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 "opt_inet.h"
  20 #include "opt_inet6.h"
  21
  22 #include <sys/param.h>
  23 #include <sys/proc.h>
  24 #include <sys/caps.h>
  25 #include <sys/systm.h>
  26 #include <sys/mbuf.h>
  27 #include <sys/socket.h>
  28 #include <sys/conf.h>
  29 #include <sys/device.h>
  30 #include <sys/filio.h>
  31 #include <sys/sockio.h>
  32 #include <sys/ttycom.h>
  33 #include <sys/signalvar.h>
  34 #include <sys/filedesc.h>
  35 #include <sys/kernel.h>
  36 #include <sys/sysctl.h>
  37 #include <sys/uio.h>
  38 #include <sys/vnode.h>
  39 #include <sys/malloc.h>
  40 #include <sys/mplock2.h>
  41 #include <sys/devfs.h>
  42 #include <sys/queue.h>
  43
  44 #include <net/bpf.h>
  45 #include <net/if.h>
  46 #include <net/if_types.h>
  47 #include <net/if_clone.h>
  48 #include <net/ifq_var.h>
  49 #include <net/netisr.h>
  50 #include <net/route.h>
  51
  52 #ifdef INET
  53 #include <netinet/in.h>
  54 #endif
  55
  56 #include "if_tunvar.h"
  57 #include "if_tun.h"
  58
  59 #define TUN             "tun"
  60 #define TUNDEBUG        if (tundebug) if_printf
  61
  62 /* module */
  63 static int              tunmodevent(module_t, int, void *);
  64
  65 /* device */
  66 static struct tun_softc *tuncreate(cdev_t, int);
  67 static void             tundestroy(struct tun_softc *sc);
  68
  69 /* clone */
  70 static int              tun_clone_create(struct if_clone *, int,
  71                                          caddr_t, caddr_t);
  72 static int              tun_clone_destroy(struct ifnet *);
  73
  74 /* network interface */
  75 static int              tunifinit(struct ifnet *);
  76 static void             tunifstart(struct ifnet *, struct ifaltq_subque *);
  77 static int              tunifoutput(struct ifnet *, struct mbuf *,
  78                                     struct sockaddr *, struct rtentry *rt);
  79 static int              tunifioctl(struct ifnet *, u_long,
  80                                    caddr_t, struct ucred *);
  81
  82 /* character device */
  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 static d_clone_t        tunclone;
  90
  91 static struct dev_ops   tun_ops = {
  92         { TUN, 0, 0 },
  93         .d_open =       tunopen,
  94         .d_close =      tunclose,
  95         .d_read =       tunread,
  96         .d_write =      tunwrite,
  97         .d_ioctl =      tunioctl,
  98         .d_kqfilter =   tunkqfilter
  99 };
 100
 101 /* kqueue support */
 102 static void             tun_filter_detach(struct knote *);
 103 static int              tun_filter_read(struct knote *, long);
 104 static int              tun_filter_write(struct knote *, long);
 105
 106 static struct filterops tun_read_filtops = {
 107         FILTEROP_ISFD,
 108         NULL,
 109         tun_filter_detach,
 110         tun_filter_read
 111 };
 112 static struct filterops tun_write_filtops = {
 113         FILTEROP_ISFD,
 114         NULL,
 115         tun_filter_detach,
 116         tun_filter_write
 117 };
 118
 119 static int              tundebug = 0;   /* debug flag */
 120 static int              tunrefcnt = 0;  /* module reference counter */
 121
 122 static MALLOC_DEFINE(M_TUN, TUN, "Tunnel Interface");
 123
 124 static DEVFS_DEFINE_CLONE_BITMAP(tun);
 125
 126 struct if_clone tun_cloner = IF_CLONE_INITIALIZER(
 127         TUN, tun_clone_create, tun_clone_destroy, 0, IF_MAXUNIT);
 128
 129 static SLIST_HEAD(,tun_softc) tun_listhead =
 130         SLIST_HEAD_INITIALIZER(&tun_listhead);
 131
 132 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0,
 133            "Enable debug output");
 134 SYSCTL_DECL(_net_link);
 135 SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW, 0,
 136             "IP tunnel software network interface");
