sys/net/if_tuntap.c

   1 /*      $NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $  */
   2 /*-
   3  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   4  *
   5  * Copyright (C) 1999-2000 by Maksim Yevmenkin <m_evmenkin@yahoo.com>
   6  * All rights reserved.
   7  * Copyright (c) 2019 Kyle Evans <kevans@FreeBSD.org>
   8  *
   9  * Redistribution and use in source and binary forms, with or without
  10  * modification, are permitted provided that the following conditions
  11  * are met:
  12  *
  13  * 1. Redistributions of source code must retain the above copyright
  14  *    notice, this list of conditions and the following disclaimer.
  15  * 2. Redistributions in binary form must reproduce the above copyright
  16  *    notice, this list of conditions and the following disclaimer in the
  17  *    documentation and/or other materials provided with the distribution.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  * BASED ON:
  32  * -------------------------------------------------------------------------
  33  *
  34  * Copyright (c) 1988, Julian Onions <jpo@cs.nott.ac.uk>
  35  * Nottingham University 1987.
  36  *
  37  * This source may be freely distributed, however I would be interested
  38  * in any changes that are made.
  39  *
  40  * This driver takes packets off the IP i/f and hands them up to a
  41  * user process to have its wicked way with. This driver has it's
  42  * roots in a similar driver written by Phil Cockcroft (formerly) at
  43  * UCL. This driver is based much more on read/write/poll mode of
  44  * operation though.
  45  *
  46  * $FreeBSD$
  47  */
  48
  49 #include "opt_inet.h"
  50 #include "opt_inet6.h"
  51
  52 #include <sys/param.h>
  53 #include <sys/lock.h>
  54 #include <sys/priv.h>
  55 #include <sys/proc.h>
  56 #include <sys/systm.h>
  57 #include <sys/jail.h>
  58 #include <sys/mbuf.h>
  59 #include <sys/module.h>
  60 #include <sys/socket.h>
  61 #include <sys/fcntl.h>
  62 #include <sys/filio.h>
  63 #include <sys/sockio.h>
  64 #include <sys/sx.h>
  65 #include <sys/syslog.h>
  66 #include <sys/ttycom.h>
  67 #include <sys/poll.h>
  68 #include <sys/selinfo.h>
  69 #include <sys/signalvar.h>
  70 #include <sys/filedesc.h>
  71 #include <sys/kernel.h>
  72 #include <sys/sysctl.h>
  73 #include <sys/conf.h>
  74 #include <sys/uio.h>
  75 #include <sys/malloc.h>
  76 #include <sys/random.h>
  77 #include <sys/ctype.h>
  78
  79 #include <net/ethernet.h>
  80 #include <net/if.h>
  81 #include <net/if_var.h>
  82 #include <net/if_clone.h>
  83 #include <net/if_dl.h>
  84 #include <net/if_media.h>
  85 #include <net/if_types.h>
  86 #include <net/netisr.h>
  87 #include <net/route.h>
  88 #include <net/vnet.h>
  89 #ifdef INET
  90 #include <netinet/in.h>
  91 #endif
  92 #include <net/bpf.h>
  93 #include <net/if_tap.h>
  94 #include <net/if_tun.h>
  95
  96 #include <sys/queue.h>
  97 #include <sys/condvar.h>
  98 #include <security/mac/mac_framework.h>
  99
 100 struct tuntap_driver;
 101
 102 /*
 103  * tun_list is protected by global tunmtx.  Other mutable fields are
 104  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
 105  * static for the duration of a tunnel interface.
 106  */
 107 struct tuntap_softc {
 108         TAILQ_ENTRY(tuntap_softc)        tun_list;
 109         struct cdev                     *tun_alias;
 110         struct cdev                     *tun_dev;
 111         u_short                          tun_flags;     /* misc flags */
 112 #define TUN_OPEN        0x0001
 113 #define TUN_INITED      0x0002
 114 #define TUN_UNUSED1     0x0008
 115 #define TUN_DSTADDR     0x0010
 116 #define TUN_LMODE       0x0020
 117 #define TUN_RWAIT       0x0040
 118 #define TUN_ASYNC       0x0080
 119 #define TUN_IFHEAD      0x0100
 120 #define TUN_DYING       0x0200
 121 #define TUN_L2          0x0400
 122 #define TUN_VMNET       0x0800
 123
 124 #define TUN_DRIVER_IDENT_MASK   (TUN_L2 | TUN_VMNET)
 125 #define TUN_READY               (TUN_OPEN | TUN_INITED)
 126
 127         pid_t                    tun_pid;       /* owning pid */
 128         struct ifnet            *tun_ifp;       /* the interface */
 129         struct sigio            *tun_sigio;     /* async I/O info */
 130         struct tuntap_driver    *tun_drv;       /* appropriate driver */
 131         struct selinfo           tun_rsel;      /* read select */
 132         struct mtx               tun_mtx;       /* softc field mutex */
 133         struct cv                tun_cv;        /* for ref'd dev destroy */
 134         struct ether_addr        tun_ether;     /* remote address */
 135         int                      tun_busy;      /* busy count */
 136 };
 137 #define TUN2IFP(sc)     ((sc)->tun_ifp)
 138
 139 #define TUNDEBUG        if (tundebug) if_printf
 140
 141 #define TUN_LOCK(tp)            mtx_lock(&(tp)->tun_mtx)
 142 #define TUN_UNLOCK(tp)          mtx_unlock(&(tp)->tun_mtx)
 143 #define TUN_LOCK_ASSERT(tp)     mtx_assert(&(tp)->tun_mtx, MA_OWNED);
 144
 145 #define TUN_VMIO_FLAG_MASK      0x0fff
 146
 147 /*
 148  * All mutable global variables in if_tun are locked using tunmtx, with
 149  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
 150  * which are static after setup.
