719 			tp->tun_flags &= ~TUN_PREPADDR;
    720 		} else
    721 			tp->tun_flags &= ~TUN_IFHEAD;
    722 		break;
    723 
    724 	case TUNGIFHEAD:
    725 		*(int *)data = (tp->tun_flags & TUN_IFHEAD);
    726 		break;
    727 
    728 	case FIONBIO:
    729 		if (*(int *)data)
    730 			tp->tun_flags |= TUN_NBIO;
    731 		else
    732 			tp->tun_flags &= ~TUN_NBIO;
    733 		break;
    734 
    735 	case FIOASYNC:
    736 		if (*(int *)data)
    737 			tp->tun_flags |= TUN_ASYNC;
    738 		else
    739 			tp->tun_flags &= ~TUN_ASYNC;
    740 		break;
    741 
    742 	case FIONREAD:
    743 		if (tp->tun_if.if_snd.ifq_head)
    744 			*(int *)data = tp->tun_if.if_snd.ifq_head->m_pkthdr.len;
    745 		else
    746 			*(int *)data = 0;
    747 		break;
    748 
    749 	case TIOCSPGRP:
    750 	case FIOSETOWN:
    751 		error = fsetown(&tp->tun_pgid, cmd, data);
    752 		break;
    753 
    754 	case TIOCGPGRP:
    755 	case FIOGETOWN:
    756 		error = fgetown(tp->tun_pgid, cmd, data);
    757 		break;
    758 
    759 	default:
    760 		error = ENOTTY;
    761 	}
    762 
    763 out:
    764 	mutex_exit(&tp->tun_lock);
    765 
    766 	return error;
    767 }
    768 
    769 /*
    770  * The cdevsw read interface - reads a packet at a time, or at
    771  * least as much of a packet as can be read.
    772  */
    773 int
    774 tunread(dev_t dev, struct uio *uio, int ioflag)
    775 {
    776 	struct tun_softc *tp;
    777 	struct ifnet	*ifp;
    778 	struct mbuf	*m, *m0;
    779 	int		error = 0, len;
    780 
    781 	tp = tun_find_unit(dev);
    782 
    783 	/* interface was "destroyed" already */
    784 	if (tp == NULL) {
    785 		return ENXIO;
    786 	}
    787 
    788 	ifp = &tp->tun_if;
    789 
    790 	TUNDEBUG ("%s: read\n", ifp->if_xname);
    791 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
    792 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags);
    793 		error = EHOSTDOWN;
    794 		goto out;
    795 	}
    796 
    797 	tp->tun_flags &= ~TUN_RWAIT;
    798 
    799 	do {
    800 		IFQ_DEQUEUE(&ifp->if_snd, m0);
    801 		if (m0 == 0) {
    802 			if (tp->tun_flags & TUN_NBIO) {
    803 				error = EWOULDBLOCK;
    804 				goto out;
    805 			}
    806 			tp->tun_flags |= TUN_RWAIT;
    807 			if (cv_wait_sig(&tp->tun_cv, &tp->tun_lock)) {
    808 				error = EINTR;
    809 				goto out;
    810 			}
    811 		}
    812 	} while (m0 == 0);
    813 
    814 	mutex_exit(&tp->tun_lock);
    815 
    816 	/* Copy the mbuf chain */
    817 	while (m0 && uio->uio_resid > 0 && error == 0) {
    818 		len = uimin(uio->uio_resid, m0->m_len);
    819 		if (len != 0)
    820 			error = uiomove(mtod(m0, void *), len, uio);
    821 		m0 = m = m_free(m0);
    822 	}
    823 
    824 	if (m0) {
    825 		TUNDEBUG("Dropping mbuf\n");
    826 		m_freem(m0);
    827 	}
    828 	if (error)
    829 		ifp->if_ierrors++;
    830 
    831 	return error;
    832 
    833 out:
    834 	mutex_exit(&tp->tun_lock);
    835 
    836 	return error;
    837 }
    838 
    839 /*
    840  * the cdevsw write interface - an atomic write is a packet - or else!