 137 SYSCTL_INT(_net_link_tun, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0,
 138            "Enable debug output");
 139 SYSCTL_INT(_net_link_tun, OID_AUTO, refcnt, CTLFLAG_RD, &tunrefcnt, 0,
 140            "Number of opened devices");
 141
 142 DEV_MODULE(if_tun, tunmodevent, NULL);
 143
 144 /*
 145  * tunmodevent - module event handler
 146  */
 147 static int
 148 tunmodevent(module_t mod, int type, void *data)
 149 {
 150         static cdev_t dev = NULL;
 151         struct tun_softc *sc, *sc_tmp;
 152
 153         switch (type) {
 154         case MOD_LOAD:
 155                 dev = make_autoclone_dev(&tun_ops, &DEVFS_CLONE_BITMAP(tun),
 156                                          tunclone, UID_UUCP, GID_DIALER,
 157                                          0600, TUN);
 158
 159                 SLIST_INIT(&tun_listhead);
 160                 if_clone_attach(&tun_cloner);
 161                 break;
 162
 163         case MOD_UNLOAD:
 164                 if (tunrefcnt > 0)
 165                         return (EBUSY);
 166
 167                 if_clone_detach(&tun_cloner);
 168
 169                 SLIST_FOREACH_MUTABLE(sc, &tun_listhead, tun_link, sc_tmp)
 170                         tundestroy(sc);
 171
 172                 dev_ops_remove_all(&tun_ops);
 173                 destroy_autoclone_dev(dev, &DEVFS_CLONE_BITMAP(tun));
 174                 break;
 175
 176         default:
 177                 return (EOPNOTSUPP);
 178         }
 179
 180         return (0);
 181 }
 182
 183 static int
 184 tunclone(struct dev_clone_args *ap)
 185 {
 186         char ifname[IFNAMSIZ];
 187         int unit;
 188
 189         unit = devfs_clone_bitmap_get(&DEVFS_CLONE_BITMAP(tun), 0);
 190         ksnprintf(ifname, IFNAMSIZ, "%s%d", TUN, unit);
 191         ap->a_dev = make_only_dev(&tun_ops, unit, UID_UUCP, GID_DIALER,
 192                                   0600, "%s", ifname);
 193
 194         /*
 195          * Use the if_clone framework to create cloned device/interface,
 196          * so the two clone methods (autoclone device /dev/tun; ifconfig
 197          * clone) are consistent and can be mix used.
 198          *
 199          * Need to pass the cdev_t because the device created by
 200          * 'make_only_dev()' doesn't appear in '/dev' yet so that it can't
 201          * be found by 'devfs_find_device_by_name()' in 'tun_clone_create()'.
 202          */
 203         return (if_clone_create(ifname, IFNAMSIZ, NULL, (caddr_t)ap->a_dev));
 204 }
 205
 206 static struct tun_softc *
 207 tuncreate(cdev_t dev, int flags)
 208 {
 209         struct tun_softc *sc;
 210         struct ifnet *ifp;
 211         int unit = minor(dev);
 212
 213         sc = kmalloc(sizeof(*sc), M_TUN, M_WAITOK | M_ZERO);
 214         dev->si_drv1 = sc;
 215         sc->tun_dev = dev;
 216         sc->tun_flags = TUN_INITED;
 217         sc->tun_flags |= flags;
 218
 219         reference_dev(dev);  /* device association */
 220
 221         ifp = sc->tun_ifp = if_alloc(IFT_PPP);
 222         if (ifp == NULL) {
 223                 kprintf("%s: failed to if_alloc() interface for %s%d",
 224                         __func__, TUN, unit);
 225                 return (NULL);
 226         }
 227
 228         if_initname(ifp, TUN, unit);
 229         ifp->if_mtu = TUNMTU;
 230         ifp->if_ioctl = tunifioctl;
 231         ifp->if_output = tunifoutput;
 232         ifp->if_start = tunifstart;
 233         ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST;
 234         ifp->if_softc = sc;
 235         ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
 236         ifq_set_ready(&ifp->if_snd);
 237
 238         if_attach(ifp, NULL);
 239         bpfattach(ifp, DLT_NULL, sizeof(uint32_t));
 240
 241         SLIST_INSERT_HEAD(&tun_listhead, sc, tun_link);
