Home | History | Annotate | Line # | Download | only in net
if_tun.c revision 1.131
      1 /*	$NetBSD: if_tun.c,v 1.131 2016/09/07 10:24:57 ozaki-r 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 1.131 2016/09/07 10:24:57 ozaki-r Exp $");
     19 
     20 #ifdef _KERNEL_OPT
     21 #include "opt_inet.h"
     22 #endif
     23 
     24 #include <sys/param.h>
     25 #include <sys/proc.h>
     26 #include <sys/systm.h>
     27 #include <sys/mbuf.h>
     28 #include <sys/buf.h>
     29 #include <sys/protosw.h>
     30 #include <sys/socket.h>
     31 #include <sys/ioctl.h>
     32 #include <sys/errno.h>
     33 #include <sys/syslog.h>
     34 #include <sys/select.h>
     35 #include <sys/poll.h>
     36 #include <sys/file.h>
     37 #include <sys/signalvar.h>
     38 #include <sys/conf.h>
     39 #include <sys/kauth.h>
     40 #include <sys/mutex.h>
     41 #include <sys/cpu.h>
     42 #include <sys/device.h>
     43 #include <sys/module.h>
     44 
     45 #include <net/if.h>
     46 #include <net/if_types.h>
     47 #include <net/netisr.h>
     48 #include <net/route.h>
     49 
     50 
     51 #ifdef INET
     52 #include <netinet/in.h>
     53 #include <netinet/in_systm.h>
     54 #include <netinet/in_var.h>
     55 #include <netinet/ip.h>
     56 #include <netinet/if_inarp.h>
     57 #endif
     58 
     59 
     60 #include <sys/time.h>
     61 #include <net/bpf.h>
     62 
     63 #include <net/if_tun.h>
     64 
     65 #include "ioconf.h"
     66 
     67 #define TUNDEBUG	if (tundebug) printf
     68 int	tundebug = 0;
     69 
     70 extern int ifqmaxlen;
     71 
     72 static LIST_HEAD(, tun_softc) tun_softc_list;
     73 static LIST_HEAD(, tun_softc) tunz_softc_list;
     74 static kmutex_t tun_softc_lock;
     75 
     76 static int	tun_ioctl(struct ifnet *, u_long, void *);
     77 static int	tun_output(struct ifnet *, struct mbuf *,
     78 			const struct sockaddr *, const struct rtentry *rt);
     79 static int	tun_clone_create(struct if_clone *, int);
     80 static int	tun_clone_destroy(struct ifnet *);
     81 
     82 static struct if_clone tun_cloner =
     83     IF_CLONE_INITIALIZER("tun", tun_clone_create, tun_clone_destroy);
     84 
     85 static void tunattach0(struct tun_softc *);
     86 static void tun_enable(struct tun_softc *);
     87 static void tun_i_softintr(void *);
     88 static void tun_o_softintr(void *);
     89 #ifdef ALTQ
     90 static void tunstart(struct ifnet *);
     91 #endif
     92 static struct tun_softc *tun_find_unit(dev_t);
     93 static struct tun_softc *tun_find_zunit(int);
     94 
     95 static dev_type_open(tunopen);
     96 static dev_type_close(tunclose);
     97 static dev_type_read(tunread);
     98 static dev_type_write(tunwrite);
     99 static dev_type_ioctl(tunioctl);
    100 static dev_type_poll(tunpoll);
    101 static dev_type_kqfilter(tunkqfilter);
    102 
    103 const struct cdevsw tun_cdevsw = {
    104 	.d_open = tunopen,
    105 	.d_close = tunclose,
    106 	.d_read = tunread,
    107 	.d_write = tunwrite,
    108 	.d_ioctl = tunioctl,
    109 	.d_stop = nostop,
    110 	.d_tty = notty,
    111 	.d_poll = tunpoll,
    112 	.d_mmap = nommap,
    113 	.d_kqfilter = tunkqfilter,
    114 	.d_discard = nodiscard,
    115 	.d_flag = D_OTHER
    116 };
    117 
    118 #ifdef _MODULE
    119 devmajor_t tun_bmajor = -1, tun_cmajor = -1;
    120 #endif
    121 
    122 void
    123 tunattach(int unused)
    124 {
    125 
    126 	/*
    127 	 * Nothing to do here, initialization is handled by the
    128 	 * module initialization code in tuninit() below).
