ad4ef23f5b94bb1e8794f11f467aac1c04df9a9c
[dragonfly.git] / sys / net / tun / if_tun.c
1 /*      $NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $  */
2
3 /*
4  * Copyright (c) 1988, Julian Onions <jpo@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 it's
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  * $FreeBSD: src/sys/net/if_tun.c,v 1.74.2.8 2002/02/13 00:43:11 dillon Exp $
17  */
18
19 #include "use_tun.h"
20 #include "opt_atalk.h"
21 #include "opt_inet.h"
22 #include "opt_inet6.h"
23 #include "opt_ipx.h"
24
25 #include <sys/param.h>
26 #include <sys/proc.h>
27 #include <sys/priv.h>
28 #include <sys/systm.h>
29 #include <sys/mbuf.h>
30 #include <sys/socket.h>
31 #include <sys/conf.h>
32 #include <sys/device.h>
33 #include <sys/filio.h>
34 #include <sys/sockio.h>
35 #include <sys/thread2.h>
36 #include <sys/ttycom.h>
37 #include <sys/signalvar.h>
38 #include <sys/filedesc.h>
39 #include <sys/kernel.h>
40 #include <sys/sysctl.h>
41 #include <sys/uio.h>
42 #include <sys/vnode.h>
43 #include <sys/malloc.h>
44
45 #include <sys/mplock2.h>
46
47 #include <net/if.h>
48 #include <net/if_types.h>
49 #include <net/ifq_var.h>
50 #include <net/netisr.h>
51 #include <net/route.h>
52 #include <sys/devfs.h>
53
54 #ifdef INET
55 #include <netinet/in.h>
56 #endif
57
58 #include <net/bpf.h>
59
60 #include "if_tunvar.h"
61 #include "if_tun.h"
62
63 static MALLOC_DEFINE(M_TUN, "tun", "Tunnel Interface");
64
65 static void tunattach (void *);
66 PSEUDO_SET(tunattach, if_tun);
67
68 static void tuncreate (cdev_t dev);
69
70 #define TUNDEBUG        if (tundebug) if_printf
71 static int tundebug = 0;
72 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
73
74 static int tunoutput (struct ifnet *, struct mbuf *, struct sockaddr *,
75             struct rtentry *rt);
76 static int tunifioctl (struct ifnet *, u_long, caddr_t, struct ucred *);
77 static int tuninit (struct ifnet *);
78 static void tunstart(struct ifnet *);
79 static void tun_filter_detach(struct knote *);
80 static int tun_filter_read(struct knote *, long);
81 static int tun_filter_write(struct knote *, long);
82
83 static  d_open_t        tunopen;
84 static  d_close_t       tunclose;
85 static  d_read_t        tunread;
86 static  d_write_t       tunwrite;
87 static  d_ioctl_t       tunioctl;
88 static  d_kqfilter_t    tunkqfilter;
89
90 static d_clone_t tunclone;
91 DEVFS_DECLARE_CLONE_BITMAP(tun);
92
93 #if NTUN <= 1
94 #define TUN_PREALLOCATED_UNITS  4
95 #else
96 #define TUN_PREALLOCATED_UNITS  NTUN
97 #endif
98
99 static struct dev_ops tun_ops = {
100         { "tun", 0, 0 },
101         .d_open =       tunopen,
102         .d_close =      tunclose,
103         .d_read =       tunread,
104         .d_write =      tunwrite,
105         .d_ioctl =      tunioctl,
106         .d_kqfilter =   tunkqfilter
107 };
108
109 static void
110 tunattach(void *dummy)
111 {
112         int i;
113         make_autoclone_dev(&tun_ops, &DEVFS_CLONE_BITMAP(tun),
114                 tunclone, UID_UUCP, GID_DIALER, 0600, "tun");
115         for (i = 0; i < TUN_PREALLOCATED_UNITS; i++) {
116                 make_dev(&tun_ops, i, UID_UUCP, GID_DIALER, 0600, "tun%d", i);
117                 devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(tun), i);
118         }
119         /* Doesn't need uninit because unloading is not possible, see PSEUDO_SET */
120 }
121
122 static int
123 tunclone(struct dev_clone_args *ap)
124 {
125         int unit;
126
127         unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(tun), 0);
128         ap->a_dev = make_only_dev(&tun_ops, unit, UID_UUCP, GID_DIALER, 0600,
129                                                                 "tun%d", unit);
130
131         return 0;
132 }
133
134 static void
135 tuncreate(cdev_t dev)
136 {
137         struct tun_softc *sc;
138         struct ifnet *ifp;