 151  */
 152 static struct mtx tunmtx;
 153 static eventhandler_tag arrival_tag;
 154 static eventhandler_tag clone_tag;
 155 static const char tunname[] = "tun";
 156 static const char tapname[] = "tap";
 157 static const char vmnetname[] = "vmnet";
 158 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
 159 static int tundebug = 0;
 160 static int tundclone = 1;
 161 static int tap_allow_uopen = 0; /* allow user open() */
 162 static int tapuponopen = 0;     /* IFF_UP on open() */
 163 static int tapdclone = 1;       /* enable devfs cloning */
 164
 165 static TAILQ_HEAD(,tuntap_softc)        tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
 166 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
 167
 168 static struct sx tun_ioctl_sx;
 169 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
 170
 171 SYSCTL_DECL(_net_link);
 172 /* tun */
 173 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW, 0,
 174     "IP tunnel software network interface");
 175 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
 176     "Enable legacy devfs interface creation");
 177
 178 /* tap */
 179 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW, 0,
 180     "Ethernet tunnel software network interface");
 181 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
 182     "Allow user to open /dev/tap (based on node permissions)");
 183 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
 184     "Bring interface up when /dev/tap is opened");
 185 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
 186     "Enable legacy devfs interface creation");
 187 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
 188
 189 static int      tun_create_device(struct tuntap_driver *drv, int unit,
 190     struct ucred *cr, struct cdev **dev, const char *name);
 191 static int      tun_busy_locked(struct tuntap_softc *tp);
 192 static void     tun_unbusy_locked(struct tuntap_softc *tp);
 193 static int      tun_busy(struct tuntap_softc *tp);
 194 static void     tun_unbusy(struct tuntap_softc *tp);
 195
 196 static int      tuntap_name2info(const char *name, int *unit, int *flags);
 197 static void     tunclone(void *arg, struct ucred *cred, char *name,
 198                     int namelen, struct cdev **dev);
 199 static void     tuncreate(struct cdev *dev);
 200 static void     tundtor(void *data);
 201 static void     tunrename(void *arg, struct ifnet *ifp);
 202 static int      tunifioctl(struct ifnet *, u_long, caddr_t);
 203 static void     tuninit(struct ifnet *);
 204 static void     tunifinit(void *xtp);
 205 static int      tuntapmodevent(module_t, int, void *);
 206 static int      tunoutput(struct ifnet *, struct mbuf *,
 207                     const struct sockaddr *, struct route *ro);
 208 static void     tunstart(struct ifnet *);
 209 static void     tunstart_l2(struct ifnet *);
 210
 211 static int      tun_clone_match(struct if_clone *ifc, const char *name);
 212 static int      tap_clone_match(struct if_clone *ifc, const char *name);
 213 static int      vmnet_clone_match(struct if_clone *ifc, const char *name);
 214 static int      tun_clone_create(struct if_clone *, char *, size_t, caddr_t);
 215 static int      tun_clone_destroy(struct if_clone *, struct ifnet *);
 216
 217 static d_open_t         tunopen;
 218 static d_read_t         tunread;
 219 static d_write_t        tunwrite;
 220 static d_ioctl_t        tunioctl;
 221 static d_poll_t         tunpoll;
 222 static d_kqfilter_t     tunkqfilter;
 223
 224 static int              tunkqread(struct knote *, long);
 225 static int              tunkqwrite(struct knote *, long);
 226 static void             tunkqdetach(struct knote *);
 227
 228 static struct filterops tun_read_filterops = {
 229         .f_isfd =       1,
 230         .f_attach =     NULL,
 231         .f_detach =     tunkqdetach,
 232         .f_event =      tunkqread,
 233 };
 234
 235 static struct filterops tun_write_filterops = {
 236         .f_isfd =       1,
 237         .f_attach =     NULL,
 238         .f_detach =     tunkqdetach,
 239         .f_event =      tunkqwrite,
 240 };
 241
 242 static struct tuntap_driver {
 243         struct cdevsw            cdevsw;
 244         int                      ident_flags;
 245         struct unrhdr           *unrhdr;
 246         struct clonedevs        *clones;
 247         ifc_match_t             *clone_match_fn;
 248         ifc_create_t            *clone_create_fn;
 249         ifc_destroy_t           *clone_destroy_fn;
 250 } tuntap_drivers[] = {
 251         {
 252                 .ident_flags =  0,
 253                 .cdevsw =       {
 254                     .d_version =        D_VERSION,
 255                     .d_flags =          D_NEEDMINOR,
 256                     .d_open =           tunopen,
 257                     .d_read =           tunread,
 258                     .d_write =          tunwrite,
 259                     .d_ioctl =          tunioctl,
 260                     .d_poll =           tunpoll,
 261                     .d_kqfilter =       tunkqfilter,
 262                     .d_name =           tunname,
 263                 },
 264                 .clone_match_fn =       tun_clone_match,
 265                 .clone_create_fn =      tun_clone_create,
 266                 .clone_destroy_fn =     tun_clone_destroy,
 267         },
 268         {
 269                 .ident_flags =  TUN_L2,
 270                 .cdevsw =       {
 271                     .d_version =        D_VERSION,
 272                     .d_flags =          D_NEEDMINOR,
 273                     .d_open =           tunopen,
 274                     .d_read =           tunread,
 275                     .d_write =          tunwrite,
 276                     .d_ioctl =          tunioctl,
 277                     .d_poll =           tunpoll,
 278                     .d_kqfilter =       tunkqfilter,
 279                     .d_name =           tapname,
 280                 },
 281                 .clone_match_fn =       tap_clone_match,
 282                 .clone_create_fn =      tun_clone_create,
 283                 .clone_destroy_fn =     tun_clone_destroy,
 284         },
 285         {
 286                 .ident_flags =  TUN_L2 | TUN_VMNET,
 287                 .cdevsw =       {
 288                     .d_version =        D_VERSION,
 289                     .d_flags =          D_NEEDMINOR,
 290                     .d_open =           tunopen,
 291                     .d_read =           tunread,
 292                     .d_write =          tunwrite,
 293                     .d_ioctl =          tunioctl,
 294                     .d_poll =           tunpoll,
 295                     .d_kqfilter =       tunkqfilter,
 296                     .d_name =           vmnetname,
 297                 },
 298                 .clone_match_fn =       vmnet_clone_match,
 299                 .clone_create_fn =      tun_clone_create,
 300                 .clone_destroy_fn =     tun_clone_destroy,
 301         },
 302 };
 303
 304 struct tuntap_driver_cloner {
 305         SLIST_ENTRY(tuntap_driver_cloner)        link;
 306         struct tuntap_driver                    *drv;
 307         struct if_clone                         *cloner;
 308 };
 309
 310 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
 311     SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
 312
 313 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
 314
 315 /*
 316  * Mechanism for marking a tunnel device as busy so that we can safely do some
 317  * orthogonal operations (such as operations on devices) without racing against
 318  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
 319  * open, to be woken up when the condition is alleviated.
 320  */
 321 static int
 322 tun_busy_locked(struct tuntap_softc *tp)
 323 {
 324
 325         TUN_LOCK_ASSERT(tp);
 326         if ((tp->tun_flags & TUN_DYING) != 0) {
 327                 /*
 328                  * Perhaps unintuitive, but the device is busy going away.
 329                  * Other interpretations of EBUSY from tun_busy make little
 330                  * sense, since making a busy device even more busy doesn't
 331                  * sound like a problem.
 332                  */
 333                 return (EBUSY);
 334         }
 335
 336         ++tp->tun_busy;
 337         return (0);
 338 }
 339
 340 static void
 341 tun_unbusy_locked(struct tuntap_softc *tp)
 342 {
 343
 344         TUN_LOCK_ASSERT(tp);
 345         KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
 346
 347         --tp->tun_busy;
 348         /* Wake up anything that may be waiting on our busy tunnel. */
 349         if (tp->tun_busy == 0)
 350                 cv_broadcast(&tp->tun_cv);
 351 }
 352
 353 static int
 354 tun_busy(struct tuntap_softc *tp)
 355 {
 356         int ret;
 357
 358         TUN_LOCK(tp);
 359         ret = tun_busy_locked(tp);
 360         TUN_UNLOCK(tp);
 361         return (ret);
 362 }
 363
 364
 365 static void
 366 tun_unbusy(struct tuntap_softc *tp)
 367 {
 368
 369         TUN_LOCK(tp);
 370         tun_unbusy_locked(tp);
 371         TUN_UNLOCK(tp);
 372 }
 373
 374 /*
 375  * Sets unit and/or flags given the device name.  Must be called with correct
 376  * vnet context.
 377  */
 378 static int
 379 tuntap_name2info(const char *name, int *outunit, int *outflags)
 380 {
 381         struct tuntap_driver *drv;
 382         struct tuntap_driver_cloner *drvc;
 383         char *dname;
 384         int flags, unit;
 385         bool found;
 386
 387         if (name == NULL)
 388                 return (EINVAL);
 389
 390         /*
 391          * Needed for dev_stdclone, but dev_stdclone will not modify, it just
 392          * wants to be able to pass back a char * through the second param. We
 393          * will always set that as NULL here, so we'll fake it.
 394          */
 395         dname = __DECONST(char *, name);
 396         found = false;
 397
 398         KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
 399             ("tuntap_driver_cloners failed to initialize"));
 400         SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
 401                 KASSERT(drvc->drv != NULL,
 402                     ("tuntap_driver_cloners entry not properly initialized"));
 403                 drv = drvc->drv;
 404
 405                 if (strcmp(name, drv->cdevsw.d_name) == 0) {
 406                         found = true;
 407                         unit = -1;
 408                         flags = drv->ident_flags;
 409                         break;
 410                 }
 411
 412                 if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
 413                         found = true;
 414                         flags = drv->ident_flags;
 415                         break;
 416                 }
 417         }
 418
 419         if (!found)
 420                 return (ENXIO);
 421
 422         if (outunit != NULL)
 423                 *outunit = unit;
 424         if (outflags != NULL)
 425                 *outflags = flags;
 426         return (0);
 427 }
 428
 429 /*
 430  * Get driver information from a set of flags specified.  Masks the identifying
 431  * part of the flags and compares it against all of the available
 432  * tuntap_drivers. Must be called with correct vnet context.