    129 	 */
    130 }
    131 
    132 static void
    133 tuninit(void)
    134 {
    135 
    136 	mutex_init(&tun_softc_lock, MUTEX_DEFAULT, IPL_NET);
    137 	LIST_INIT(&tun_softc_list);
    138 	LIST_INIT(&tunz_softc_list);
    139 	if_clone_attach(&tun_cloner);
    140 #ifdef _MODULE
    141 	devsw_attach("tun", NULL, &tun_bmajor, &tun_cdevsw, &tun_cmajor);
    142 #endif
    143 }
    144 
    145 static int
    146 tundetach(void)
    147 {
    148 	int error = 0;
    149 
    150 	if (!LIST_EMPTY(&tun_softc_list) || !LIST_EMPTY(&tunz_softc_list))
    151 		error = EBUSY;
    152 
    153 #ifdef _MODULE
    154 	if (error == 0)
    155 		error = devsw_detach(NULL, &tun_cdevsw);
    156 #endif
    157 	if (error == 0) {
    158 		if_clone_detach(&tun_cloner);
    159 		mutex_destroy(&tun_softc_lock);
    160 	}
    161 
    162 	return error;
    163 }
    164 
    165 /*
    166  * Find driver instance from dev_t.
    167  * Returns with tp locked (if found).
    168  */
    169 static struct tun_softc *
    170 tun_find_unit(dev_t dev)
    171 {
    172 	struct tun_softc *tp;
    173 	int unit = minor(dev);
    174 
    175 	mutex_enter(&tun_softc_lock);
    176 	LIST_FOREACH(tp, &tun_softc_list, tun_list)
    177 		if (unit == tp->tun_unit)
    178 			break;
    179 	if (tp)
    180 		mutex_enter(&tp->tun_lock);
    181 	mutex_exit(&tun_softc_lock);
    182 
    183 	return (tp);
    184 }
    185 
    186 /*
    187  * Find zombie driver instance by unit number.
    188  * Remove tp from list and return it unlocked (if found).
    189  */
    190 static struct tun_softc *
    191 tun_find_zunit(int unit)
    192 {
    193 	struct tun_softc *tp;
    194 
    195 	mutex_enter(&tun_softc_lock);
    196 	LIST_FOREACH(tp, &tunz_softc_list, tun_list)
    197 		if (unit == tp->tun_unit)
    198 			break;
    199 	if (tp)
    200 		LIST_REMOVE(tp, tun_list);
    201 	mutex_exit(&tun_softc_lock);
    202 #ifdef DIAGNOSTIC
    203 	if (tp != NULL && (tp->tun_flags & (TUN_INITED|TUN_OPEN)) != TUN_OPEN)
    204 		printf("tun%d: inconsistent flags: %x\n", unit, tp->tun_flags);
    205 #endif
    206 
    207 	return (tp);
    208 }
    209 
    210 static int
    211 tun_clone_create(struct if_clone *ifc, int unit)
    212 {
    213 	struct tun_softc *tp;
    214 
    215 	if ((tp = tun_find_zunit(unit)) == NULL) {
    216 		/* Allocate a new instance */
    217 		tp = malloc(sizeof(*tp), M_DEVBUF, M_WAITOK|M_ZERO);
    218 
    219 		tp->tun_unit = unit;
    220 		mutex_init(&tp->tun_lock, MUTEX_DEFAULT, IPL_NET);
    221 		selinit(&tp->tun_rsel);
    222 		selinit(&tp->tun_wsel);
    223 	} else {
    224 		/* Revive tunnel instance; clear ifp part */
    225 		(void)memset(&tp->tun_if, 0, sizeof(struct ifnet));
    226 	}
    227 
    228 	if_initname(&tp->tun_if, ifc->ifc_name, unit);
    229 	tunattach0(tp);
    230 	tp->tun_flags |= TUN_INITED;
    231 	tp->tun_osih = softint_establish(SOFTINT_CLOCK, tun_o_softintr, tp);
    232 	tp->tun_isih = softint_establish(SOFTINT_CLOCK, tun_i_softintr, tp);
    233 
    234 	mutex_enter(&tun_softc_lock);
    235 	LIST_INSERT_HEAD(&tun_softc_list, tp, tun_list);
    236 	mutex_exit(&tun_softc_lock);
    237 
    238 	return (0);
    239 }
    240 
    241 static void
    242 tunattach0(struct tun_softc *tp)
    243 {
    244 	struct ifnet *ifp;
    245 
    246 	ifp = &tp->tun_if;
    247 	ifp->if_softc = tp;
    248 	ifp->if_mtu = TUNMTU;
    249 	ifp->if_ioctl = tun_ioctl;
    250 	ifp->if_output = tun_output;
    251 #ifdef ALTQ
    252 	ifp->if_start = tunstart;
    253 #endif
    254 	ifp->if_flags = IFF_POINTOPOINT;
    255 	ifp->if_type = IFT_TUNNEL;
    256 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
    257 	ifp->if_collisions = 0;
    258 	ifp->if_ierrors = 0;
    259 	ifp->if_oerrors = 0;
    260 	ifp->if_ipackets = 0;
    261 	ifp->if_opackets = 0;
    262 	ifp->if_ibytes   = 0;
    263 	ifp->if_obytes   = 0;
    264 	ifp->if_dlt = DLT_NULL;
    265 	IFQ_SET_READY(&ifp->if_snd);
    266 	if_attach(ifp);
    267 	if_alloc_sadl(ifp);
    268 	bpf_attach(ifp, DLT_NULL, sizeof(uint32_t));
    269 }
    270 
    271 static int
    272 tun_clone_destroy(struct ifnet *ifp)
