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