139
140 #if 0
141         dev = make_dev(&tun_ops, minor(dev),
142             UID_UUCP, GID_DIALER, 0600, "tun%d", lminor(dev));
143 #endif
144
145         MALLOC(sc, struct tun_softc *, sizeof(*sc), M_TUN, M_WAITOK | M_ZERO);
146         sc->tun_flags = TUN_INITED;
147
148         ifp = &sc->tun_if;
149         if_initname(ifp, "tun", lminor(dev));
150         ifp->if_mtu = TUNMTU;
151         ifp->if_ioctl = tunifioctl;
152         ifp->if_output = tunoutput;
153         ifp->if_start = tunstart;
154         ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST;
155         ifp->if_type = IFT_PPP;
156         ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
157         ifq_set_ready(&ifp->if_snd);
158         ifp->if_softc = sc;
159         if_attach(ifp, NULL);
160         bpfattach(ifp, DLT_NULL, sizeof(u_int));
161         dev->si_drv1 = sc;
162 }
163
164 /*
165  * tunnel open - must be superuser & the device must be
166  * configured in
167  */
168 static  int
169 tunopen(struct dev_open_args *ap)
170 {
171         cdev_t dev = ap->a_head.a_dev;
172         struct ifnet    *ifp;
173         struct tun_softc *tp;
174         int     error;
175
176         if ((error = priv_check_cred(ap->a_cred, PRIV_ROOT, 0)) != 0)
177                 return (error);
178
179         tp = dev->si_drv1;
180         if (!tp) {
181                 tuncreate(dev);
182                 tp = dev->si_drv1;
183         }
184         if (tp->tun_flags & TUN_OPEN)
185                 return EBUSY;
186         tp->tun_pid = curproc->p_pid;
187         ifp = &tp->tun_if;
188         tp->tun_flags |= TUN_OPEN;
189         TUNDEBUG(ifp, "open\n");
190         return (0);
191 }
192
193 /*
194  * tunclose - close the device - mark i/f down & delete
195  * routing info
196  */
197 static  int
198 tunclose(struct dev_close_args *ap)
199 {
200         cdev_t dev = ap->a_head.a_dev;
201         struct tun_softc *tp;
202         struct ifnet    *ifp;
203
204         tp = dev->si_drv1;
205         ifp = &tp->tun_if;
206
207         tp->tun_flags &= ~TUN_OPEN;
208         tp->tun_pid = 0;
209
210         /* Junk all pending output. */
211         ifq_purge(&ifp->if_snd);
212
213         if (ifp->if_flags & IFF_UP)
214                 if_down(ifp);
215         ifp->if_flags &= ~IFF_RUNNING;
216         if_purgeaddrs_nolink(ifp);
217
218         funsetown(tp->tun_sigio);
219         KNOTE(&tp->tun_rkq.ki_note, 0);
220
221         TUNDEBUG(ifp, "closed\n");
222 #if 0
223         if (dev->si_uminor >= TUN_PREALLOCATED_UNITS) {
224                 devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(tun), dev->si_uminor);
225         }
226 #endif
227         return (0);
228 }
229
230 static int
231 tuninit(struct ifnet *ifp)
232 {
233         struct tun_softc *tp = ifp->if_softc;
234         struct ifaddr_container *ifac;
235         int error = 0;
236
237         TUNDEBUG(ifp, "tuninit\n");
238
239         ifp->if_flags |= IFF_UP | IFF_RUNNING;
240         getmicrotime(&ifp->if_lastchange);
241
242         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
243                 struct ifaddr *ifa = ifac->ifa;
244
245                 if (ifa->ifa_addr == NULL) {
246                         error = EFAULT;
247                         /* XXX: Should maybe return straight off? */
248                 } else {
249 #ifdef INET
250                         if (ifa->ifa_addr->sa_family == AF_INET) {
251                             struct sockaddr_in *si;
252
253                             si = (struct sockaddr_in *)ifa->ifa_addr;
254                             if (si->sin_addr.s_addr)
255                                     tp->tun_flags |= TUN_IASET;
256                         }
257 #endif
258                 }
259         }
260         return (error);
261 }
262
263 /*
264  * Process an ioctl request.