    273 {
    274 	struct tun_softc *tp = (void *)ifp;
    275 	int zombie = 0;
    276 
    277 	IF_PURGE(&ifp->if_snd);
    278 	ifp->if_flags &= ~IFF_RUNNING;
    279 
    280 	mutex_enter(&tun_softc_lock);
    281 	mutex_enter(&tp->tun_lock);
    282 	LIST_REMOVE(tp, tun_list);
    283 	if (tp->tun_flags & TUN_OPEN) {
    284 		/* Hang on to storage until last close */
    285 		zombie = 1;
    286 		tp->tun_flags &= ~TUN_INITED;
    287 		LIST_INSERT_HEAD(&tunz_softc_list, tp, tun_list);
    288 	}
    289 	mutex_exit(&tun_softc_lock);
    290 
    291 	if (tp->tun_flags & TUN_RWAIT) {
    292 		tp->tun_flags &= ~TUN_RWAIT;
    293 		wakeup((void *)tp);
    294 	}
    295 	selnotify(&tp->tun_rsel, 0, 0);
    296 
    297 	mutex_exit(&tp->tun_lock);
    298 
    299 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    300 		fownsignal(tp->tun_pgid, SIGIO, POLL_HUP, 0, NULL);
    301 
    302 	bpf_detach(ifp);
    303 	if_detach(ifp);
    304 
    305 	if (!zombie) {
    306 		seldestroy(&tp->tun_rsel);
    307 		seldestroy(&tp->tun_wsel);
    308 		softint_disestablish(tp->tun_osih);
    309 		softint_disestablish(tp->tun_isih);
    310 		mutex_destroy(&tp->tun_lock);
    311 		free(tp, M_DEVBUF);
    312 	}
    313 
    314 	return (0);
    315 }
    316 
    317 /*
    318  * tunnel open - must be superuser & the device must be
    319  * configured in
    320  */
    321 static int
    322 tunopen(dev_t dev, int flag, int mode, struct lwp *l)
    323 {
    324 	struct ifnet	*ifp;
    325 	struct tun_softc *tp;
    326 	int	error;
    327 
    328 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE_TUN,
    329 	    KAUTH_REQ_NETWORK_INTERFACE_TUN_ADD, NULL, NULL, NULL);
    330 	if (error)
    331 		return (error);
    332 
    333 	tp = tun_find_unit(dev);
    334 
    335 	if (tp == NULL) {
    336 		(void)tun_clone_create(&tun_cloner, minor(dev));
    337 		tp = tun_find_unit(dev);
    338 		if (tp == NULL) {
    339 			error = ENXIO;
    340 			goto out_nolock;
    341 		}
    342 	}
    343 
    344 	if (tp->tun_flags & TUN_OPEN) {
    345 		error = EBUSY;
    346 		goto out;
    347 	}
    348 
    349 	ifp = &tp->tun_if;
    350 	tp->tun_flags |= TUN_OPEN;
    351 	TUNDEBUG("%s: open\n", ifp->if_xname);
    352 out:
    353 	mutex_exit(&tp->tun_lock);
    354 out_nolock:
    355 	return (error);
    356 }
    357 
    358 /*
    359  * tunclose - close the device - mark i/f down & delete
    360  * routing info
    361  */
    362 int
    363 tunclose(dev_t dev, int flag, int mode,
    364     struct lwp *l)
    365 {
    366 	struct tun_softc *tp;
    367 	struct ifnet	*ifp;
    368 
    369 	if ((tp = tun_find_zunit(minor(dev))) != NULL) {
    370 		/* interface was "destroyed" before the close */
    371 		seldestroy(&tp->tun_rsel);
    372 		seldestroy(&tp->tun_wsel);
    373 		softint_disestablish(tp->tun_osih);
    374 		softint_disestablish(tp->tun_isih);
    375 		mutex_destroy(&tp->tun_lock);
    376 		free(tp, M_DEVBUF);
    377 		goto out_nolock;
    378 	}
    379 
    380 	if ((tp = tun_find_unit(dev)) == NULL)
    381 		goto out_nolock;
    382 
    383 	ifp = &tp->tun_if;
    384 
    385 	tp->tun_flags &= ~TUN_OPEN;
    386 
    387 	tp->tun_pgid = 0;
    388 	selnotify(&tp->tun_rsel, 0, 0);
    389 
    390 	TUNDEBUG ("%s: closed\n", ifp->if_xname);
    391 	mutex_exit(&tp->tun_lock);
    392 
    393 	/*
    394 	 * junk all pending output
    395 	 */
    396 	IFQ_PURGE(&ifp->if_snd);
    397 
    398 	if (ifp->if_flags & IFF_UP) {
    399 		if_down(ifp);
    400 		if (ifp->if_flags & IFF_RUNNING) {
    401 			/* find internet addresses and delete routes */
    402 			struct ifaddr *ifa;
    403 			IFADDR_READER_FOREACH(ifa, ifp) {
    404 #if defined(INET) || defined(INET6)
    405 				if (ifa->ifa_addr->sa_family == AF_INET ||
    406 				    ifa->ifa_addr->sa_family == AF_INET6) {
    407 					rtinit(ifa, (int)RTM_DELETE,
    408 					       tp->tun_flags & TUN_DSTADDR
    409 							? RTF_HOST
    410 							: 0);
    411 				}
    412 #endif
    413 			}
    414 		}
    415 	}
    416 out_nolock:
    417 	return (0);
    418 }
    419 
    420 /*
    421  * Call at splnet().