 433  */
 434 static struct tuntap_driver *
 435 tuntap_driver_from_flags(int tun_flags)
 436 {
 437         struct tuntap_driver *drv;
 438         struct tuntap_driver_cloner *drvc;
 439
 440         KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
 441             ("tuntap_driver_cloners failed to initialize"));
 442         SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
 443                 KASSERT(drvc->drv != NULL,
 444                     ("tuntap_driver_cloners entry not properly initialized"));
 445                 drv = drvc->drv;
 446                 if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->ident_flags)
 447                         return (drv);
 448         }
 449
 450         return (NULL);
 451 }
 452
 453
 454
 455 static int
 456 tun_clone_match(struct if_clone *ifc, const char *name)
 457 {
 458         int tunflags;
 459
 460         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 461                 if ((tunflags & TUN_L2) == 0)
 462                         return (1);
 463         }
 464
 465         return (0);
 466 }
 467
 468 static int
 469 tap_clone_match(struct if_clone *ifc, const char *name)
 470 {
 471         int tunflags;
 472
 473         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 474                 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
 475                         return (1);
 476         }
 477
 478         return (0);
 479 }
 480
 481 static int
 482 vmnet_clone_match(struct if_clone *ifc, const char *name)
 483 {
 484         int tunflags;
 485
 486         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 487                 if ((tunflags & TUN_VMNET) != 0)
 488                         return (1);
 489         }
 490
 491         return (0);
 492 }
 493
 494 static int
 495 tun_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
 496 {
 497         struct tuntap_driver *drv;
 498         struct cdev *dev;
 499         int err, i, tunflags, unit;
 500
 501         tunflags = 0;
 502         /* The name here tells us exactly what we're creating */
 503         err = tuntap_name2info(name, &unit, &tunflags);
 504         if (err != 0)
 505                 return (err);
 506
 507         drv = tuntap_driver_from_flags(tunflags);
 508         if (drv == NULL)
 509                 return (ENXIO);
 510
 511         if (unit != -1) {
 512                 /* If this unit number is still available that's okay. */
 513                 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
 514                         return (EEXIST);
 515         } else {
 516                 unit = alloc_unr(drv->unrhdr);
 517         }
 518
 519         snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
 520
 521         /* find any existing device, or allocate new unit number */
 522         dev = NULL;
 523         i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
 524         /* No preexisting struct cdev *, create one */
 525         if (i != 0)
 526                 i = tun_create_device(drv, unit, NULL, &dev, name);
 527         if (i == 0)
 528                 tuncreate(dev);
 529
 530         return (i);
 531 }
 532
 533 static void
 534 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
 535     struct cdev **dev)
 536 {
 537         char devname[SPECNAMELEN + 1];
 538         struct tuntap_driver *drv;
 539         int append_unit, i, u, tunflags;
 540         bool mayclone;
 541
 542         if (*dev != NULL)
 543                 return;
 544
 545         tunflags = 0;
 546         CURVNET_SET(CRED_TO_VNET(cred));
 547         if (tuntap_name2info(name, &u, &tunflags) != 0)
 548                 goto out;       /* Not recognized */
 549
 550         if (u != -1 && u > IF_MAXUNIT)
 551                 goto out;       /* Unit number too high */
 552
 553         mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE, 0) == 0;
 554         if ((tunflags & TUN_L2) != 0) {
 555                 /* tap/vmnet allow user open with a sysctl */
 556                 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
 557         } else {
 558                 mayclone = mayclone && tundclone;
 559         }
 560
 561         /*
 562          * If tun cloning is enabled, only the superuser can create an
 563          * interface.
 564          */
 565         if (!mayclone)
 566                 goto out;
 567
 568         if (u == -1)
 569                 append_unit = 1;
 570         else
 571                 append_unit = 0;
 572
 573         drv = tuntap_driver_from_flags(tunflags);
 574         if (drv == NULL)
 575                 goto out;
 576
 577         /* find any existing device, or allocate new unit number */
 578         i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
 579         if (i) {
 580                 if (append_unit) {
 581                         namelen = snprintf(devname, sizeof(devname), "%s%d",
 582                             name, u);
 583                         name = devname;
 584                 }
 585
 586                 i = tun_create_device(drv, u, cred, dev, name);
 587         }
 588         if (i == 0)
 589                 if_clone_create(name, namelen, NULL);
 590 out:
 591         CURVNET_RESTORE();
 592 }
 593
 594 static void
 595 tun_destroy(struct tuntap_softc *tp)
 596 {
 597
 598         TUN_LOCK(tp);
 599         tp->tun_flags |= TUN_DYING;
 600         if (tp->tun_busy != 0)
 601                 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
 602         else
 603                 TUN_UNLOCK(tp);
 604
 605         CURVNET_SET(TUN2IFP(tp)->if_vnet);
 606
 607         /* destroy_dev will take care of any alias. */
 608         destroy_dev(tp->tun_dev);
 609         seldrain(&tp->tun_rsel);
 610         knlist_clear(&tp->tun_rsel.si_note, 0);
 611         knlist_destroy(&tp->tun_rsel.si_note);
 612         if ((tp->tun_flags & TUN_L2) != 0) {
 613                 ether_ifdetach(TUN2IFP(tp));
 614         } else {
 615                 bpfdetach(TUN2IFP(tp));
 616                 if_detach(TUN2IFP(tp));
 617         }
 618         sx_xlock(&tun_ioctl_sx);
 619         TUN2IFP(tp)->if_softc = NULL;
 620         sx_xunlock(&tun_ioctl_sx);
 621         free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
 622         if_free(TUN2IFP(tp));
 623         mtx_destroy(&tp->tun_mtx);
 624         cv_destroy(&tp->tun_cv);
 625         free(tp, M_TUN);
 626         CURVNET_RESTORE();
 627 }
 628
 629 static int
 630 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp)
 631 {
 632         struct tuntap_softc *tp = ifp->if_softc;
 633
 634         mtx_lock(&tunmtx);
 635         TAILQ_REMOVE(&tunhead, tp, tun_list);
 636         mtx_unlock(&tunmtx);
 637         tun_destroy(tp);
 638
 639         return (0);
 640 }
 641
 642 static void
 643 vnet_tun_init(const void *unused __unused)
 644 {
 645         struct tuntap_driver *drv;
 646         struct tuntap_driver_cloner *drvc;
 647         int i;
 648
 649         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 650                 drv = &tuntap_drivers[i];
 651                 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
 652
 653                 drvc->drv = drv;
 654                 drvc->cloner = if_clone_advanced(drv->cdevsw.d_name, 0,
 655                     drv->clone_match_fn, drv->clone_create_fn,
 656                     drv->clone_destroy_fn);
 657                 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
 658         };
 659 }
 660 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 661                 vnet_tun_init, NULL);
 662
 663 static void
 664 vnet_tun_uninit(const void *unused __unused)
 665 {
 666         struct tuntap_driver_cloner *drvc;
 667
 668         while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
 669                 drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
 670                 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
 671
 672                 if_clone_detach(drvc->cloner);
 673                 free(drvc, M_TUN);
 674         }
 675 }
 676 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 677     vnet_tun_uninit, NULL);
 678
 679 static void
 680 tun_uninit(const void *unused __unused)
 681 {
 682         struct tuntap_driver *drv;
 683         struct tuntap_softc *tp;
 684         int i;
 685
 686         EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
 687         EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
 688         drain_dev_clone_events();
 689
 690         mtx_lock(&tunmtx);
 691         while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
 692                 TAILQ_REMOVE(&tunhead, tp, tun_list);
 693                 mtx_unlock(&tunmtx);
 694                 tun_destroy(tp);
 695                 mtx_lock(&tunmtx);
 696         }
 697         mtx_unlock(&tunmtx);
 698         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 699                 drv = &tuntap_drivers[i];
 700                 delete_unrhdr(drv->unrhdr);
 701                 clone_cleanup(&drv->clones);
 702         }
 703         mtx_destroy(&tunmtx);
 704 }
 705 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
 706
 707 static struct tuntap_driver *
 708 tuntap_driver_from_ifnet(const struct ifnet *ifp)
 709 {
 710         struct tuntap_driver *drv;
 711         int i;
 712
 713         if (ifp == NULL)
 714                 return (NULL);
 715
 716         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 717                 drv = &tuntap_drivers[i];