265  *
266  * MPSAFE
267  */
268 int
269 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
270 {
271         struct ifreq *ifr = (struct ifreq *)data;
272         struct tun_softc *tp = ifp->if_softc;
273         struct ifstat *ifs;
274         int error = 0;
275
276         switch(cmd) {
277         case SIOCGIFSTATUS:
278                 ifs = (struct ifstat *)data;
279                 if (tp->tun_pid)
280                         ksprintf(ifs->ascii + strlen(ifs->ascii),
281                             "\tOpened by PID %d\n", tp->tun_pid);
282                 break;
283         case SIOCSIFADDR:
284                 error = tuninit(ifp);
285                 TUNDEBUG(ifp, "address set, error=%d\n", error);
286                 break;
287         case SIOCSIFDSTADDR:
288                 error = tuninit(ifp);
289                 TUNDEBUG(ifp, "destination address set, error=%d\n", error);
290                 break;
291         case SIOCSIFMTU:
292                 ifp->if_mtu = ifr->ifr_mtu;
293                 TUNDEBUG(ifp, "mtu set\n");
294                 break;
295         case SIOCSIFFLAGS:
296         case SIOCADDMULTI:
297         case SIOCDELMULTI:
298                 break;
299         default:
300                 error = EINVAL;
301         }
302         return (error);
303 }
304
305 /*
306  * tunoutput - queue packets from higher level ready to put out.
307  *
308  * MPSAFE
309  */
310 static int
311 tunoutput_serialized(struct ifnet *ifp, struct mbuf *m0, struct sockaddr *dst,
312                      struct rtentry *rt)
313 {
314         struct tun_softc *tp = ifp->if_softc;
315         int error;
316         struct altq_pktattr pktattr;
317
318         TUNDEBUG(ifp, "tunoutput\n");
319
320         if ((tp->tun_flags & TUN_READY) != TUN_READY) {
321                 TUNDEBUG(ifp, "not ready 0%o\n", tp->tun_flags);
322                 m_freem (m0);
323                 return EHOSTDOWN;
324         }
325
326         /*
327          * if the queueing discipline needs packet classification,
328          * do it before prepending link headers.
329          */
330         ifq_classify(&ifp->if_snd, m0, dst->sa_family, &pktattr);
331
332         /* BPF write needs to be handled specially */
333         if (dst->sa_family == AF_UNSPEC) {
334                 dst->sa_family = *(mtod(m0, int *));
335                 m0->m_len -= sizeof(int);
336                 m0->m_pkthdr.len -= sizeof(int);
337                 m0->m_data += sizeof(int);
338         }
339
340         if (ifp->if_bpf) {
341                 /*
342                  * We need to prepend the address family as
343                  * a four byte field.