    422  */
    423 static void
    424 tun_enable(struct tun_softc *tp)
    425 {
    426 	struct ifnet	*ifp = &tp->tun_if;
    427 	struct ifaddr	*ifa;
    428 
    429 	TUNDEBUG("%s: %s\n", __func__, ifp->if_xname);
    430 
    431 	mutex_enter(&tp->tun_lock);
    432 	ifp->if_flags |= IFF_UP | IFF_RUNNING;
    433 
    434 	tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR);
    435 	IFADDR_READER_FOREACH(ifa, ifp) {
    436 #ifdef INET
    437 		if (ifa->ifa_addr->sa_family == AF_INET) {
    438 			struct sockaddr_in *sin;
    439 
    440 			sin = satosin(ifa->ifa_addr);
    441 			if (sin && sin->sin_addr.s_addr)
    442 				tp->tun_flags |= TUN_IASET;
    443 
    444 			if (ifp->if_flags & IFF_POINTOPOINT) {
    445 				sin = satosin(ifa->ifa_dstaddr);
    446 				if (sin && sin->sin_addr.s_addr)
    447 					tp->tun_flags |= TUN_DSTADDR;
    448 			}
    449 		}
    450 #endif
    451 #ifdef INET6
    452 		if (ifa->ifa_addr->sa_family == AF_INET6) {
    453 			struct sockaddr_in6 *sin;
    454 
    455 			sin = (struct sockaddr_in6 *)ifa->ifa_addr;
    456 			if (!IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
    457 				tp->tun_flags |= TUN_IASET;
    458 
    459 			if (ifp->if_flags & IFF_POINTOPOINT) {
    460 				sin = (struct sockaddr_in6 *)ifa->ifa_dstaddr;
    461 				if (sin &&
    462 				    !IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
    463 					tp->tun_flags |= TUN_DSTADDR;
    464 			} else
    465 				tp->tun_flags &= ~TUN_DSTADDR;
    466 		}
    467 #endif /* INET6 */
    468 	}
    469 	mutex_exit(&tp->tun_lock);
    470 }
    471 
    472 /*
    473  * Process an ioctl request.
    474  */
    475 static int
    476 tun_ioctl(struct ifnet *ifp, u_long cmd, void *data)
    477 {
    478 	int		error = 0, s;
    479 	struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc);
    480 	struct ifreq *ifr = (struct ifreq *)data;
    481 	struct ifaddr *ifa = (struct ifaddr *)data;
    482 
    483 	s = splnet();
    484 
    485 	switch (cmd) {
    486 	case SIOCINITIFADDR:
    487 		tun_enable(tp);
    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 	splx(s);
    528 	return (error);
    529 }
    530 
    531 /*
    532  * tun_output - queue packets from higher level ready to put out.