    841  */
    842 int
    843 tunwrite(dev_t dev, struct uio *uio, int ioflag)
    844 {
    845 	struct tun_softc *tp;
    846 	struct ifnet	*ifp;
    847 	struct mbuf	*top, **mp, *m;
    848 	pktqueue_t	*pktq;
    849 	struct sockaddr	dst;
    850 	int		error = 0, tlen, mlen;
    851 	uint32_t	family;
    852 
    853 	tp = tun_find_unit(dev);
    854 	if (tp == NULL) {
    855 		/* Interface was "destroyed" already. */
    856 		return ENXIO;
    857 	}
    858 
    859 	/* Unlock until we've got the data */
    860 	mutex_exit(&tp->tun_lock);
    861 
    862 	ifp = &tp->tun_if;
    863 
    864 	TUNDEBUG("%s: tunwrite\n", ifp->if_xname);
    865 
    866 	if (tp->tun_flags & TUN_PREPADDR) {
    867 		if (uio->uio_resid < sizeof(dst)) {
    868 			error = EIO;
    869 			goto out0;
    870 		}
    871 		error = uiomove((void *)&dst, sizeof(dst), uio);
    872 		if (dst.sa_len > sizeof(dst)) {
    873 			/* Duh.. */
    874 			int n = dst.sa_len - sizeof(dst);
    875 			while (n--) {
    876 				char discard;
    877 				error = uiomove(&discard, 1, uio);
    878 				if (error) {
    879 					goto out0;
    880 				}
    881 			}
    882 		}
    883 	} else if (tp->tun_flags & TUN_IFHEAD) {
    884 		if (uio->uio_resid < sizeof(family)){
    885 			error = EIO;
    886 			goto out0;
    887 		}
    888 		error = uiomove((void *)&family, sizeof(family), uio);
    889 		dst.sa_family = ntohl(family);
    890 	} else {
    891 #ifdef INET
    892 		dst.sa_family = AF_INET;
    893 #endif
    894 	}
    895 
    896 	if (uio->uio_resid > TUNMTU) {
    897 		TUNDEBUG("%s: len=%lu!\n", ifp->if_xname,
    898 		    (unsigned long)uio->uio_resid);
    899 		error = EIO;
    900 		goto out0;
    901 	}
    902 
    903 	switch (dst.sa_family) {
    904 #ifdef INET
    905 	case AF_INET:
    906 		pktq = ip_pktq;
    907 		break;
    908 #endif
    909 #ifdef INET6
    910 	case AF_INET6:
    911 		pktq = ip6_pktq;
    912 		break;
    913 #endif
    914 	default:
    915 		error = EAFNOSUPPORT;
    916 		goto out0;
    917 	}
    918 
    919 	tlen = uio->uio_resid;
    920 
    921 	/* get a header mbuf */
    922 	MGETHDR(m, M_DONTWAIT, MT_DATA);
    923 	if (m == NULL) {
    924 		return ENOBUFS;
    925 	}
    926 	mlen = MHLEN;
    927 
    928 	top = NULL;
    929 	mp = &top;
    930 	while (error == 0 && uio->uio_resid > 0) {
    931 		m->m_len = uimin(mlen, uio->uio_resid);
    932 		error = uiomove(mtod(m, void *), m->m_len, uio);
    933 		*mp = m;
    934 		mp = &m->m_next;
    935 		if (error == 0 && uio->uio_resid > 0) {
    936 			MGET(m, M_DONTWAIT, MT_DATA);
    937 			if (m == NULL) {
    938 				error = ENOBUFS;
    939 				break;
    940 			}
    941 			mlen = MLEN;
    942 		}
    943 	}
    944 	if (error) {
    945 		if (top != NULL)
    946 			m_freem(top);
    947 		ifp->if_ierrors++;
    948 		goto out0;
    949 	}
    950 
    951 	top->m_pkthdr.len = tlen;
    952 	m_set_rcvif(top, ifp);
    953 
    954 	bpf_mtap_af(ifp, dst.sa_family, top, BPF_D_IN);
    955 
    956 	if ((error = pfil_run_hooks(ifp->if_pfil, &top, ifp, PFIL_IN)) != 0)
    957 		goto out0;
    958 	if (top == NULL)
    959 		goto out0;
    960 
    961 	mutex_enter(&tp->tun_lock);
    962 	if ((tp->tun_flags & TUN_INITED) == 0) {
    963 		/* Interface was destroyed */
    964 		error = ENXIO;
    965 		goto out;
    966 	}
    967 	if (__predict_false(!pktq_enqueue(pktq, top, 0))) {
    968 		ifp->if_collisions++;
    969 		mutex_exit(&tp->tun_lock);
    970 		error = ENOBUFS;
    971 		m_freem(top);
    972 		goto out0;
    973 	}
    974 	ifp->if_ipackets++;
    975 	ifp->if_ibytes += tlen;
    976 out:
    977 	mutex_exit(&tp->tun_lock);
    978 out0:
    979 	return error;
    980 }
    981 
    982 #ifdef ALTQ
    983 /*
    984  * Start packet transmission on the interface.
    985  * when the interface queue is rate-limited by ALTQ or TBR,
    986  * if_start is needed to drain packets from the queue in order
    987  * to notify readers when outgoing packets become ready.
    988  */
    989 static void
    990 tunstart(struct ifnet *ifp)
    991 {
    992 	struct tun_softc *tp = ifp->if_softc;
    993 
    994 	if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd))
    995 		return;
    996 
    997 	mutex_enter(&tp->tun_lock);
    998 	if (!IF_IS_EMPTY(&ifp->if_snd)) {
    999 		if (tp->tun_flags & TUN_RWAIT) {
   1000 			tp->tun_flags &= ~TUN_RWAIT;
   1001 			cv_broadcast(&tp->tun_cv);
   1002 		}
   1003 		if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
   1004 			softint_schedule(tp->tun_osih);
   1005 
   1006 		selnotify(&tp->tun_rsel, 0, NOTE_SUBMIT);
   1007 	}
   1008 	mutex_exit(&tp->tun_lock);
   1009 }
   1010 #endif /* ALTQ */
   1011 /*
   1012  * tunpoll - the poll interface, this is only useful on reads
   1013  * really. The write detect always returns true, write never blocks
   1014  * anyway, it either accepts the packet or drops it.