344                  */
345                 uint32_t af = dst->sa_family;
346
347                 bpf_ptap(ifp->if_bpf, m0, &af, sizeof(af));
348         }
349
350         /* prepend sockaddr? this may abort if the mbuf allocation fails */
351         if (tp->tun_flags & TUN_LMODE) {
352                 /* allocate space for sockaddr */
353                 M_PREPEND(m0, dst->sa_len, MB_DONTWAIT);
354
355                 /* if allocation failed drop packet */
356                 if (m0 == NULL){
357                         IF_DROP(&ifp->if_snd);
358                         ifp->if_oerrors++;
359                         return (ENOBUFS);
360                 } else {
361                         bcopy(dst, m0->m_data, dst->sa_len);
362                 }
363         }
364
365         if (tp->tun_flags & TUN_IFHEAD) {
366                 /* Prepend the address family */
367                 M_PREPEND(m0, 4, MB_DONTWAIT);
368
369                 /* if allocation failed drop packet */
370                 if (m0 == NULL){
371                         IF_DROP(&ifp->if_snd);
372                         ifp->if_oerrors++;
373                         return ENOBUFS;
374                 } else
375                         *(u_int32_t *)m0->m_data = htonl(dst->sa_family);
376         } else {
377 #ifdef INET
378                 if (dst->sa_family != AF_INET)
379 #endif
380                 {
381                         m_freem(m0);
382                         return EAFNOSUPPORT;
383                 }
384         }
385
386         error = ifq_handoff(ifp, m0, &pktattr);
387         if (error) {
388                 ifp->if_collisions++;
389         } else {
390                 ifp->if_opackets++;
391                 if (tp->tun_flags & TUN_RWAIT) {
392                         tp->tun_flags &= ~TUN_RWAIT;
393                         wakeup((caddr_t)tp);
394                 }
395                 get_mplock();
396                 if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio)
397                         pgsigio(tp->tun_sigio, SIGIO, 0);
398                 rel_mplock();
399                 ifnet_deserialize_all(ifp);
400                 KNOTE(&tp->tun_rkq.ki_note, 0);
401                 ifnet_serialize_all(ifp);
402         }
403         return (error);
404 }
405
406 static int
407 tunoutput(struct ifnet *ifp, struct mbuf *m0, struct sockaddr *dst,
408           struct rtentry *rt)
409 {
410         int error;
411
412         ifnet_serialize_all(ifp);
413         error = tunoutput_serialized(ifp, m0, dst, rt);
414         ifnet_deserialize_all(ifp);
415
416         return error;
417 }
418
419 /*
420  * the ops interface is now pretty minimal.
421  */
422 static  int
423 tunioctl(struct dev_ioctl_args *ap)
424 {
425         cdev_t dev = ap->a_head.a_dev;
426         struct tun_softc *tp = dev->si_drv1;
427         struct tuninfo *tunp;
428
429         switch (ap->a_cmd) {
430         case TUNSIFINFO:
431                 tunp = (struct tuninfo *)ap->a_data;
432                 if (tunp->mtu < IF_MINMTU)
433                         return (EINVAL);
434                 tp->tun_if.if_mtu = tunp->mtu;
435                 tp->tun_if.if_type = tunp->type;
436                 tp->tun_if.if_baudrate = tunp->baudrate;
437                 break;
438         case TUNGIFINFO:
439                 tunp = (struct tuninfo *)ap->a_data;
440                 tunp->mtu = tp->tun_if.if_mtu;
441                 tunp->type = tp->tun_if.if_type;
442                 tunp->baudrate = tp->tun_if.if_baudrate;
443                 break;
444         case TUNSDEBUG:
445                 tundebug = *(int *)ap->a_data;
446                 break;
447         case TUNGDEBUG:
448                 *(int *)ap->a_data = tundebug;
449                 break;
450         case TUNSLMODE:
451                 if (*(int *)ap->a_data) {
452                         tp->tun_flags |= TUN_LMODE;
453                         tp->tun_flags &= ~TUN_IFHEAD;
454                 } else
455                         tp->tun_flags &= ~TUN_LMODE;
456                 break;
457         case TUNSIFHEAD:
458                 if (*(int *)ap->a_data) {
459                         tp->tun_flags |= TUN_IFHEAD;
460                         tp->tun_flags &= ~TUN_LMODE;
461                 } else 
