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/syslog.h>
  67 #include <sys/ttycom.h>
  68 #include <sys/poll.h>
  69 #include <sys/selinfo.h>
  70 #include <sys/signalvar.h>
  71 #include <sys/filedesc.h>
  72 #include <sys/kernel.h>
  73 #include <sys/sysctl.h>
  74 #include <sys/conf.h>
  75 #include <sys/uio.h>
  76 #include <sys/malloc.h>
  77 #include <sys/random.h>
  78 #include <sys/ctype.h>
  79
  80 #include <net/ethernet.h>
  81 #include <net/if.h>
  82 #include <net/if_var.h>
  83 #include <net/if_clone.h>
  84 #include <net/if_dl.h>
  85 #include <net/if_media.h>
  86 #include <net/if_types.h>
  87 #include <net/if_vlan_var.h>
  88 #include <net/netisr.h>
  89 #include <net/route.h>
  90 #include <net/vnet.h>
  91 #ifdef INET
  92 #include <netinet/in.h>
  93 #include <netinet/ip.h>
  94 #include <netinet/ip6.h>
  95 #include <netinet6/ip6_var.h>
  96 #include <netinet/udp.h>
  97 #include <netinet/tcp.h>
  98 #endif
  99 #include <net/bpf.h>
 100 #include <net/if_tap.h>
 101 #include <net/if_tun.h>
 102
 103 #include <dev/virtio/network/virtio_net.h>
 104
 105 #include <sys/queue.h>
 106 #include <sys/condvar.h>
 107 #include <security/mac/mac_framework.h>
 108
 109 struct tuntap_driver;
 110
 111 /*
 112  * tun_list is protected by global tunmtx.  Other mutable fields are
 113  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
 114  * static for the duration of a tunnel interface.
 115  */
 116 struct tuntap_softc {
 117         TAILQ_ENTRY(tuntap_softc)        tun_list;
 118         struct cdev                     *tun_alias;
 119         struct cdev                     *tun_dev;
 120         u_short                          tun_flags;     /* misc flags */
 121 #define TUN_OPEN        0x0001
 122 #define TUN_INITED      0x0002
 123 #define TUN_UNUSED1     0x0008
 124 #define TUN_DSTADDR     0x0010
 125 #define TUN_LMODE       0x0020
 126 #define TUN_RWAIT       0x0040
 127 #define TUN_ASYNC       0x0080
 128 #define TUN_IFHEAD      0x0100
 129 #define TUN_DYING       0x0200
 130 #define TUN_L2          0x0400
 131 #define TUN_VMNET       0x0800
 132
 133 #define TUN_DRIVER_IDENT_MASK   (TUN_L2 | TUN_VMNET)
 134 #define TUN_READY               (TUN_OPEN | TUN_INITED)
 135
 136         pid_t                    tun_pid;       /* owning pid */
 137         struct ifnet            *tun_ifp;       /* the interface */
 138         struct sigio            *tun_sigio;     /* async I/O info */
 139         struct tuntap_driver    *tun_drv;       /* appropriate driver */
 140         struct selinfo           tun_rsel;      /* read select */
 141         struct mtx               tun_mtx;       /* softc field mutex */
 142         struct cv                tun_cv;        /* for ref'd dev destroy */
 143         struct ether_addr        tun_ether;     /* remote address */
 144         int                      tun_busy;      /* busy count */
 145         int                      tun_vhdrlen;   /* virtio-net header length */
 146 };
 147 #define TUN2IFP(sc)     ((sc)->tun_ifp)
 148
 149 #define TUNDEBUG        if (tundebug) if_printf
 150
 151 #define TUN_LOCK(tp)            mtx_lock(&(tp)->tun_mtx)
 152 #define TUN_UNLOCK(tp)          mtx_unlock(&(tp)->tun_mtx)
 153 #define TUN_LOCK_ASSERT(tp)     mtx_assert(&(tp)->tun_mtx, MA_OWNED);
 154
 155 #define TUN_VMIO_FLAG_MASK      0x0fff
 156
 157 /*
 158  * Interface capabilities of a tap device that supports the virtio-net
 159  * header.
 160  */
 161 #define TAP_VNET_HDR_CAPS       (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6       \
 162                                 | IFCAP_VLAN_HWCSUM                     \
 163                                 | IFCAP_TSO | IFCAP_LRO                 \
 164                                 | IFCAP_VLAN_HWTSO)
 165
 166 #define TAP_ALL_OFFLOAD         (CSUM_TSO | CSUM_TCP | CSUM_UDP |\
 167                                     CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
 168
 169
 170 /*
 171  * All mutable global variables in if_tun are locked using tunmtx, with
 172  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
 173  * which are static after setup.