 242         TUNDEBUG(ifp, "created, minor = %#x, flags = 0x%x\n",
 243                  unit, sc->tun_flags);
 244         return (sc);
 245 }
 246
 247 static void
 248 tundestroy(struct tun_softc *sc)
 249 {
 250         cdev_t dev = sc->tun_dev;
 251         struct ifnet *ifp = sc->tun_ifp;
 252         int unit = minor(dev);
 253
 254         TUNDEBUG(ifp, "destroyed, minor = %#x. Module refcnt = %d\n",
 255                  unit, tunrefcnt);
 256
 257         bpfdetach(ifp);
 258         if_detach(ifp);
 259         if_free(ifp);
 260
 261         sc->tun_dev = NULL;
 262         dev->si_drv1 = NULL;
 263         release_dev(dev);  /* device disassociation */
 264
 265         /* Also destroy the cloned device */
 266         destroy_dev(dev);
 267         devfs_clone_bitmap_put(&DEVFS_CLONE_BITMAP(tun), unit);
 268
 269         SLIST_REMOVE(&tun_listhead, sc, tun_softc, tun_link);
 270         kfree(sc, M_TUN);
 271 }
 272
 273 /*
 274  * tunnel open - must be superuser & the device must be configured in
 275  */
 276 static int
 277 tunopen(struct dev_open_args *ap)
 278 {
 279         cdev_t dev = ap->a_head.a_dev;
 280         struct tun_softc *sc;
 281         int error;
 282
 283         if ((error = caps_priv_check(ap->a_cred, SYSCAP_RESTRICTEDROOT)) != 0)
 284                 return (error);
 285
 286         sc = dev->si_drv1;
 287         if (sc == NULL && (sc = tuncreate(dev, 0)) == NULL)
 288                 return (ENOMEM);
 289         if (sc->tun_flags & TUN_OPEN)
 290                 return (EBUSY);
 291
 292         sc->tun_pid = curproc->p_pid;
 293         sc->tun_flags |= TUN_OPEN;
 294         tunrefcnt++;
 295
 296         TUNDEBUG(sc->tun_ifp, "opened, minor = %#x. Module refcnt = %d\n",
 297                  minor(dev), tunrefcnt);
 298         return (0);
 299 }
 300
 301 /*
 302  * close the device - mark interface down & delete routing info
 303  */
 304 static int
 305 tunclose(struct dev_close_args *ap)
 306 {
 307         cdev_t dev = ap->a_head.a_dev;
 308         struct tun_softc *sc = dev->si_drv1;
 309         struct ifnet *ifp;
 310         int unit = minor(dev);
 311         char ifname[IFNAMSIZ];
 312
 313         KASSERT(sc != NULL,
 314                 ("try closing the already destroyed %s%d", TUN, unit));
 315         ifp = sc->tun_ifp;
 316
 317         sc->tun_flags &= ~TUN_OPEN;
 318         sc->tun_pid = 0;
 319
 320         /* Junk all pending output. */
 321         ifq_purge_all(&ifp->if_snd);
 322
 323         if (ifp->if_flags & IFF_UP)
 324                 if_down(ifp);
 325         ifp->if_flags &= ~IFF_RUNNING;
 326
 327         if ((sc->tun_flags & TUN_MANUALMAKE) == 0) {
 328                 if_purgeaddrs_nolink(ifp);
 329
 330                 EVENTHANDLER_INVOKE(ifnet_detach_event, ifp);
 331
 332                 /* Announce the departure of the interface. */
 333                 rt_ifannouncemsg(ifp, IFAN_DEPARTURE);
 334         }
 335
 336         funsetown(&sc->tun_sigio);
 337         KNOTE(&sc->tun_rkq.ki_note, 0);
 338
 339         tunrefcnt--;
 340         if (tunrefcnt < 0) {
 341                 tunrefcnt = 0;
 342                 if_printf(ifp, ". Module refcnt = %d is out of sync! "
 343                           "Force refcnt to be 0.\n", tunrefcnt);
 344         }
 345
 346         TUNDEBUG(ifp, "closed, minor = %#x. Module refcnt = %d\n",
 347                  unit, tunrefcnt);
 348
 349         /* Only auto-destroy if the interface was not manually created. */
 350         if ((sc->tun_flags & TUN_MANUALMAKE) == 0) {
 351                 ksnprintf(ifname, IFNAMSIZ, "%s%d", TUN, unit);
 352                 if_clone_destroy(ifname);
 353         }
 354
 355         return (0);
 356 }
 357
 358
 359 /*
 360  * Interface clone support
 361  *
 362  * Create and destroy tun device/interface via ifconfig(8).