462                         tp->tun_flags &= ~TUN_IFHEAD;
463                 break;
464         case TUNGIFHEAD:
465                 *(int *)ap->a_data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
466                 break;
467         case TUNSIFMODE:
468                 /* deny this if UP */
469                 if (tp->tun_if.if_flags & IFF_UP)
470                         return(EBUSY);
471
472                 switch (*(int *)ap->a_data & ~IFF_MULTICAST) {
473                 case IFF_POINTOPOINT:
474                 case IFF_BROADCAST:
475                         tp->tun_if.if_flags &= ~(IFF_BROADCAST|IFF_POINTOPOINT);
476                         tp->tun_if.if_flags |= *(int *)ap->a_data;
477                         break;
478                 default:
479                         return(EINVAL);
480                 }
481                 break;
482         case TUNSIFPID:
483                 tp->tun_pid = curproc->p_pid;
484                 break;
485         case FIOASYNC:
486                 if (*(int *)ap->a_data)
487                         tp->tun_flags |= TUN_ASYNC;
488                 else
489                         tp->tun_flags &= ~TUN_ASYNC;
490                 break;
491         case FIONREAD:
492                 if (!ifq_is_empty(&tp->tun_if.if_snd)) {
493                         struct mbuf *mb;
494
495                         mb = ifq_poll(&tp->tun_if.if_snd);
496                         for( *(int *)ap->a_data = 0; mb != 0; mb = mb->m_next) 
497                                 *(int *)ap->a_data += mb->m_len;
498                 } else {
499                         *(int *)ap->a_data = 0;
500                 }
501                 break;
502         case FIOSETOWN:
503                 return (fsetown(*(int *)ap->a_data, &tp->tun_sigio));
504
505         case FIOGETOWN:
506                 *(int *)ap->a_data = fgetown(tp->tun_sigio);
507                 return (0);
508
509         /* This is deprecated, FIOSETOWN should be used instead. */
510         case TIOCSPGRP:
511                 return (fsetown(-(*(int *)ap->a_data), &tp->tun_sigio));
512
513         /* This is deprecated, FIOGETOWN should be used instead. */
514         case TIOCGPGRP:
515                 *(int *)ap->a_data = -fgetown(tp->tun_sigio);
516                 return (0);
517
518         default:
519                 return (ENOTTY);
520         }
521         return (0);
522 }
523
524 /*
525  * The ops read interface - reads a packet at a time, or at
526  * least as much of a packet as can be read.
527  */
528 static  int
529 tunread(struct dev_read_args *ap)
530 {
531         cdev_t dev = ap->a_head.a_dev;
532         struct uio *uio = ap->a_uio;
533         struct tun_softc *tp = dev->si_drv1;
534         struct ifnet    *ifp = &tp->tun_if;
535         struct mbuf     *m0;
536         int             error=0, len;
537
538         TUNDEBUG(ifp, "read\n");
539         if ((tp->tun_flags & TUN_READY) != TUN_READY) {
540                 TUNDEBUG(ifp, "not ready 0%o\n", tp->tun_flags);
541                 return EHOSTDOWN;
542         }
543
544         tp->tun_flags &= ~TUN_RWAIT;
545
546         ifnet_serialize_all(ifp);
547
548         while ((m0 = ifq_dequeue(&ifp->if_snd, NULL)) == NULL) {
549                 if (ap->a_ioflag & IO_NDELAY) {
550                         ifnet_deserialize_all(ifp);
551                         return EWOULDBLOCK;
552                 }
553                 tp->tun_flags |= TUN_RWAIT;
554                 ifnet_deserialize_all(ifp);
555                 if ((error = tsleep(tp, PCATCH, "tunread", 0)) != 0)
556                         return error;
557                 ifnet_serialize_all(ifp);
558         }
559
560         ifnet_deserialize_all(ifp);
561
562         while (m0 && uio->uio_resid > 0 && error == 0) {
563                 len = (int)szmin(uio->uio_resid, m0->m_len);
564                 if (len != 0)
565                         error = uiomove(mtod(m0, caddr_t), (size_t)len, uio);
566                 m0 = m_free(m0);
567         }
568
569         if (m0) {
570                 TUNDEBUG(ifp, "Dropping mbuf\n");
571                 m_freem(m0);
572         }
573         return error;
574 }
575
576 /*
577  * the ops write interface - an atomic write is a packet - or else!