 718                 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
 719                         return (drv);
 720         }
 721
 722         return (NULL);
 723 }
 724
 725 static int
 726 tuntapmodevent(module_t mod, int type, void *data)
 727 {
 728         struct tuntap_driver *drv;
 729         int i;
 730
 731         switch (type) {
 732         case MOD_LOAD:
 733                 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
 734                 for (i = 0; i < nitems(tuntap_drivers); ++i) {
 735                         drv = &tuntap_drivers[i];
 736                         clone_setup(&drv->clones);
 737                         drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
 738                 }
 739                 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
 740                    tunrename, 0, 1000);
 741                 if (arrival_tag == NULL)
 742                         return (ENOMEM);
 743                 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
 744                 if (clone_tag == NULL)
 745                         return (ENOMEM);
 746                 break;
 747         case MOD_UNLOAD:
 748                 /* See tun_uninit, so it's done after the vnet_sysuninit() */
 749                 break;
 750         default:
 751                 return EOPNOTSUPP;
 752         }
 753         return 0;
 754 }
 755
 756 static moduledata_t tuntap_mod = {
 757         "if_tuntap",
 758         tuntapmodevent,
 759         0
 760 };
 761
 762 /* We'll only ever have these two, so no need for a macro. */
 763 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
 764 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
 765
 766 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 767 MODULE_VERSION(if_tuntap, 1);
 768 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 769 MODULE_VERSION(if_tun, 1);
 770 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 771 MODULE_VERSION(if_tap, 1);
 772
 773 static int
 774 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
 775     struct cdev **dev, const char *name)
 776 {
 777         struct make_dev_args args;
 778         struct tuntap_softc *tp;
 779         int error;
 780
 781         tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
 782         mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
 783         cv_init(&tp->tun_cv, "tun_condvar");
 784         tp->tun_flags = drv->ident_flags;
 785         tp->tun_drv = drv;
 786
 787         make_dev_args_init(&args);
 788         if (cr != NULL)
 789                 args.mda_flags = MAKEDEV_REF;
 790         args.mda_devsw = &drv->cdevsw;
 791         args.mda_cr = cr;
 792         args.mda_uid = UID_UUCP;
 793         args.mda_gid = GID_DIALER;
 794         args.mda_mode = 0600;
 795         args.mda_unit = unit;
 796         args.mda_si_drv1 = tp;
 797         error = make_dev_s(&args, dev, "%s", name);
 798         if (error != 0) {
 799                 free(tp, M_TUN);
 800                 return (error);
 801         }
 802
 803         KASSERT((*dev)->si_drv1 != NULL,
 804             ("Failed to set si_drv1 at %s creation", name));
 805         tp->tun_dev = *dev;
 806         knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
 807         mtx_lock(&tunmtx);
 808         TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
 809         mtx_unlock(&tunmtx);
 810         return (0);
 811 }
 812
 813 static void
 814 tunstart(struct ifnet *ifp)
 815 {
 816         struct tuntap_softc *tp = ifp->if_softc;
 817         struct mbuf *m;
 818
 819         TUNDEBUG(ifp, "starting\n");
 820         if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
 821                 IFQ_LOCK(&ifp->if_snd);
 822                 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
 823                 if (m == NULL) {
 824                         IFQ_UNLOCK(&ifp->if_snd);
 825                         return;
 826                 }
 827                 IFQ_UNLOCK(&ifp->if_snd);
 828         }
 829
 830         TUN_LOCK(tp);
 831         if (tp->tun_flags & TUN_RWAIT) {
 832                 tp->tun_flags &= ~TUN_RWAIT;
 833                 wakeup(tp);
 834         }
 835         selwakeuppri(&tp->tun_rsel, PZERO + 1);
 836         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 837         if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
 838                 TUN_UNLOCK(tp);
 839                 pgsigio(&tp->tun_sigio, SIGIO, 0);
 840         } else
 841                 TUN_UNLOCK(tp);
 842 }
 843
 844 /*
 845  * tunstart_l2
 846  *
 847  * queue packets from higher level ready to put out
 848  */
 849 static void
 850 tunstart_l2(struct ifnet *ifp)
 851 {
 852         struct tuntap_softc     *tp = ifp->if_softc;
 853
 854         TUNDEBUG(ifp, "starting\n");
 855
 856         /*
 857          * do not junk pending output if we are in VMnet mode.
 858          * XXX: can this do any harm because of queue overflow?
 859          */
 860
 861         TUN_LOCK(tp);
 862         if (((tp->tun_flags & TUN_VMNET) == 0) &&
 863             ((tp->tun_flags & TUN_READY) != TUN_READY)) {
 864                 struct mbuf *m;
 865
 866                 /* Unlocked read. */
 867                 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
 868
 869                 for (;;) {
 870                         IF_DEQUEUE(&ifp->if_snd, m);
 871                         if (m != NULL) {
 872                                 m_freem(m);
 873                                 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 874                         } else
 875                                 break;
 876                 }
 877                 TUN_UNLOCK(tp);
 878
 879                 return;
 880         }
 881
 882         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
 883
 884         if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
 885                 if (tp->tun_flags & TUN_RWAIT) {
 886                         tp->tun_flags &= ~TUN_RWAIT;
 887                         wakeup(tp);
 888                 }
 889
 890                 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
 891                         TUN_UNLOCK(tp);
 892                         pgsigio(&tp->tun_sigio, SIGIO, 0);
 893                         TUN_LOCK(tp);
 894                 }
 895
 896                 selwakeuppri(&tp->tun_rsel, PZERO+1);
 897                 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 898                 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
 899         }
 900
 901         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
 902         TUN_UNLOCK(tp);
 903 } /* tunstart_l2 */
 904
 905 /* XXX: should return an error code so it can fail. */
 906 static void
 907 tuncreate(struct cdev *dev)
 908 {
 909         struct tuntap_driver *drv;
 910         struct tuntap_softc *tp;
 911         struct ifnet *ifp;
 912         struct ether_addr eaddr;
 913         int iflags;
 914         u_char type;
 915
 916         tp = dev->si_drv1;
 917         KASSERT(tp != NULL,
 918             ("si_drv1 should have been initialized at creation"));
 919
 920         drv = tp->tun_drv;
 921         iflags = IFF_MULTICAST;
 922         if ((tp->tun_flags & TUN_L2) != 0) {
 923                 type = IFT_ETHER;
 924                 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
 925         } else {
 926                 type = IFT_PPP;
 927                 iflags |= IFF_POINTOPOINT;
 928         }
 929         ifp = tp->tun_ifp = if_alloc(type);
 930         if (ifp == NULL)
 931                 panic("%s%d: failed to if_alloc() interface.\n",
 932                     drv->cdevsw.d_name, dev2unit(dev));
 933         ifp->if_softc = tp;
 934         if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
 935         ifp->if_ioctl = tunifioctl;
 936         ifp->if_flags = iflags;
 937         IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
 938         ifp->if_capabilities |= IFCAP_LINKSTATE;
 939         ifp->if_capenable |= IFCAP_LINKSTATE;
 940
 941         if ((tp->tun_flags & TUN_L2) != 0) {
 942                 ifp->if_init = tunifinit;
 943                 ifp->if_start = tunstart_l2;
 944
 945                 ether_gen_addr(ifp, &eaddr);
 946                 ether_ifattach(ifp, eaddr.octet);
 947         } else {
 948                 ifp->if_mtu = TUNMTU;
 949                 ifp->if_start = tunstart;
 950                 ifp->if_output = tunoutput;
 951
 952                 ifp->if_snd.ifq_drv_maxlen = 0;
 953                 IFQ_SET_READY(&ifp->if_snd);
 954
 955                 if_attach(ifp);
 956                 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
 957         }
 958
 959         TUN_LOCK(tp);
 960         tp->tun_flags |= TUN_INITED;
 961         TUN_UNLOCK(tp);
 962
 963         TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
 964             ifp->if_xname, dev2unit(dev));
 965 }
 966
 967 static void
 968 tunrename(void *arg __unused, struct ifnet *ifp)
 969 {
 970         struct tuntap_softc *tp;
 971         int error;
 972
 973         if ((ifp->if_flags & IFF_RENAMING) == 0)
 974                 return;
 975
 976         if (tuntap_driver_from_ifnet(ifp) == NULL)
 977                 return;
 978
 979         /*
 980          * We need to grab the ioctl sx long enough to make sure the softc is
 981          * still there.  If it is, we can safely try to busy the tun device.