 363  */
 364
 365 static struct tun_softc *
 366 tunfind(int unit)
 367 {
 368         struct tun_softc *sc;
 369
 370         SLIST_FOREACH(sc, &tun_listhead, tun_link) {
 371                 if (minor(sc->tun_dev) == unit)
 372                         return (sc);
 373         }
 374         return (NULL);
 375 }
 376
 377 static int
 378 tun_clone_create(struct if_clone *ifc __unused, int unit,
 379                  caddr_t params __unused, caddr_t data)
 380 {
 381         struct tun_softc *sc;
 382         cdev_t dev = (cdev_t)data;
 383         int flags;
 384
 385         if (tunfind(unit) != NULL)
 386                 return (EEXIST);
 387
 388         if (dev == NULL) {
 389                 if (!devfs_clone_bitmap_chk(&DEVFS_CLONE_BITMAP(tun), unit)) {
 390                         devfs_clone_bitmap_set(&DEVFS_CLONE_BITMAP(tun), unit);
 391                         dev = make_dev(&tun_ops, unit, UID_UUCP, GID_DIALER,
 392                                        0600, "%s%d", TUN, unit);
 393                 } else {
 394                         dev = devfs_find_device_by_name("%s%d", TUN, unit);
 395                         if (dev == NULL)
 396                                 return (ENOENT);
 397                 }
 398                 flags = TUN_MANUALMAKE;
 399         } else {
 400                 flags = 0;
 401         }
 402
 403         if ((sc = tuncreate(dev, flags)) == NULL)
 404                 return (ENOMEM);
 405
 406         sc->tun_flags |= TUN_CLONE;
 407
 408         TUNDEBUG(sc->tun_ifp, "clone created, minor = %#x, flags = 0x%x\n",
 409                  minor(sc->tun_dev), sc->tun_flags);
 410
 411         return (0);
 412 }
 413
 414 static int
 415 tun_clone_destroy(struct ifnet *ifp)
 416 {
 417         struct tun_softc *sc = ifp->if_softc;
 418
 419         if (sc->tun_flags & TUN_OPEN)
 420                 return (EBUSY);
 421         if ((sc->tun_flags & TUN_CLONE) == 0)
 422                 return (EINVAL);
 423
 424         TUNDEBUG(ifp, "clone destroyed, minor = %#x, flags = 0x%x\n",
 425                  minor(sc->tun_dev), sc->tun_flags);
 426         tundestroy(sc);
 427
 428         return (0);
 429 }
 430
 431
 432 /*
 433  * Network interface functions
 434  */
 435
 436 static int
 437 tunifinit(struct ifnet *ifp)
 438 {
 439 #ifdef INET
 440         struct tun_softc *sc = ifp->if_softc;
 441 #endif
 442         struct ifaddr_container *ifac;
 443         int error = 0;
 444
 445         TUNDEBUG(ifp, "initialize\n");
 446
 447         ifp->if_flags |= IFF_UP | IFF_RUNNING;
 448         getmicrotime(&ifp->if_lastchange);
 449
 450         TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
 451                 struct ifaddr *ifa = ifac->ifa;
 452
 453                 if (ifa->ifa_addr == NULL) {
 454                         error = EFAULT;
 455                         /* XXX: Should maybe return straight off? */
 456                 } else {
 457 #ifdef INET
 458                         if (ifa->ifa_addr->sa_family == AF_INET) {
 459                                 struct sockaddr_in *si;
 460
 461                                 si = (struct sockaddr_in *)ifa->ifa_addr;
 462                                 if (si->sin_addr.s_addr)
 463                                         sc->tun_flags |= TUN_IASET;
 464                         }
 465 #endif
 466                 }
 467         }
 468         return (error);
 469 }
 470
 471 /*
 472  * Process an ioctl request.
 473  *
 474  * MPSAFE
 475  */
 476 static int
 477 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
 478 {
 479         struct ifreq *ifr = (struct ifreq *)data;
 480         struct tun_softc *sc = ifp->if_softc;
 481         struct ifstat *ifs;
 482         int error = 0;
 483
 484         switch (cmd) {
 485         case SIOCGIFSTATUS:
 486                 ifs = (struct ifstat *)data;
 487                 if (sc->tun_pid)
 488                         ksprintf(ifs->ascii + strlen(ifs->ascii),
 489                             "\tOpened by PID %d\n", sc->tun_pid);
 490                 break;
 491         case SIOCSIFADDR:
 492                 error = tunifinit(ifp);
 493                 TUNDEBUG(ifp, "address set, error=%d\n", error);
 494                 break;
 495         case SIOCSIFDSTADDR:
 496                 error = tunifinit(ifp);
 497                 TUNDEBUG(ifp, "destination address set, error=%d\n", error);
 498                 break;
 499         case SIOCSIFMTU:
 500                 ifp->if_mtu = ifr->ifr_mtu;
 501                 TUNDEBUG(ifp, "mtu set\n");
 502                 break;
 503         case SIOCSIFFLAGS:
 504         case SIOCADDMULTI:
 505         case SIOCDELMULTI:
 506                 break;
 507         default:
 508                 error = EINVAL;
 509         }
 510         return (error);
 511 }
 512
 513 /*
 514  * Start packet transmission on the interface.
 515  * when the interface queue is rate-limited by ALTQ,
 516  * if_start is needed to drain packets from the queue in order
 517  * to notify readers when outgoing packets become ready.