578  */
579 static  int
580 tunwrite(struct dev_write_args *ap)
581 {
582         cdev_t dev = ap->a_head.a_dev;
583         struct uio *uio = ap->a_uio;
584         struct tun_softc *tp = dev->si_drv1;
585         struct ifnet    *ifp = &tp->tun_if;
586         struct mbuf     *top, **mp, *m;
587         int             error=0;
588         size_t          tlen, mlen;
589         uint32_t        family;
590         int             isr;
591
592         TUNDEBUG(ifp, "tunwrite\n");
593
594         if (uio->uio_resid == 0)
595                 return 0;
596
597         if (uio->uio_resid > TUNMRU) {
598                 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
599                 return EIO;
600         }
601         tlen = uio->uio_resid;
602
603         /* get a header mbuf */
604         MGETHDR(m, MB_DONTWAIT, MT_DATA);
605         if (m == NULL)
606                 return ENOBUFS;
607         mlen = MHLEN;
608
609         top = 0;
610         mp = &top;
611         while (error == 0 && uio->uio_resid > 0) {
612                 m->m_len = (int)szmin(mlen, uio->uio_resid);
613                 error = uiomove(mtod (m, caddr_t), (size_t)m->m_len, uio);
614                 *mp = m;
615                 mp = &m->m_next;
616                 if (uio->uio_resid > 0) {
617                         MGET (m, MB_DONTWAIT, MT_DATA);
618                         if (m == 0) {
619                                 error = ENOBUFS;
620                                 break;
621                         }
622                         mlen = MLEN;
623                 }
624         }
625         if (error) {
626                 if (top)
627                         m_freem (top);
628                 ifp->if_ierrors++;
629                 return error;
630         }
631
632         top->m_pkthdr.len = (int)tlen;
633         top->m_pkthdr.rcvif = ifp;
634
635         if (ifp->if_bpf) {
636                 if (tp->tun_flags & TUN_IFHEAD) {
637                         /*
638                          * Conveniently, we already have a 4-byte address
639                          * family prepended to our packet !
640                          * Inconveniently, it's in the wrong byte order !
641                          */
642                         if ((top = m_pullup(top, sizeof(family))) == NULL)
643                                 return ENOBUFS;
644                         *mtod(top, u_int32_t *) =
645                             ntohl(*mtod(top, u_int32_t *));
646                         bpf_mtap(ifp->if_bpf, top);
647                         *mtod(top, u_int32_t *) =
648                             htonl(*mtod(top, u_int32_t *));
649                 } else {
650                         /*
651                          * We need to prepend the address family as
652                          * a four byte field.