 982          * The busy may fail if the device is currently dying, in which case
 983          * we do nothing.  If it doesn't fail, the busy count stops the device
 984          * from dying until we've created the alias (that will then be
 985          * subsequently destroyed).
 986          */
 987         sx_xlock(&tun_ioctl_sx);
 988         tp = ifp->if_softc;
 989         if (tp == NULL) {
 990                 sx_xunlock(&tun_ioctl_sx);
 991                 return;
 992         }
 993         error = tun_busy(tp);
 994         sx_xunlock(&tun_ioctl_sx);
 995         if (error != 0)
 996                 return;
 997         if (tp->tun_alias != NULL) {
 998                 destroy_dev(tp->tun_alias);
 999                 tp->tun_alias = NULL;
1000         }
1001
1002         if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1003                 goto out;
1004
1005         /*
1006          * Failure's ok, aliases are created on a best effort basis.  If a
1007          * tun user/consumer decides to rename the interface to conflict with
1008          * another device (non-ifnet) on the system, we will assume they know
1009          * what they are doing.  make_dev_alias_p won't touch tun_alias on
1010          * failure, so we use it but ignore the return value.
1011          */
1012         make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1013             ifp->if_xname);
1014 out:
1015         tun_unbusy(tp);
1016 }
1017
1018 static int
1019 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1020 {
1021         struct ifnet    *ifp;
1022         struct tuntap_softc *tp;
1023         int error, tunflags;
1024
1025         tunflags = 0;
1026         CURVNET_SET(TD_TO_VNET(td));
1027         error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1028         if (error != 0) {
1029                 CURVNET_RESTORE();
1030                 return (error); /* Shouldn't happen */
1031         }
1032
1033         if ((tunflags & TUN_L2) != 0) {
1034                 /* Restrict? */
1035                 if (tap_allow_uopen == 0) {
1036                         error = priv_check(td, PRIV_NET_TAP);
1037                         if (error != 0) {
1038                                 CURVNET_RESTORE();
1039                                 return (error);
1040                         }
1041                 }
1042         }
1043
1044         tp = dev->si_drv1;
1045         KASSERT(tp != NULL,
1046             ("si_drv1 should have been initialized at creation"));
1047
1048         TUN_LOCK(tp);
1049         if ((tp->tun_flags & TUN_INITED) == 0) {
1050                 TUN_UNLOCK(tp);
1051                 CURVNET_RESTORE();
1052                 return (ENXIO);
1053         }
1054         if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1055                 TUN_UNLOCK(tp);
1056                 CURVNET_RESTORE();
1057                 return (EBUSY);
1058         }
1059
1060         error = tun_busy_locked(tp);
1061         KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1062         ifp = TUN2IFP(tp);
1063
1064         if ((tp->tun_flags & TUN_L2) != 0) {
1065                 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1066                     sizeof(tp->tun_ether.octet));
1067
1068                 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1069                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1070
1071                 if (tapuponopen)
1072                         ifp->if_flags |= IFF_UP;
1073         }
1074
1075         tp->tun_pid = td->td_proc->p_pid;
1076         tp->tun_flags |= TUN_OPEN;
1077
1078         if_link_state_change(ifp, LINK_STATE_UP);
1079         TUNDEBUG(ifp, "open\n");
1080         TUN_UNLOCK(tp);
1081
1082         /*
1083          * This can fail with either ENOENT or EBUSY.  This is in the middle of
1084          * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1085          * the only cdevpriv dtor being set will be tundtor and the softc being
1086          * passed is constant for a given cdev.  We ignore the possible error
1087          * because of this as either "unlikely" or "not actually a problem."
1088          */
1089         (void)devfs_set_cdevpriv(tp, tundtor);
1090         CURVNET_RESTORE();
1091         return (0);
1092 }
1093
1094 /*
1095  * tundtor - tear down the device - mark i/f down & delete
1096  * routing info
1097  */
1098 static void
1099 tundtor(void *data)
1100 {
1101         struct proc *p;
1102         struct tuntap_softc *tp;
1103         struct ifnet *ifp;
1104         bool l2tun;
1105
1106         tp = data;
1107         p = curproc;
1108         ifp = TUN2IFP(tp);
1109
1110         TUN_LOCK(tp);
1111
1112         /*
1113          * Realistically, we can't be obstinate here.  This only means that the
1114          * tuntap device was closed out of order, and the last closer wasn't the
1115          * controller.  These are still good to know about, though, as software
1116          * should avoid multiple processes with a tuntap device open and
1117          * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1118          * parent).
1119          */
1120         if (p->p_pid != tp->tun_pid) {
1121                 log(LOG_INFO,
1122                     "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1123                     p->p_pid, p->p_comm, tp->tun_dev->si_name);
1124         }
1125
1126         /*
1127          * junk all pending output
1128          */
1129         CURVNET_SET(ifp->if_vnet);
1130
1131         l2tun = false;
1132         if ((tp->tun_flags & TUN_L2) != 0) {
1133                 l2tun = true;
1134                 IF_DRAIN(&ifp->if_snd);
1135         } else {
1136                 IFQ_PURGE(&ifp->if_snd);
1137         }
1138
1139         /* For vmnet, we won't do most of the address/route bits */
1140         if ((tp->tun_flags & TUN_VMNET) != 0 ||
1141             (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1142                 goto out;
1143
1144         if (ifp->if_flags & IFF_UP) {
1145                 TUN_UNLOCK(tp);
1146                 if_down(ifp);
1147                 TUN_LOCK(tp);
1148         }
1149
1150         /* Delete all addresses and routes which reference this interface. */
1151         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1152                 struct ifaddr *ifa;
1153
1154                 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1155                 TUN_UNLOCK(tp);
1156                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1157                         /* deal w/IPv4 PtP destination; unlocked read */
1158                         if (!l2tun && ifa->ifa_addr->sa_family == AF_INET) {
1159                                 rtinit(ifa, (int)RTM_DELETE,
1160                                     tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
1161                         } else {
1162                                 rtinit(ifa, (int)RTM_DELETE, 0);
1163                         }
1164                 }
1165                 if_purgeaddrs(ifp);
1166                 TUN_LOCK(tp);
1167         }
1168
1169 out:
1170         if_link_state_change(ifp, LINK_STATE_DOWN);
1171         CURVNET_RESTORE();
1172
1173         funsetown(&tp->tun_sigio);
1174         selwakeuppri(&tp->tun_rsel, PZERO + 1);
1175         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1176         TUNDEBUG (ifp, "closed\n");
1177         tp->tun_flags &= ~TUN_OPEN;
1178         tp->tun_pid = 0;
1179
1180         tun_unbusy_locked(tp);
1181         TUN_UNLOCK(tp);
1182 }
1183
1184 static void
1185 tuninit(struct ifnet *ifp)
1186 {
1187         struct tuntap_softc *tp = ifp->if_softc;
1188 #ifdef INET
1189         struct ifaddr *ifa;
1190 #endif
1191
1192         TUNDEBUG(ifp, "tuninit\n");
1193
1194         TUN_LOCK(tp);
1195         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1196         if ((tp->tun_flags & TUN_L2) == 0) {
1197                 ifp->if_flags |= IFF_UP;
1198                 getmicrotime(&ifp->if_lastchange);
1199 #ifdef INET
1200                 if_addr_rlock(ifp);
1201                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1202                         if (ifa->ifa_addr->sa_family == AF_INET) {
1203                                 struct sockaddr_in *si;
1204
1205                                 si = (struct sockaddr_in *)ifa->ifa_dstaddr;
1206                                 if (si && si->sin_addr.s_addr) {
1207                                         tp->tun_flags |= TUN_DSTADDR;
1208                                         break;
1209                                 }
1210                         }
1211                 }
1212                 if_addr_runlock(ifp);
1213 #endif
1214                 TUN_UNLOCK(tp);
1215         } else {
1216                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1217                 TUN_UNLOCK(tp);
1218                 /* attempt to start output */
1219                 tunstart_l2(ifp);
1220         }
1221
1222 }
1223
1224 /*
1225  * Used only for l2 tunnel.