 518  */
 519 static void
 520 tunifstart(struct ifnet *ifp, struct ifaltq_subque *ifsq)
 521 {
 522         struct tun_softc *sc = ifp->if_softc;
 523         struct mbuf *m;
 524
 525         ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
 526
 527         if (!ifq_is_enabled(&ifp->if_snd))
 528                 return;
 529
 530         m = ifsq_poll(ifsq);
 531         if (m != NULL) {
 532                 if (sc->tun_flags & TUN_RWAIT) {
 533                         sc->tun_flags &= ~TUN_RWAIT;
 534                         wakeup((caddr_t)sc);
 535                 }
 536                 if (sc->tun_flags & TUN_ASYNC && sc->tun_sigio)
 537                         pgsigio(sc->tun_sigio, SIGIO, 0);
 538                 ifsq_deserialize_hw(ifsq);
 539                 KNOTE(&sc->tun_rkq.ki_note, 0);
 540                 ifsq_serialize_hw(ifsq);
 541         }
 542 }
 543
 544 /*
 545  * tunifoutput - queue packets from higher level ready to put out.
 546  *
 547  * MPSAFE
 548  */
 549 static int
 550 tunifoutput_serialized(struct ifnet *ifp, struct mbuf *m0,
 551                        struct sockaddr *dst, struct rtentry *rt)
 552 {
 553         struct tun_softc *sc = ifp->if_softc;
 554         int error;
 555         struct altq_pktattr pktattr;
 556
 557         TUNDEBUG(ifp, "output\n");
 558
 559         if ((sc->tun_flags & TUN_READY) != TUN_READY) {
 560                 TUNDEBUG(ifp, "not ready, flags = 0x%x\n", sc->tun_flags);
 561                 m_freem(m0);
 562                 return (EHOSTDOWN);
 563         }
 564
 565         /*
 566          * if the queueing discipline needs packet classification,
 567          * do it before prepending link headers.
 568          */
 569         ifq_classify(&ifp->if_snd, m0, dst->sa_family, &pktattr);
 570
 571         /* BPF write needs to be handled specially */
 572         if (dst->sa_family == AF_UNSPEC) {
 573                 dst->sa_family = *(mtod(m0, int *));
 574                 m0->m_len -= sizeof(int);
 575                 m0->m_pkthdr.len -= sizeof(int);
 576                 m0->m_data += sizeof(int);
 577         }
 578
 579         if (ifp->if_bpf) {
 580                 bpf_gettoken();
 581                 if (ifp->if_bpf) {
 582                         /*
 583                          * We need to prepend the address family as
 584                          * a four byte field.
 585                          */
 586                         uint32_t af = dst->sa_family;
 587
 588                         bpf_ptap(ifp->if_bpf, m0, &af, sizeof(af));
 589                 }
 590                 bpf_reltoken();
 591         }
 592
 593         /* prepend sockaddr? this may abort if the mbuf allocation fails */
 594         if (sc->tun_flags & TUN_LMODE) {
 595                 /* allocate space for sockaddr */
 596                 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
 597
 598                 /* if allocation failed drop packet */
 599                 if (m0 == NULL){
 600                         IFNET_STAT_INC(ifp, oerrors, 1);
 601                         return (ENOBUFS);
 602                 } else {
 603                         bcopy(dst, m0->m_data, dst->sa_len);
 604                 }
 605         }
 606
 607         if (sc->tun_flags & TUN_IFHEAD) {
 608                 /* Prepend the address family */
 609                 M_PREPEND(m0, 4, M_NOWAIT);
 610
 611                 /* if allocation failed drop packet */
 612                 if (m0 == NULL){
 613                         IFNET_STAT_INC(ifp, oerrors, 1);
 614                         return (ENOBUFS);
 615                 } else {
 616                         *(u_int32_t *)m0->m_data = htonl(dst->sa_family);
 617                 }
 618         } else {
 619 #ifdef INET
 620                 if (dst->sa_family != AF_INET)
 621 #endif
 622                 {
 623                         m_freem(m0);
 624                         return (EAFNOSUPPORT);
 625                 }
 626         }
 627
 628         error = ifq_handoff(ifp, m0, &pktattr);
 629         if (error) {
 630                 IFNET_STAT_INC(ifp, collisions, 1);
 631         } else {
 632                 IFNET_STAT_INC(ifp, opackets, 1);
 633                 if (sc->tun_flags & TUN_RWAIT) {
 634                         sc->tun_flags &= ~TUN_RWAIT;
 635                         wakeup((caddr_t)sc);
 636                 }
 637                 get_mplock();
 638                 if (sc->tun_flags & TUN_ASYNC && sc->tun_sigio)
 639                         pgsigio(sc->tun_sigio, SIGIO, 0);
 640                 rel_mplock();
 641                 ifnet_deserialize_all(ifp);
 642                 KNOTE(&sc->tun_rkq.ki_note, 0);
 643                 ifnet_serialize_all(ifp);
 644         }
 645         return (error);
 646 }
 647
 648 static int
 649 tunifoutput(struct ifnet *ifp, struct mbuf *m0, struct sockaddr *dst,
 650             struct rtentry *rt)
 651 {
 652         int error;
 653
 654         ifnet_serialize_all(ifp);
 655         error = tunifoutput_serialized(ifp, m0, dst, rt);
 656         ifnet_deserialize_all(ifp);
 657
 658         return (error);
 659 }
 660
 661
 662 /*
 663  * the ops interface is now pretty minimal.