    533  */
    534 static int
    535 tun_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
    536     const struct rtentry *rt)
    537 {
    538 	struct tun_softc *tp = ifp->if_softc;
    539 	int		s;
    540 	int		error;
    541 #if defined(INET) || defined(INET6)
    542 	int		mlen;
    543 	uint32_t	*af;
    544 #endif
    545 
    546 	s = splnet();
    547 	mutex_enter(&tp->tun_lock);
    548 	TUNDEBUG ("%s: tun_output\n", ifp->if_xname);
    549 
    550 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
    551 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname,
    552 			  tp->tun_flags);
    553 		error = EHOSTDOWN;
    554 		goto out;
    555 	}
    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);
    564 
    565 	switch(dst->sa_family) {
    566 #ifdef INET6
    567 	case AF_INET6:
    568 #endif
    569 #ifdef INET
    570 	case AF_INET:
    571 #endif
    572 #if defined(INET) || defined(INET6)
    573 		if (tp->tun_flags & TUN_PREPADDR) {
    574 			/* Simple link-layer header */
    575 			M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
    576 			if (m0 == NULL) {
    577 				IF_DROP(&ifp->if_snd);
    578 				error = ENOBUFS;
    579 				goto out;
    580 			}
    581 			bcopy(dst, mtod(m0, char *), dst->sa_len);
    582 		}
    583 
    584 		if (tp->tun_flags & TUN_IFHEAD) {
    585 			/* Prepend the address family */
    586 			M_PREPEND(m0, sizeof(*af), M_DONTWAIT);
    587 			if (m0 == NULL) {
    588 				IF_DROP(&ifp->if_snd);
    589 				error = ENOBUFS;
    590 				goto out;
    591 			}
    592 			af = mtod(m0,uint32_t *);
    593 			*af = htonl(dst->sa_family);
    594 		} else {
    595 #ifdef INET
    596 			if (dst->sa_family != AF_INET)
    597 #endif
    598 			{
    599 				error = EAFNOSUPPORT;
    600 				goto out;
    601 			}
    602 		}
    603 		/* FALLTHROUGH */
    604 	case AF_UNSPEC:
    605 		IFQ_ENQUEUE(&ifp->if_snd, m0, error);
    606 		if (error) {
    607 			ifp->if_collisions++;
    608 			error = EAFNOSUPPORT;
    609 			m0 = NULL;
    610 			goto out;
    611 		}
    612 		mlen = m0->m_pkthdr.len;
    613 		ifp->if_opackets++;
    614 		ifp->if_obytes += mlen;
    615 		break;
    616 #endif
    617 	default:
    618 		error = EAFNOSUPPORT;
    619 		goto out;
    620 	}
    621 
    622 	if (tp->tun_flags & TUN_RWAIT) {
    623 		tp->tun_flags &= ~TUN_RWAIT;
    624 		wakeup((void *)tp);
    625 	}
    626 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    627 		softint_schedule(tp->tun_isih);
    628 
    629 	selnotify(&tp->tun_rsel, 0, 0);
    630 out:
    631 	mutex_exit(&tp->tun_lock);
    632 	splx(s);
    633 
    634 	if (error && m0) {
    635 		m_freem(m0);
    636 	}
    637 	return 0;
    638 }
    639 
    640 static void
    641 tun_i_softintr(void *cookie)
    642 {
    643 	struct tun_softc *tp = cookie;
    644 
    645 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    646 		fownsignal(tp->tun_pgid, SIGIO, POLL_IN, POLLIN|POLLRDNORM,
    647 		    NULL);
    648 }
    649 
    650 static void
    651 tun_o_softintr(void *cookie)
    652 {
    653 	struct tun_softc *tp = cookie;
    654 
    655 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
    656 		fownsignal(tp->tun_pgid, SIGIO, POLL_OUT, POLLOUT|POLLWRNORM,
    657 		    NULL);
    658 }
    659 
    660 /*
    661  * the cdevsw interface is now pretty minimal.
    662  */
    663 int
    664 tunioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    665 {
    666 	struct tun_softc *tp;
    667 	int s, error = 0;
    668 
    669 	s = splnet();
    670 	tp = tun_find_unit(dev);
    671 
    672 	/* interface was "destroyed" already */
    673 	if (tp == NULL) {
    674 		error = ENXIO;
    675 		goto out_nolock;
    676 	}
    677 
    678 	switch (cmd) {
    679 	case TUNSDEBUG:
    680 		tundebug = *(int *)data;
    681 		break;
    682 
    683 	case TUNGDEBUG:
    684 		*(int *)data = tundebug;
    685 		break;
    686 
    687 	case TUNSIFMODE:
    688 		switch (*(int *)data & (IFF_POINTOPOINT|IFF_BROADCAST)) {
    689 		case IFF_POINTOPOINT:
    690 		case IFF_BROADCAST:
    691 			if (tp->tun_if.if_flags & IFF_UP) {
    692 				error = EBUSY;
    693 				goto out;
    694 			}
    695 			tp->tun_if.if_flags &=
    696 				~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
    697 			tp->tun_if.if_flags |= *(int *)data;
    698 			break;
    699 		default:
    700 			error = EINVAL;
    701 			goto out;
    702 		}
    703 		break;
    704 
    705 	case TUNSLMODE:
    706 		if (*(int *)data) {
    707 			tp->tun_flags |= TUN_PREPADDR;
    708 			tp->tun_flags &= ~TUN_IFHEAD;
    709 		} else
    710 			tp->tun_flags &= ~TUN_PREPADDR;
    711 		break;
    712 
    713 	case TUNSIFHEAD:
    714 		if (*(int *)data) {
    715 			tp->tun_flags |= TUN_IFHEAD;
    716 			tp->tun_flags &= ~TUN_PREPADDR;
    717 		} else
    718 			tp->tun_flags &= ~TUN_IFHEAD;
    719 		break;
    720 
    721 	case TUNGIFHEAD:
    722 		*(int *)data = (tp->tun_flags & TUN_IFHEAD);
    723 		break;
    724 
    725 	case FIONBIO:
    726 		if (*(int *)data)
    727 			tp->tun_flags |= TUN_NBIO;
    728 		else
    729 			tp->tun_flags &= ~TUN_NBIO;
    730 		break;
    731 
    732 	case FIOASYNC:
    733 		if (*(int *)data)
    734 			tp->tun_flags |= TUN_ASYNC;
    735 		else
    736 			tp->tun_flags &= ~TUN_ASYNC;
    737 		break;
    738 
    739 	case FIONREAD:
    740 		if (tp->tun_if.if_snd.ifq_head)
    741 			*(int *)data = tp->tun_if.if_snd.ifq_head->m_pkthdr.len;
    742 		else
    743 			*(int *)data = 0;
    744 		break;
    745 
    746 	case TIOCSPGRP:
    747 	case FIOSETOWN:
    748 		error = fsetown(&tp->tun_pgid, cmd, data);
    749 		break;
    750 
    751 	case TIOCGPGRP:
    752 	case FIOGETOWN:
    753 		error = fgetown(tp->tun_pgid, cmd, data);
    754 		break;
    755 
    756 	default:
    757 		error = ENOTTY;
    758 	}
    759 
    760 out:
    761 	mutex_exit(&tp->tun_lock);
    762 out_nolock:
    763 	splx(s);
    764 	return (error);
    765 }
    766 
    767 /*
    768  * The cdevsw read interface - reads a packet at a time, or at
    769  * least as much of a packet as can be read.