653                          */
654                         static const uint32_t af = AF_INET;
655
656                         bpf_ptap(ifp->if_bpf, top, &af, sizeof(af));
657                 }
658         }
659
660         if (tp->tun_flags & TUN_IFHEAD) {
661                 if (top->m_len < sizeof(family) &&
662                     (top = m_pullup(top, sizeof(family))) == NULL)
663                                 return ENOBUFS;
664                 family = ntohl(*mtod(top, u_int32_t *));
665                 m_adj(top, sizeof(family));
666         } else
667                 family = AF_INET;
668
669         ifp->if_ibytes += top->m_pkthdr.len;
670         ifp->if_ipackets++;
671
672         switch (family) {
673 #ifdef INET
674         case AF_INET:
675                 isr = NETISR_IP;
676                 break;
677 #endif
678 #ifdef INET6
679         case AF_INET6:
680                 isr = NETISR_IPV6;
681                 break;
682 #endif
683 #ifdef IPX
684         case AF_IPX:
685                 isr = NETISR_IPX;
686                 break;
687 #endif
688 #ifdef NETATALK
689         case AF_APPLETALK:
690                 isr = NETISR_ATALK2;
691                 break;
692 #endif
693         default:
694                 m_freem(m);
695                 return (EAFNOSUPPORT);
696         }
697
698         netisr_queue(isr, top);
699         return (0);
700 }
701
702 static struct filterops tun_read_filtops =
703         { FILTEROP_ISFD, NULL, tun_filter_detach, tun_filter_read };
704 static struct filterops tun_write_filtops =
705         { FILTEROP_ISFD, NULL, tun_filter_detach, tun_filter_write };
706
707 static int
708 tunkqfilter(struct dev_kqfilter_args *ap)
709 {
710         cdev_t dev = ap->a_head.a_dev;
711         struct tun_softc *tp = dev->si_drv1;
712         struct knote *kn = ap->a_kn;
713         struct klist *klist;
714
715         ap->a_result = 0;
716         ifnet_serialize_all(&tp->tun_if);
717
718         switch (kn->kn_filter) {
719         case EVFILT_READ:
720                 kn->kn_fop = &tun_read_filtops;
721                 kn->kn_hook = (caddr_t)tp;
722                 break;
723         case EVFILT_WRITE:
724                 kn->kn_fop = &tun_write_filtops;
725                 kn->kn_hook = (caddr_t)tp;
726                 break;
727         default:
728                 ifnet_deserialize_all(&tp->tun_if);
729                 ap->a_result = EOPNOTSUPP;
730                 return (0);
731         }
732
733         klist = &tp->tun_rkq.ki_note;
734         knote_insert(klist, kn);
735         ifnet_deserialize_all(&tp->tun_if);
736
737         return (0);
738 }
739
740 static void
741 tun_filter_detach(struct knote *kn)
742 {
743         struct tun_softc *tp = (struct tun_softc *)kn->kn_hook;
744         struct klist *klist = &tp->tun_rkq.ki_note;
745
746         knote_remove(klist, kn);
747 }
748
749 static int
750 tun_filter_write(struct knote *kn, long hint)
751 {
752         /* Always ready for a write */
753         return (1);
754 }
755
756 static int
757 tun_filter_read(struct knote *kn, long hint)
758 {
759         struct tun_softc *tp = (struct tun_softc *)kn->kn_hook;
760         int ready = 0;
761
762         ifnet_serialize_all(&tp->tun_if);
763         if (!ifq_is_empty(&tp->tun_if.if_snd))
764                 ready = 1;
765         ifnet_deserialize_all(&tp->tun_if);
766
767         return (ready);
768 }
769
770 /*
771  * Start packet transmission on the interface.
772  * when the interface queue is rate-limited by ALTQ,
773  * if_start is needed to drain packets from the queue in order
774  * to notify readers when outgoing packets become ready.
775  */
776 static void
777 tunstart(struct ifnet *ifp)
778 {
779         struct tun_softc *tp = ifp->if_softc;
780         struct mbuf *m;
781
782         if (!ifq_is_enabled(&ifp->if_snd))
783                 return;
784
785         m = ifq_poll(&ifp->if_snd);
786         if (m != NULL) {
787                 if (tp->tun_flags & TUN_RWAIT) {
788                         tp->tun_flags &= ~TUN_RWAIT;
789                         wakeup((caddr_t)tp);
790                 }
791                 if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio)
792                         pgsigio(tp->tun_sigio, SIGIO, 0);
793                 ifnet_deserialize_tx(ifp);
794                 KNOTE(&tp->tun_rkq.ki_note, 0);
795                 ifnet_serialize_tx(ifp);
796         }
797 }