1226  */
1227 static void
1228 tunifinit(void *xtp)
1229 {
1230         struct tuntap_softc *tp;
1231
1232         tp = (struct tuntap_softc *)xtp;
1233         tuninit(tp->tun_ifp);
1234 }
1235
1236 /*
1237  * Process an ioctl request.
1238  */
1239 static int
1240 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1241 {
1242         struct ifreq *ifr = (struct ifreq *)data;
1243         struct tuntap_softc *tp;
1244         struct ifstat *ifs;
1245         struct ifmediareq       *ifmr;
1246         int             dummy, error = 0;
1247         bool            l2tun;
1248
1249         ifmr = NULL;
1250         sx_xlock(&tun_ioctl_sx);
1251         tp = ifp->if_softc;
1252         if (tp == NULL) {
1253                 error = ENXIO;
1254                 goto bad;
1255         }
1256         l2tun = (tp->tun_flags & TUN_L2) != 0;
1257         switch(cmd) {
1258         case SIOCGIFSTATUS:
1259                 ifs = (struct ifstat *)data;
1260                 TUN_LOCK(tp);
1261                 if (tp->tun_pid)
1262                         snprintf(ifs->ascii, sizeof(ifs->ascii),
1263                             "\tOpened by PID %d\n", tp->tun_pid);
1264                 else
1265                         ifs->ascii[0] = '\0';
1266                 TUN_UNLOCK(tp);
1267                 break;
1268         case SIOCSIFADDR:
1269                 if (l2tun)
1270                         error = ether_ioctl(ifp, cmd, data);
1271                 else
1272                         tuninit(ifp);
1273                 if (error == 0)
1274                     TUNDEBUG(ifp, "address set\n");
1275                 break;
1276         case SIOCSIFMTU:
1277                 ifp->if_mtu = ifr->ifr_mtu;
1278                 TUNDEBUG(ifp, "mtu set\n");
1279                 break;
1280         case SIOCSIFFLAGS:
1281         case SIOCADDMULTI:
1282         case SIOCDELMULTI:
1283                 break;
1284         case SIOCGIFMEDIA:
1285                 if (!l2tun) {
1286                         error = EINVAL;
1287                         break;
1288                 }
1289
1290                 ifmr = (struct ifmediareq *)data;
1291                 dummy = ifmr->ifm_count;
1292                 ifmr->ifm_count = 1;
1293                 ifmr->ifm_status = IFM_AVALID;
1294                 ifmr->ifm_active = IFM_ETHER;
1295                 if (tp->tun_flags & TUN_OPEN)
1296                         ifmr->ifm_status |= IFM_ACTIVE;
1297                 ifmr->ifm_current = ifmr->ifm_active;
1298                 if (dummy >= 1) {
1299                         int media = IFM_ETHER;
1300                         error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1301                 }
1302                 break;
1303         default:
1304                 if (l2tun) {
1305                         error = ether_ioctl(ifp, cmd, data);
1306                 } else {
1307                         error = EINVAL;
1308                 }
1309         }
1310 bad:
1311         sx_xunlock(&tun_ioctl_sx);
1312         return (error);
1313 }
1314
1315 /*
1316  * tunoutput - queue packets from higher level ready to put out.
1317  */
1318 static int
1319 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1320     struct route *ro)
1321 {
1322         struct tuntap_softc *tp = ifp->if_softc;
1323         u_short cached_tun_flags;
1324         int error;
1325         u_int32_t af;
1326
1327         TUNDEBUG (ifp, "tunoutput\n");
1328
1329 #ifdef MAC
1330         error = mac_ifnet_check_transmit(ifp, m0);
1331         if (error) {
1332                 m_freem(m0);
1333                 return (error);
1334         }
1335 #endif
1336
1337         /* Could be unlocked read? */
1338         TUN_LOCK(tp);
1339         cached_tun_flags = tp->tun_flags;
1340         TUN_UNLOCK(tp);
1341         if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1342                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1343                 m_freem (m0);
1344                 return (EHOSTDOWN);
1345         }
1346
1347         if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1348                 m_freem (m0);
1349                 return (EHOSTDOWN);
1350         }
1351
1352         /* BPF writes need to be handled specially. */
1353         if (dst->sa_family == AF_UNSPEC)
1354                 bcopy(dst->sa_data, &af, sizeof(af));
1355         else
1356                 af = dst->sa_family;
1357
1358         if (bpf_peers_present(ifp->if_bpf))
1359                 bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1360
1361         /* prepend sockaddr? this may abort if the mbuf allocation fails */
1362         if (cached_tun_flags & TUN_LMODE) {
1363                 /* allocate space for sockaddr */
1364                 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1365
1366                 /* if allocation failed drop packet */
1367                 if (m0 == NULL) {
1368                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1369                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1370                         return (ENOBUFS);
1371                 } else {
1372                         bcopy(dst, m0->m_data, dst->sa_len);
1373                 }
1374         }
1375
1376         if (cached_tun_flags & TUN_IFHEAD) {
1377                 /* Prepend the address family */
1378                 M_PREPEND(m0, 4, M_NOWAIT);
1379
1380                 /* if allocation failed drop packet */
1381                 if (m0 == NULL) {
1382                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1383                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1384                         return (ENOBUFS);
1385                 } else
1386                         *(u_int32_t *)m0->m_data = htonl(af);
1387         } else {
1388 #ifdef INET
1389                 if (af != AF_INET)
1390 #endif
1391                 {
1392                         m_freem(m0);
1393                         return (EAFNOSUPPORT);
1394                 }
1395         }
1396
1397         error = (ifp->if_transmit)(ifp, m0);
1398         if (error)
1399                 return (ENOBUFS);
1400         if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1401         return (0);
1402 }
1403
1404 /*
1405  * the cdevsw interface is now pretty minimal.