    770  */
    771 int
    772 tunread(dev_t dev, struct uio *uio, int ioflag)
    773 {
    774 	struct tun_softc *tp;
    775 	struct ifnet	*ifp;
    776 	struct mbuf	*m, *m0;
    777 	int		error = 0, len, s, index;
    778 
    779 	s = splnet();
    780 	tp = tun_find_unit(dev);
    781 
    782 	/* interface was "destroyed" already */
    783 	if (tp == NULL) {
    784 		error = ENXIO;
    785 		goto out_nolock;
    786 	}
    787 
    788 	index = tp->tun_if.if_index;
    789 	ifp = &tp->tun_if;
    790 
    791 	TUNDEBUG ("%s: read\n", ifp->if_xname);
    792 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
    793 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags);
    794 		error = EHOSTDOWN;
    795 		goto out;
    796 	}
    797 
    798 	tp->tun_flags &= ~TUN_RWAIT;
    799 
    800 	do {
    801 		IFQ_DEQUEUE(&ifp->if_snd, m0);
    802 		if (m0 == 0) {
    803 			if (tp->tun_flags & TUN_NBIO) {
    804 				error = EWOULDBLOCK;
    805 				goto out;
    806 			}
    807 			tp->tun_flags |= TUN_RWAIT;
    808 			if (mtsleep((void *)tp, PZERO|PCATCH|PNORELOCK,
    809 					"tunread", 0, &tp->tun_lock) != 0) {
    810 				error = EINTR;
    811 				goto out_nolock;
    812 			} else {
    813 				/*
    814 				 * Maybe the interface was destroyed while
    815 				 * we were sleeping, so let's ensure that
    816 				 * we're looking at the same (valid) tun
    817 				 * interface before looping.
    818 				 */
    819 				tp = tun_find_unit(dev);
    820 				if (tp == NULL) {
    821 					error = ENXIO;
    822 					goto out_nolock;
    823 				}
    824 				if (tp->tun_if.if_index != index) {
    825 					error = ENXIO;
    826 					goto out;
    827 				}
    828 			}
    829 		}
    830 	} while (m0 == 0);
    831 
    832 	mutex_exit(&tp->tun_lock);
    833 	splx(s);
    834 
    835 	/* Copy the mbuf chain */
    836 	while (m0 && uio->uio_resid > 0 && error == 0) {
    837 		len = min(uio->uio_resid, m0->m_len);
    838 		if (len != 0)
    839 			error = uiomove(mtod(m0, void *), len, uio);
    840 		MFREE(m0, m);
    841 		m0 = m;
    842 	}
    843 
    844 	if (m0) {
    845 		TUNDEBUG("Dropping mbuf\n");
    846 		m_freem(m0);
    847 	}
    848 	if (error)
    849 		ifp->if_ierrors++;
    850 
    851 	return (error);
    852 
    853 out:
    854 	mutex_exit(&tp->tun_lock);
    855 out_nolock:
    856 	splx(s);
    857 	return (error);
    858 }
    859 
    860 /*
    861  * the cdevsw write interface - an atomic write is a packet - or else!