   1015  */
   1016 int
   1017 tunpoll(dev_t dev, int events, struct lwp *l)
   1018 {
   1019 	struct tun_softc *tp;
   1020 	struct ifnet	*ifp;
   1021 	int revents = 0;
   1022 
   1023 	tp = tun_find_unit(dev);
   1024 	if (tp == NULL) {
   1025 		/* Interface was "destroyed" already. */
   1026 		return 0;
   1027 	}
   1028 	ifp = &tp->tun_if;
   1029 
   1030 	TUNDEBUG("%s: tunpoll\n", ifp->if_xname);
   1031 
   1032 	if (events & (POLLIN | POLLRDNORM)) {
   1033 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
   1034 			TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname,
   1035 			    ifp->if_snd.ifq_len);
   1036 			revents |= events & (POLLIN | POLLRDNORM);
   1037 		} else {
   1038 			TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname);
   1039 			selrecord(l, &tp->tun_rsel);
   1040 		}
   1041 	}
   1042 
   1043 	if (events & (POLLOUT | POLLWRNORM))
   1044 		revents |= events & (POLLOUT | POLLWRNORM);
   1045 
   1046 	mutex_exit(&tp->tun_lock);
   1047 
   1048 	return revents;
   1049 }
   1050 
   1051 static void
   1052 filt_tunrdetach(struct knote *kn)
   1053 {
   1054 	struct tun_softc *tp = kn->kn_hook;
   1055 
   1056 	mutex_enter(&tp->tun_lock);
   1057 	SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext);
   1058 	mutex_exit(&tp->tun_lock);
   1059 }
   1060 
   1061 static int
   1062 filt_tunread(struct knote *kn, long hint)
   1063 {
   1064 	struct tun_softc *tp = kn->kn_hook;
   1065 	struct ifnet *ifp = &tp->tun_if;
   1066 	struct mbuf *m;
   1067 	int ready;
   1068 
   1069 	if (hint & NOTE_SUBMIT)
   1070 		KASSERT(mutex_owned(&tp->tun_lock));
   1071 	else
   1072 		mutex_enter(&tp->tun_lock);
   1073 
   1074 	IF_POLL(&ifp->if_snd, m);
   1075 	ready = (m != NULL);
   1076 	for (kn->kn_data = 0; m != NULL; m = m->m_next)
   1077 		kn->kn_data += m->m_len;
   1078 
   1079 	if (hint & NOTE_SUBMIT)
   1080 		KASSERT(mutex_owned(&tp->tun_lock));
   1081 	else
   1082 		mutex_exit(&tp->tun_lock);
   1083 
   1084 	return ready;
   1085 }
   1086 
   1087 static const struct filterops tunread_filtops = {
   1088 	.f_isfd = 1,
   1089 	.f_attach = NULL,
   1090 	.f_detach = filt_tunrdetach,
   1091 	.f_event = filt_tunread,
   1092 };
   1093 
   1094 static const struct filterops tun_seltrue_filtops = {
   1095 	.f_isfd = 1,
   1096 	.f_attach = NULL,
   1097 	.f_detach = filt_tunrdetach,
   1098 	.f_event = filt_seltrue,
   1099 };
   1100 
   1101 int
   1102 tunkqfilter(dev_t dev, struct knote *kn)
   1103 {
   1104 	struct tun_softc *tp;
   1105 	struct klist *klist;
   1106 	int rv = 0;
   1107 
   1108 	tp = tun_find_unit(dev);
   1109 	if (tp == NULL)
   1110 		goto out_nolock;
   1111 
   1112 	switch (kn->kn_filter) {
   1113 	case EVFILT_READ:
   1114 		klist = &tp->tun_rsel.sel_klist;
   1115 		kn->kn_fop = &tunread_filtops;
   1116 		break;
   1117 
   1118 	case EVFILT_WRITE:
   1119 		klist = &tp->tun_rsel.sel_klist;
   1120 		kn->kn_fop = &tun_seltrue_filtops;
   1121 		break;
   1122 
   1123 	default:
   1124 		rv = EINVAL;
   1125 		goto out;
   1126 	}
   1127 
   1128 	kn->kn_hook = tp;
   1129 
   1130 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
   1131 
   1132 out:
   1133 	mutex_exit(&tp->tun_lock);
   1134 out_nolock:
   1135 	return rv;
   1136 }
   1137 
   1138 /*
   1139  * Module infrastructure
   1140  */
   1141 #include "if_module.h"
   1142 
   1143 IF_MODULE(MODULE_CLASS_DRIVER, tun, "")
   1144