 664  */
 665 static int
 666 tunioctl(struct dev_ioctl_args *ap)
 667 {
 668         cdev_t dev = ap->a_head.a_dev;
 669         caddr_t data = ap->a_data;
 670         struct tun_softc *sc = dev->si_drv1;
 671         struct ifnet *ifp = sc->tun_ifp;
 672         struct ifreq *ifr;
 673         struct tuninfo *tunp;
 674         int error = 0;
 675
 676         switch (ap->a_cmd) {
 677         case TUNSIFINFO:
 678                 tunp = (struct tuninfo *)data;
 679                 if (ifp->if_type != tunp->type)
 680                         return (EPROTOTYPE);
 681                 if (tunp->mtu < IF_MINMTU)
 682                         return (EINVAL);
 683                 ifp->if_mtu = tunp->mtu;
 684                 ifp->if_baudrate = tunp->baudrate;
 685                 break;
 686
 687         case TUNGIFINFO:
 688                 tunp = (struct tuninfo *)data;
 689                 tunp->mtu = ifp->if_mtu;
 690                 tunp->type = ifp->if_type;
 691                 tunp->baudrate = ifp->if_baudrate;
 692                 break;
 693
 694         case TUNGIFNAME:
 695                 ifr = (struct ifreq *)data;
 696                 strlcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ);
 697                 break;
 698
 699         case TUNSDEBUG:
 700                 tundebug = *(int *)data;
 701                 break;
 702
 703         case TUNGDEBUG:
 704                 *(int *)data = tundebug;
 705                 break;
 706
 707         case TUNSLMODE:
 708                 if (*(int *)data) {
 709                         sc->tun_flags |= TUN_LMODE;
 710                         sc->tun_flags &= ~TUN_IFHEAD;
 711                 } else {
 712                         sc->tun_flags &= ~TUN_LMODE;
 713                 }
 714                 break;
 715
 716         case TUNSIFHEAD:
 717                 if (*(int *)data) {
 718                         sc->tun_flags |= TUN_IFHEAD;
 719                         sc->tun_flags &= ~TUN_LMODE;
 720                 } else {
 721                         sc->tun_flags &= ~TUN_IFHEAD;
 722                 }
 723                 break;
 724
 725         case TUNGIFHEAD:
 726                 *(int *)data = (sc->tun_flags & TUN_IFHEAD) ? 1 : 0;
 727                 break;
 728
 729         case TUNSIFMODE:
 730                 /* deny this if UP */
 731                 if (ifp->if_flags & IFF_UP)
 732                         return (EBUSY);
 733
 734                 switch (*(int *)data & ~IFF_MULTICAST) {
 735                 case IFF_POINTOPOINT:
 736                 case IFF_BROADCAST:
 737                         ifp->if_flags &= ~(IFF_BROADCAST | IFF_POINTOPOINT);
 738                         ifp->if_flags |= *(int *)data;
 739                         break;
 740                 default:
 741                         return (EINVAL);
 742                 }
 743                 break;
 744
 745         case TUNSIFPID:
 746                 sc->tun_pid = curproc->p_pid;
 747                 break;
 748
 749         case FIOASYNC:
 750                 if (*(int *)data)
 751                         sc->tun_flags |= TUN_ASYNC;
 752                 else
 753                         sc->tun_flags &= ~TUN_ASYNC;
 754                 break;
 755
 756         case FIONREAD:
 757                 *(int *)data = ifsq_poll_pktlen(
 758                     ifq_get_subq_default(&ifp->if_snd));
 759                 break;
 760
 761         case FIOSETOWN:
 762                 error = fsetown(*(int *)data, &sc->tun_sigio);
 763                 break;
 764
 765         case FIOGETOWN:
 766                 *(int *)data = fgetown(&sc->tun_sigio);
 767                 break;
 768
 769         /* This is deprecated, FIOSETOWN should be used instead. */
 770         case TIOCSPGRP:
 771                 error = fsetown(-(*(int *)data), &sc->tun_sigio);
 772                 break;
 773
 774         /* This is deprecated, FIOGETOWN should be used instead. */
 775         case TIOCGPGRP:
 776                 *(int *)data = -fgetown(&sc->tun_sigio);
 777                 break;
 778
 779         default:
 780                 error = ENOTTY;
 781                 break;
 782         }
 783
 784         return (error);
 785 }
 786
 787 /*
 788  * The ops read interface - reads a packet at a time, or at
 789  * least as much of a packet as can be read.