    862  */
    863 int
    864 tunwrite(dev_t dev, struct uio *uio, int ioflag)
    865 {
    866 	struct tun_softc *tp;
    867 	struct ifnet	*ifp;
    868 	struct mbuf	*top, **mp, *m;
    869 	pktqueue_t	*pktq;
    870 	struct sockaddr	dst;
    871 	int		error = 0, s, tlen, mlen;
    872 	uint32_t	family;
    873 
    874 	s = splnet();
    875 	tp = tun_find_unit(dev);
    876 
    877 	/* interface was "destroyed" already */
    878 	if (tp == NULL) {
    879 		error = ENXIO;
    880 		goto out_nolock;
    881 	}
    882 
    883 	/* Unlock until we've got the data */
    884 	mutex_exit(&tp->tun_lock);
    885 	splx(s);
    886 
    887 	ifp = &tp->tun_if;
    888 
    889 	TUNDEBUG("%s: tunwrite\n", ifp->if_xname);
    890 
    891 	if (tp->tun_flags & TUN_PREPADDR) {
    892 		if (uio->uio_resid < sizeof(dst)) {
    893 			error = EIO;
    894 			goto out0;
    895 		}
    896 		error = uiomove((void *)&dst, sizeof(dst), uio);
    897 		if (dst.sa_len > sizeof(dst)) {
    898 			/* Duh.. */
    899 			char discard;
    900 			int n = dst.sa_len - sizeof(dst);
    901 			while (n--)
    902 				if ((error = uiomove(&discard, 1, uio)) != 0) {
    903 					goto out0;
    904 				}
    905 		}
    906 	} else if (tp->tun_flags & TUN_IFHEAD) {
    907 		if (uio->uio_resid < sizeof(family)){
    908 			error = EIO;
    909 			goto out0;
    910 		}
    911 		error = uiomove((void *)&family, sizeof(family), uio);
    912 		dst.sa_family = ntohl(family);
    913 	} else {
    914 #ifdef INET
    915 		dst.sa_family = AF_INET;
    916 #endif
    917 	}
    918 
    919 	if (uio->uio_resid > TUNMTU) {
    920 		TUNDEBUG("%s: len=%lu!\n", ifp->if_xname,
    921 		    (unsigned long)uio->uio_resid);
    922 		error = EIO;
    923 		goto out0;
    924 	}
    925 
    926 	switch (dst.sa_family) {
    927 #ifdef INET
    928 	case AF_INET:
    929 		pktq = ip_pktq;
    930 		break;
    931 #endif
    932 #ifdef INET6
    933 	case AF_INET6:
    934 		pktq = ip6_pktq;
    935 		break;
    936 #endif
    937 	default:
    938 		error = EAFNOSUPPORT;
    939 		goto out0;
    940 	}
    941 
    942 	tlen = uio->uio_resid;
    943 
    944 	/* get a header mbuf */
    945 	MGETHDR(m, M_DONTWAIT, MT_DATA);
    946 	if (m == NULL) {
    947 		error = ENOBUFS;
    948 		goto out0;
    949 	}
    950 	mlen = MHLEN;
    951 
    952 	top = NULL;
    953 	mp = &top;
    954 	while (error == 0 && uio->uio_resid > 0) {
    955 		m->m_len = min(mlen, uio->uio_resid);
    956 		error = uiomove(mtod(m, void *), m->m_len, uio);
    957 		*mp = m;
    958 		mp = &m->m_next;
    959 		if (error == 0 && uio->uio_resid > 0) {
    960 			MGET(m, M_DONTWAIT, MT_DATA);
    961 			if (m == NULL) {
    962 				error = ENOBUFS;
    963 				break;
    964 			}
    965 			mlen = MLEN;
    966 		}
    967 	}
    968 	if (error) {
    969 		if (top != NULL)
    970 			m_freem (top);
    971 		ifp->if_ierrors++;
    972 		goto out0;
    973 	}
    974 
    975 	top->m_pkthdr.len = tlen;
    976 	m_set_rcvif(top, ifp);
    977 
    978 	bpf_mtap_af(ifp, dst.sa_family, top);
    979 
    980 	s = splnet();
    981 	mutex_enter(&tp->tun_lock);
    982 	if ((tp->tun_flags & TUN_INITED) == 0) {
    983 		/* Interface was destroyed */
    984 		error = ENXIO;
    985 		goto out;
    986 	}
    987 	if (__predict_false(!pktq_enqueue(pktq, top, 0))) {
    988 		ifp->if_collisions++;
    989 		mutex_exit(&tp->tun_lock);
    990 		error = ENOBUFS;
    991 		m_freem(top);
    992 		goto out_nolock;
    993 	}
    994 	ifp->if_ipackets++;
    995 	ifp->if_ibytes += tlen;
    996 out:
    997 	mutex_exit(&tp->tun_lock);
    998 out_nolock:
    999 	splx(s);
   1000 out0:
   1001 	return (error);
   1002 }
   1003 
   1004 #ifdef ALTQ
   1005 /*
   1006  * Start packet transmission on the interface.
   1007  * when the interface queue is rate-limited by ALTQ or TBR,
   1008  * if_start is needed to drain packets from the queue in order
   1009  * to notify readers when outgoing packets become ready.
   1010  *
   1011  * Should be called at splnet.