1406  */
1407 static  int
1408 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1409     struct thread *td)
1410 {
1411         struct ifreq ifr, *ifrp;
1412         struct tuntap_softc *tp = dev->si_drv1;
1413         struct tuninfo *tunp;
1414         int error, iflags;
1415 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1416     defined(COMPAT_FREEBSD4)
1417         int     ival;
1418 #endif
1419         bool    l2tun;
1420
1421         l2tun = (tp->tun_flags & TUN_L2) != 0;
1422         if (l2tun) {
1423                 /* tap specific ioctls */
1424                 switch(cmd) {
1425                 /* VMware/VMnet port ioctl's */
1426 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1427     defined(COMPAT_FREEBSD4)
1428                 case _IO('V', 0):
1429                         ival = IOCPARM_IVAL(data);
1430                         data = (caddr_t)&ival;
1431                         /* FALLTHROUGH */
1432 #endif
1433                 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1434                         iflags = *(int *)data;
1435                         iflags &= TUN_VMIO_FLAG_MASK;
1436                         iflags &= ~IFF_CANTCHANGE;
1437                         iflags |= IFF_UP;
1438
1439                         TUN_LOCK(tp);
1440                         TUN2IFP(tp)->if_flags = iflags |
1441                             (TUN2IFP(tp)->if_flags & IFF_CANTCHANGE);
1442                         TUN_UNLOCK(tp);
1443
1444                         return (0);
1445                 case SIOCGIFADDR:       /* get MAC address of the remote side */
1446                         TUN_LOCK(tp);
1447                         bcopy(&tp->tun_ether.octet, data,
1448                             sizeof(tp->tun_ether.octet));
1449                         TUN_UNLOCK(tp);
1450
1451                         return (0);
1452                 case SIOCSIFADDR:       /* set MAC address of the remote side */
1453                         TUN_LOCK(tp);
1454                         bcopy(data, &tp->tun_ether.octet,
1455                             sizeof(tp->tun_ether.octet));
1456                         TUN_UNLOCK(tp);
1457
1458                         return (0);
1459                 }
1460
1461                 /* Fall through to the common ioctls if unhandled */
1462         } else {
1463                 switch (cmd) {
1464                 case TUNSLMODE:
1465                         TUN_LOCK(tp);
1466                         if (*(int *)data) {
1467                                 tp->tun_flags |= TUN_LMODE;
1468                                 tp->tun_flags &= ~TUN_IFHEAD;
1469                         } else
1470                                 tp->tun_flags &= ~TUN_LMODE;
1471                         TUN_UNLOCK(tp);
1472
1473                         return (0);
1474                 case TUNSIFHEAD:
1475                         TUN_LOCK(tp);
1476                         if (*(int *)data) {
1477                                 tp->tun_flags |= TUN_IFHEAD;
1478                                 tp->tun_flags &= ~TUN_LMODE;
1479                         } else
1480                                 tp->tun_flags &= ~TUN_IFHEAD;
1481                         TUN_UNLOCK(tp);
1482
1483                         return (0);
1484                 case TUNGIFHEAD:
1485                         TUN_LOCK(tp);
1486                         *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1487                         TUN_UNLOCK(tp);
1488
1489                         return (0);
1490                 case TUNSIFMODE:
1491                         /* deny this if UP */
1492                         if (TUN2IFP(tp)->if_flags & IFF_UP)
1493                                 return (EBUSY);
1494
1495                         switch (*(int *)data & ~IFF_MULTICAST) {
1496                         case IFF_POINTOPOINT:
1497                         case IFF_BROADCAST:
1498                                 TUN_LOCK(tp);
1499                                 TUN2IFP(tp)->if_flags &=
1500                                     ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1501                                 TUN2IFP(tp)->if_flags |= *(int *)data;
1502                                 TUN_UNLOCK(tp);
1503
1504                                 break;
1505                         default:
1506                                 return (EINVAL);
1507                         }
1508
1509                         return (0);
1510                 case TUNSIFPID:
1511                         TUN_LOCK(tp);
1512                         tp->tun_pid = curthread->td_proc->p_pid;
1513                         TUN_UNLOCK(tp);
1514
1515                         return (0);
1516                 }
1517                 /* Fall through to the common ioctls if unhandled */
1518         }
1519
1520         switch (cmd) {
1521         case TUNGIFNAME:
1522                 ifrp = (struct ifreq *)data;
1523                 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1524
1525                 return (0);
1526         case TUNSIFINFO:
1527                 tunp = (struct tuninfo *)data;
1528                 if (TUN2IFP(tp)->if_type != tunp->type)
1529                         return (EPROTOTYPE);
1530                 TUN_LOCK(tp);
1531                 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1532                         strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1533                         ifr.ifr_mtu = tunp->mtu;
1534                         CURVNET_SET(TUN2IFP(tp)->if_vnet);
1535                         error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1536                             (caddr_t)&ifr, td);
1537                         CURVNET_RESTORE();
1538                         if (error) {
1539                                 TUN_UNLOCK(tp);
1540                                 return (error);
1541                         }
1542                 }
1543                 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1544                 TUN_UNLOCK(tp);
1545                 break;
1546         case TUNGIFINFO:
1547                 tunp = (struct tuninfo *)data;
1548                 TUN_LOCK(tp);
1549                 tunp->mtu = TUN2IFP(tp)->if_mtu;
1550                 tunp->type = TUN2IFP(tp)->if_type;
1551                 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1552                 TUN_UNLOCK(tp);
1553                 break;
1554         case TUNSDEBUG:
1555                 tundebug = *(int *)data;
1556                 break;
1557         case TUNGDEBUG:
1558                 *(int *)data = tundebug;
1559                 break;
1560         case FIONBIO:
1561                 break;
1562         case FIOASYNC:
1563                 TUN_LOCK(tp);
1564                 if (*(int *)data)
1565                         tp->tun_flags |= TUN_ASYNC;
1566                 else
1567                         tp->tun_flags &= ~TUN_ASYNC;
1568                 TUN_UNLOCK(tp);
1569                 break;
1570         case FIONREAD:
1571                 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1572                         struct mbuf *mb;
1573                         IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1574                         IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1575                         for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1576                                 *(int *)data += mb->m_len;
1577                         IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1578                 } else
1579                         *(int *)data = 0;
1580                 break;
1581         case FIOSETOWN:
1582                 return (fsetown(*(int *)data, &tp->tun_sigio));
1583
1584         case FIOGETOWN:
1585                 *(int *)data = fgetown(&tp->tun_sigio);
1586                 return (0);
1587
1588         /* This is deprecated, FIOSETOWN should be used instead. */
1589         case TIOCSPGRP:
1590                 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1591
1592         /* This is deprecated, FIOGETOWN should be used instead. */
1593         case TIOCGPGRP:
1594                 *(int *)data = -fgetown(&tp->tun_sigio);
1595                 return (0);
1596
1597         default:
1598                 return (ENOTTY);
1599         }
1600         return (0);
1601 }
1602
1603 /*
1604  * The cdevsw read interface - reads a packet at a time, or at
1605  * least as much of a packet as can be read.
1606  */
1607 static  int
1608 tunread(struct cdev *dev, struct uio *uio, int flag)
1609 {
1610         struct tuntap_softc *tp = dev->si_drv1;
1611         struct ifnet    *ifp = TUN2IFP(tp);
1612         struct mbuf     *m;
1613         int             error=0, len;
1614
1615         TUNDEBUG (ifp, "read\n");
1616         TUN_LOCK(tp);
1617         if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1618                 TUN_UNLOCK(tp);
1619                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1620                 return (EHOSTDOWN);
1621         }
1622
1623         tp->tun_flags &= ~TUN_RWAIT;
1624
1625         for (;;) {
1626                 IFQ_DEQUEUE(&ifp->if_snd, m);
1627                 if (m != NULL)
1628                         break;
1629                 if (flag & O_NONBLOCK) {
1630                         TUN_UNLOCK(tp);
1631                         return (EWOULDBLOCK);
1632                 }
1633                 tp->tun_flags |= TUN_RWAIT;
1634                 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1635                     "tunread", 0);
1636                 if (error != 0) {
1637                         TUN_UNLOCK(tp);
1638                         return (error);
1639                 }
1640         }
1641         TUN_UNLOCK(tp);
1642
1643         if ((tp->tun_flags & TUN_L2) != 0)
1644                 BPF_MTAP(ifp, m);
1645
1646         while (m && uio->uio_resid > 0 && error == 0) {
1647                 len = min(uio->uio_resid, m->m_len);
1648                 if (len != 0)
1649                         error = uiomove(mtod(m, void *), len, uio);
1650                 m = m_free(m);
1651         }
1652
1653         if (m) {
1654                 TUNDEBUG(ifp, "Dropping mbuf\n");
1655                 m_freem(m);
1656         }
1657         return (error);
1658 }
1659
1660 static int
1661 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m)
1662 {
1663         struct ether_header *eh;
1664         struct ifnet *ifp;
1665
1666         ifp = TUN2IFP(tp);
1667
1668         /*
1669          * Only pass a unicast frame to ether_input(), if it would
1670          * actually have been received by non-virtual hardware.