 790  */
 791 static  int
 792 tunread(struct dev_read_args *ap)
 793 {
 794         cdev_t dev = ap->a_head.a_dev;
 795         struct uio *uio = ap->a_uio;
 796         struct tun_softc *sc = dev->si_drv1;
 797         struct ifnet *ifp = sc->tun_ifp;
 798         struct ifaltq_subque *ifsq = ifq_get_subq_default(&ifp->if_snd);
 799         struct mbuf *m0;
 800         int error=0, len;
 801
 802         TUNDEBUG(ifp, "read\n");
 803         if ((sc->tun_flags & TUN_READY) != TUN_READY) {
 804                 TUNDEBUG(ifp, "not ready, flags = 0x%x\n", sc->tun_flags);
 805                 return (EHOSTDOWN);
 806         }
 807
 808         sc->tun_flags &= ~TUN_RWAIT;
 809
 810         ifnet_serialize_all(ifp);
 811
 812         while ((m0 = ifsq_dequeue(ifsq)) == NULL) {
 813                 if (ap->a_ioflag & IO_NDELAY) {
 814                         ifnet_deserialize_all(ifp);
 815                         return (EWOULDBLOCK);
 816                 }
 817                 sc->tun_flags |= TUN_RWAIT;
 818                 ifnet_deserialize_all(ifp);
 819                 if ((error = tsleep(sc, PCATCH, "tunread", 0)) != 0)
 820                         return (error);
 821                 ifnet_serialize_all(ifp);
 822         }
 823
 824         ifnet_deserialize_all(ifp);
 825
 826         while (m0 && uio->uio_resid > 0 && error == 0) {
 827                 len = (int)szmin(uio->uio_resid, m0->m_len);
 828                 if (len != 0)
 829                         error = uiomove(mtod(m0, caddr_t), (size_t)len, uio);
 830                 m0 = m_free(m0);
 831         }
 832
 833         if (m0) {
 834                 TUNDEBUG(ifp, "dropping mbuf\n");
 835                 m_freem(m0);
 836         }
 837         return (error);
 838 }
 839
 840 /*
 841  * the ops write interface - an atomic write is a packet - or else!
 842  */
 843 static  int
 844 tunwrite(struct dev_write_args *ap)
 845 {
 846         cdev_t dev = ap->a_head.a_dev;
 847         struct uio *uio = ap->a_uio;
 848         struct tun_softc *sc = dev->si_drv1;
 849         struct ifnet *ifp = sc->tun_ifp;
 850         struct mbuf *top, **mp, *m;
 851         size_t tlen;
 852         uint32_t family, mru;
 853         int error = 0;
 854         int isr;
 855
 856         TUNDEBUG(ifp, "tunwrite\n");
 857
 858         if (uio->uio_resid == 0)
 859                 return (0);
 860
 861         mru = TUNMRU;
 862         if (sc->tun_flags & TUN_IFHEAD)
 863                 mru += sizeof(family);
 864         if (uio->uio_resid > mru) {
 865                 TUNDEBUG(ifp, "len = %zd!\n", uio->uio_resid);
 866                 return (EIO);
 867         }
 868
 869         /* get a header mbuf */
 870         MGETHDR(m, M_WAITOK, MT_DATA);
 871
 872         tlen = uio->uio_resid;
 873         top = NULL;
 874         mp = &top;
 875         while (error == 0 && uio->uio_resid > 0) {
 876                 m->m_len = (int)szmin(MHLEN, uio->uio_resid);
 877                 error = uiomove(mtod(m, caddr_t), (size_t)m->m_len, uio);
 878                 *mp = m;
 879                 mp = &m->m_next;
 880                 if (uio->uio_resid > 0)
 881                         MGET(m, M_WAITOK, MT_DATA);
 882         }
 883         if (error) {
 884                 if (top)
 885                         m_freem(top);
 886                 IFNET_STAT_INC(ifp, ierrors, 1);
 887                 return (error);
 888         }
 889
 890         top->m_pkthdr.len = (int)tlen;
 891         top->m_pkthdr.rcvif = ifp;
 892
 893         if (ifp->if_bpf) {
 894                 bpf_gettoken();
 895
 896                 if (ifp->if_bpf) {
 897                         if (sc->tun_flags & TUN_IFHEAD) {
 898                                 /*
 899                                  * Conveniently, we already have a 4-byte
 900                                  * address family prepended to our packet !