   1012  */
   1013 static void
   1014 tunstart(struct ifnet *ifp)
   1015 {
   1016 	struct tun_softc *tp = ifp->if_softc;
   1017 
   1018 	if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd))
   1019 		return;
   1020 
   1021 	mutex_enter(&tp->tun_lock);
   1022 	if (!IF_IS_EMPTY(&ifp->if_snd)) {
   1023 		if (tp->tun_flags & TUN_RWAIT) {
   1024 			tp->tun_flags &= ~TUN_RWAIT;
   1025 			wakeup((void *)tp);
   1026 		}
   1027 		if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
   1028 			softint_schedule(tp->tun_osih);
   1029 
   1030 		selnotify(&tp->tun_rsel, 0, 0);
   1031 	}
   1032 	mutex_exit(&tp->tun_lock);
   1033 }
   1034 #endif /* ALTQ */
   1035 /*
   1036  * tunpoll - the poll interface, this is only useful on reads
   1037  * really. The write detect always returns true, write never blocks
   1038  * anyway, it either accepts the packet or drops it.
   1039  */
   1040 int
   1041 tunpoll(dev_t dev, int events, struct lwp *l)
   1042 {
   1043 	struct tun_softc *tp;
   1044 	struct ifnet	*ifp;
   1045 	int		s, revents = 0;
   1046 
   1047 	s = splnet();
   1048 	tp = tun_find_unit(dev);
   1049 
   1050 	/* interface was "destroyed" already */
   1051 	if (tp == NULL)
   1052 		goto out_nolock;
   1053 
   1054 	ifp = &tp->tun_if;
   1055 
   1056 	TUNDEBUG("%s: tunpoll\n", ifp->if_xname);
   1057 
   1058 	if (events & (POLLIN | POLLRDNORM)) {
   1059 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
   1060 			TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname,
   1061 			    ifp->if_snd.ifq_len);
   1062 			revents |= events & (POLLIN | POLLRDNORM);
   1063 		} else {
   1064 			TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname);
   1065 			selrecord(l, &tp->tun_rsel);
   1066 		}
   1067 	}
   1068 
   1069 	if (events & (POLLOUT | POLLWRNORM))
   1070 		revents |= events & (POLLOUT | POLLWRNORM);
   1071 
   1072 	mutex_exit(&tp->tun_lock);
   1073 out_nolock:
   1074 	splx(s);
   1075 	return (revents);
   1076 }
   1077 
   1078 static void
   1079 filt_tunrdetach(struct knote *kn)
   1080 {
   1081 	struct tun_softc *tp = kn->kn_hook;
   1082 	int s;
   1083 
   1084 	s = splnet();
   1085 	SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext);
   1086 	splx(s);
   1087 }
   1088 
   1089 static int
   1090 filt_tunread(struct knote *kn, long hint)
   1091 {
   1092 	struct tun_softc *tp = kn->kn_hook;
   1093 	struct ifnet *ifp = &tp->tun_if;
   1094 	struct mbuf *m;
   1095 	int s;
   1096 
   1097 	s = splnet();
   1098 	IF_POLL(&ifp->if_snd, m);
   1099 	if (m == NULL) {
   1100 		splx(s);
   1101 		return (0);
   1102 	}
   1103 
   1104 	for (kn->kn_data = 0; m != NULL; m = m->m_next)
   1105 		kn->kn_data += m->m_len;
   1106 
   1107 	splx(s);
   1108 	return (1);
   1109 }
   1110 
   1111 static const struct filterops tunread_filtops =
   1112 	{ 1, NULL, filt_tunrdetach, filt_tunread };
   1113 
   1114 static const struct filterops tun_seltrue_filtops =
   1115 	{ 1, NULL, filt_tunrdetach, filt_seltrue };
   1116 
   1117 int
   1118 tunkqfilter(dev_t dev, struct knote *kn)
   1119 {
   1120 	struct tun_softc *tp;
   1121 	struct klist *klist;
   1122 	int rv = 0, s;
   1123 
   1124 	s = splnet();
   1125 	tp = tun_find_unit(dev);
   1126 	if (tp == NULL)
   1127 		goto out_nolock;
   1128 
   1129 	switch (kn->kn_filter) {
   1130 	case EVFILT_READ:
   1131 		klist = &tp->tun_rsel.sel_klist;
   1132 		kn->kn_fop = &tunread_filtops;
   1133 		break;
   1134 
   1135 	case EVFILT_WRITE:
   1136 		klist = &tp->tun_rsel.sel_klist;
   1137 		kn->kn_fop = &tun_seltrue_filtops;
   1138 		break;
   1139 
   1140 	default:
   1141 		rv = EINVAL;
   1142 		goto out;
   1143 	}
   1144 
   1145 	kn->kn_hook = tp;
   1146 
   1147 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
   1148 
   1149 out:
   1150 	mutex_exit(&tp->tun_lock);
   1151 out_nolock:
   1152 	splx(s);
   1153 	return (rv);
   1154 }
   1155 
   1156 /*
   1157  * Module infrastructure
   1158  */
   1159 #include "if_module.h"
   1160 
   1161 IF_MODULE(MODULE_CLASS_DRIVER, tun, "")
   1162