1671          */
1672         if (m->m_len < sizeof(struct ether_header)) {
1673                 m_freem(m);
1674                 return (0);
1675         }
1676
1677         eh = mtod(m, struct ether_header *);
1678
1679         if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1680             !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1681             bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1682                 m_freem(m);
1683                 return (0);
1684         }
1685
1686         /* Pass packet up to parent. */
1687         CURVNET_SET(ifp->if_vnet);
1688         (*ifp->if_input)(ifp, m);
1689         CURVNET_RESTORE();
1690         /* ibytes are counted in parent */
1691         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1692         return (0);
1693 }
1694
1695 static int
1696 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1697 {
1698         struct ifnet *ifp;
1699         int family, isr;
1700
1701         ifp = TUN2IFP(tp);
1702         /* Could be unlocked read? */
1703         TUN_LOCK(tp);
1704         if (tp->tun_flags & TUN_IFHEAD) {
1705                 TUN_UNLOCK(tp);
1706                 if (m->m_len < sizeof(family) &&
1707                 (m = m_pullup(m, sizeof(family))) == NULL)
1708                         return (ENOBUFS);
1709                 family = ntohl(*mtod(m, u_int32_t *));
1710                 m_adj(m, sizeof(family));
1711         } else {
1712                 TUN_UNLOCK(tp);
1713                 family = AF_INET;
1714         }
1715
1716         BPF_MTAP2(ifp, &family, sizeof(family), m);
1717
1718         switch (family) {
1719 #ifdef INET
1720         case AF_INET:
1721                 isr = NETISR_IP;
1722                 break;
1723 #endif
1724 #ifdef INET6
1725         case AF_INET6:
1726                 isr = NETISR_IPV6;
1727                 break;
1728 #endif
1729         default:
1730                 m_freem(m);
1731                 return (EAFNOSUPPORT);
1732         }
1733         random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1734         if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1735         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1736         CURVNET_SET(ifp->if_vnet);
1737         M_SETFIB(m, ifp->if_fib);
1738         netisr_dispatch(isr, m);
1739         CURVNET_RESTORE();
1740         return (0);
1741 }
1742
1743 /*
1744  * the cdevsw write interface - an atomic write is a packet - or else!
1745  */
1746 static  int
1747 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1748 {
1749         struct tuntap_softc *tp;
1750         struct ifnet    *ifp;
1751         struct mbuf     *m;
1752         uint32_t        mru;
1753         int             align;
1754         bool            l2tun;
1755
1756         tp = dev->si_drv1;
1757         ifp = TUN2IFP(tp);
1758         TUNDEBUG(ifp, "tunwrite\n");
1759         if ((ifp->if_flags & IFF_UP) != IFF_UP)
1760                 /* ignore silently */
1761                 return (0);
1762
1763         if (uio->uio_resid == 0)
1764                 return (0);
1765
1766         l2tun = (tp->tun_flags & TUN_L2) != 0;
1767         align = 0;
1768         mru = l2tun ? TAPMRU : TUNMRU;
1769         if (l2tun)
1770                 align = ETHER_ALIGN;
1771         else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1772                 mru += sizeof(uint32_t);        /* family */
1773         if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1774                 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1775                 return (EIO);
1776         }
1777
1778         if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1779                 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1780                 return (ENOBUFS);
1781         }
1782
1783         m->m_pkthdr.rcvif = ifp;
1784 #ifdef MAC
1785         mac_ifnet_create_mbuf(ifp, m);
1786 #endif
1787
1788         if (l2tun)
1789                 return (tunwrite_l2(tp, m));
1790
1791         return (tunwrite_l3(tp, m));
1792 }
1793
1794 /*
1795  * tunpoll - the poll interface, this is only useful on reads
1796  * really. The write detect always returns true, write never blocks
1797  * anyway, it either accepts the packet or drops it.
1798  */
1799 static  int
1800 tunpoll(struct cdev *dev, int events, struct thread *td)
1801 {
1802         struct tuntap_softc *tp = dev->si_drv1;
1803         struct ifnet    *ifp = TUN2IFP(tp);
1804         int             revents = 0;
1805
1806         TUNDEBUG(ifp, "tunpoll\n");
1807
1808         if (events & (POLLIN | POLLRDNORM)) {
1809                 IFQ_LOCK(&ifp->if_snd);
1810                 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1811                         TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1812                         revents |= events & (POLLIN | POLLRDNORM);
1813                 } else {
1814                         TUNDEBUG(ifp, "tunpoll waiting\n");
1815                         selrecord(td, &tp->tun_rsel);
1816                 }
1817                 IFQ_UNLOCK(&ifp->if_snd);
1818         }
1819         revents |= events & (POLLOUT | POLLWRNORM);
1820
1821         return (revents);
1822 }
1823
1824 /*
1825  * tunkqfilter - support for the kevent() system call.
1826  */
1827 static int
1828 tunkqfilter(struct cdev *dev, struct knote *kn)
1829 {
1830         struct tuntap_softc     *tp = dev->si_drv1;
1831         struct ifnet    *ifp = TUN2IFP(tp);
1832
1833         switch(kn->kn_filter) {
1834         case EVFILT_READ:
1835                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1836                     ifp->if_xname, dev2unit(dev));
1837                 kn->kn_fop = &tun_read_filterops;
1838                 break;
1839
1840         case EVFILT_WRITE:
1841                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1842                     ifp->if_xname, dev2unit(dev));
1843                 kn->kn_fop = &tun_write_filterops;
1844                 break;
1845
1846         default:
1847                 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1848                     ifp->if_xname, dev2unit(dev));
1849                 return(EINVAL);
1850         }
1851
1852         kn->kn_hook = tp;
1853         knlist_add(&tp->tun_rsel.si_note, kn, 0);
1854
1855         return (0);
1856 }
1857
1858 /*
1859  * Return true of there is data in the interface queue.
1860  */
1861 static int
1862 tunkqread(struct knote *kn, long hint)
1863 {
1864         int                     ret;
1865         struct tuntap_softc     *tp = kn->kn_hook;
1866         struct cdev             *dev = tp->tun_dev;
1867         struct ifnet    *ifp = TUN2IFP(tp);
1868
1869         if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
1870                 TUNDEBUG(ifp,
1871                     "%s have data in the queue.  Len = %d, minor = %#x\n",
1872                     ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
1873                 ret = 1;
1874         } else {
1875                 TUNDEBUG(ifp,
1876                     "%s waiting for data, minor = %#x\n", ifp->if_xname,
1877                     dev2unit(dev));
1878                 ret = 0;
1879         }
1880
1881         return (ret);
1882 }
1883
1884 /*
1885  * Always can write, always return MTU in kn->data.
1886  */
1887 static int
1888 tunkqwrite(struct knote *kn, long hint)
1889 {
1890         struct tuntap_softc     *tp = kn->kn_hook;
1891         struct ifnet    *ifp = TUN2IFP(tp);
1892
1893         kn->kn_data = ifp->if_mtu;
1894
1895         return (1);
1896 }
1897
1898 static void
1899 tunkqdetach(struct knote *kn)
1900 {
1901         struct tuntap_softc     *tp = kn->kn_hook;
1902
1903         knlist_remove(&tp->tun_rsel.si_note, kn, 0);
1904 }