 901                                  * Inconveniently, it's in the wrong byte
 902                                  * order !
 903                                  */
 904                                 if ((top = m_pullup(top, sizeof(family)))
 905                                     == NULL) {
 906                                         bpf_reltoken();
 907                                         return (ENOBUFS);
 908                                 }
 909                                 *mtod(top, u_int32_t *) =
 910                                     ntohl(*mtod(top, u_int32_t *));
 911                                 bpf_mtap(ifp->if_bpf, top);
 912                                 *mtod(top, u_int32_t *) =
 913                                     htonl(*mtod(top, u_int32_t *));
 914                         } else {
 915                                 /*
 916                                  * We need to prepend the address family as
 917                                  * a four byte field.
 918                                  */
 919                                 static const uint32_t af = AF_INET;
 920
 921                                 bpf_ptap(ifp->if_bpf, top, &af, sizeof(af));
 922                         }
 923                 }
 924
 925                 bpf_reltoken();
 926         }
 927
 928         if (sc->tun_flags & TUN_IFHEAD) {
 929                 if (top->m_len < sizeof(family) &&
 930                     (top = m_pullup(top, sizeof(family))) == NULL)
 931                                 return (ENOBUFS);
 932                 family = ntohl(*mtod(top, u_int32_t *));
 933                 m_adj(top, sizeof(family));
 934         } else {
 935                 family = AF_INET;
 936         }
 937
 938         IFNET_STAT_INC(ifp, ibytes, top->m_pkthdr.len);
 939         IFNET_STAT_INC(ifp, ipackets, 1);
 940
 941         switch (family) {
 942 #ifdef INET
 943         case AF_INET:
 944                 isr = NETISR_IP;
 945                 break;
 946 #endif
 947 #ifdef INET6
 948         case AF_INET6:
 949                 isr = NETISR_IPV6;
 950                 break;
 951 #endif
 952         default:
 953                 m_freem(m);
 954                 return (EAFNOSUPPORT);
 955         }
 956
 957         netisr_queue(isr, top);
 958         return (0);
 959 }
 960
 961
 962 /*
 963  * tunkqfilter - support for the kevent() system call.
 964  */
 965 static int
 966 tunkqfilter(struct dev_kqfilter_args *ap)
 967 {
 968         cdev_t dev = ap->a_head.a_dev;
 969         struct tun_softc *sc = dev->si_drv1;
 970         struct ifnet *ifp = sc->tun_ifp;
 971         struct knote *kn = ap->a_kn;
 972         struct klist *klist;
 973
 974         ap->a_result = 0;
 975         ifnet_serialize_all(ifp);
 976
 977         switch (kn->kn_filter) {
 978         case EVFILT_READ:
 979                 kn->kn_fop = &tun_read_filtops;
 980                 kn->kn_hook = (caddr_t)sc;
 981                 break;
 982         case EVFILT_WRITE:
 983                 kn->kn_fop = &tun_write_filtops;
 984                 kn->kn_hook = (caddr_t)sc;
 985                 break;
 986         default:
 987                 ifnet_deserialize_all(ifp);
 988                 ap->a_result = EOPNOTSUPP;
 989                 return (0);
 990         }
 991
 992         klist = &sc->tun_rkq.ki_note;
 993         knote_insert(klist, kn);
 994         ifnet_deserialize_all(ifp);
 995
 996         return (0);
 997 }
 998
 999 static void
1000 tun_filter_detach(struct knote *kn)
1001 {
1002         struct tun_softc *sc = (struct tun_softc *)kn->kn_hook;
1003         struct klist *klist = &sc->tun_rkq.ki_note;
1004
1005         knote_remove(klist, kn);
1006 }
1007
1008 static int
1009 tun_filter_write(struct knote *kn, long hint)
1010 {
1011         /* Always ready for a write */
1012         return (1);
1013 }
1014
1015 static int
1016 tun_filter_read(struct knote *kn, long hint)
1017 {
1018         struct tun_softc *sc = (struct tun_softc *)kn->kn_hook;
1019         struct ifnet *ifp = sc->tun_ifp;
1020         int ready = 0;
1021
1022         ifnet_serialize_all(ifp);
1023         if (!ifsq_is_empty(ifq_get_subq_default(&ifp->if_snd))) {
1024                 if ((kn->kn_sfflags & NOTE_HUPONLY) == 0)
1025                         ready = 1;
1026         }
1027         ifnet_deserialize_all(ifp);
1028
1029         return (ready);
1030 }