 174  */
 175 static struct mtx tunmtx;
 176 static eventhandler_tag arrival_tag;
 177 static eventhandler_tag clone_tag;
 178 static const char tunname[] = "tun";
 179 static const char tapname[] = "tap";
 180 static const char vmnetname[] = "vmnet";
 181 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
 182 static int tundebug = 0;
 183 static int tundclone = 1;
 184 static int tap_allow_uopen = 0; /* allow user devfs cloning */
 185 static int tapuponopen = 0;     /* IFF_UP on open() */
 186 static int tapdclone = 1;       /* enable devfs cloning */
 187
 188 static TAILQ_HEAD(,tuntap_softc)        tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
 189 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
 190
 191 static struct sx tun_ioctl_sx;
 192 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
 193
 194 SYSCTL_DECL(_net_link);
 195 /* tun */
 196 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW, 0,
 197     "IP tunnel software network interface");
 198 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
 199     "Enable legacy devfs interface creation");
 200
 201 /* tap */
 202 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW, 0,
 203     "Ethernet tunnel software network interface");
 204 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
 205     "Enable legacy devfs interface creation for all users");
 206 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
 207     "Bring interface up when /dev/tap is opened");
 208 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
 209     "Enable legacy devfs interface creation");
 210 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
 211
 212 static int      tun_create_device(struct tuntap_driver *drv, int unit,
 213     struct ucred *cr, struct cdev **dev, const char *name);
 214 static int      tun_busy_locked(struct tuntap_softc *tp);
 215 static void     tun_unbusy_locked(struct tuntap_softc *tp);
 216 static int      tun_busy(struct tuntap_softc *tp);
 217 static void     tun_unbusy(struct tuntap_softc *tp);
 218
 219 static int      tuntap_name2info(const char *name, int *unit, int *flags);
 220 static void     tunclone(void *arg, struct ucred *cred, char *name,
 221                     int namelen, struct cdev **dev);
 222 static void     tuncreate(struct cdev *dev);
 223 static void     tundtor(void *data);
 224 static void     tunrename(void *arg, struct ifnet *ifp);
 225 static int      tunifioctl(struct ifnet *, u_long, caddr_t);
 226 static void     tuninit(struct ifnet *);
 227 static void     tunifinit(void *xtp);
 228 static int      tuntapmodevent(module_t, int, void *);
 229 static int      tunoutput(struct ifnet *, struct mbuf *,
 230                     const struct sockaddr *, struct route *ro);
 231 static void     tunstart(struct ifnet *);
 232 static void     tunstart_l2(struct ifnet *);
 233
 234 static int      tun_clone_match(struct if_clone *ifc, const char *name);
 235 static int      tap_clone_match(struct if_clone *ifc, const char *name);
 236 static int      vmnet_clone_match(struct if_clone *ifc, const char *name);
 237 static int      tun_clone_create(struct if_clone *, char *, size_t, caddr_t);
 238 static int      tun_clone_destroy(struct if_clone *, struct ifnet *);
 239 static void     tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
 240
 241 static d_open_t         tunopen;
 242 static d_read_t         tunread;
 243 static d_write_t        tunwrite;
 244 static d_ioctl_t        tunioctl;
 245 static d_poll_t         tunpoll;
 246 static d_kqfilter_t     tunkqfilter;
 247
 248 static int              tunkqread(struct knote *, long);
 249 static int              tunkqwrite(struct knote *, long);
 250 static void             tunkqdetach(struct knote *);
 251
 252 static struct filterops tun_read_filterops = {
 253         .f_isfd =       1,
 254         .f_attach =     NULL,
 255         .f_detach =     tunkqdetach,
 256         .f_event =      tunkqread,
 257 };
 258
 259 static struct filterops tun_write_filterops = {
 260         .f_isfd =       1,
 261         .f_attach =     NULL,
 262         .f_detach =     tunkqdetach,
 263         .f_event =      tunkqwrite,
 264 };
 265
 266 static struct tuntap_driver {
 267         struct cdevsw            cdevsw;
 268         int                      ident_flags;
 269         struct unrhdr           *unrhdr;
 270         struct clonedevs        *clones;
 271         ifc_match_t             *clone_match_fn;
 272         ifc_create_t            *clone_create_fn;
 273         ifc_destroy_t           *clone_destroy_fn;
 274 } tuntap_drivers[] = {
 275         {
 276                 .ident_flags =  0,
 277                 .cdevsw =       {
 278                     .d_version =        D_VERSION,
 279                     .d_flags =          D_NEEDMINOR,
 280                     .d_open =           tunopen,
 281                     .d_read =           tunread,
 282                     .d_write =          tunwrite,
 283                     .d_ioctl =          tunioctl,
 284                     .d_poll =           tunpoll,
 285                     .d_kqfilter =       tunkqfilter,
 286                     .d_name =           tunname,
 287                 },
 288                 .clone_match_fn =       tun_clone_match,
 289                 .clone_create_fn =      tun_clone_create,
 290                 .clone_destroy_fn =     tun_clone_destroy,
 291         },
 292         {
 293                 .ident_flags =  TUN_L2,
 294                 .cdevsw =       {
 295                     .d_version =        D_VERSION,
 296                     .d_flags =          D_NEEDMINOR,
 297                     .d_open =           tunopen,
 298                     .d_read =           tunread,
 299                     .d_write =          tunwrite,
 300                     .d_ioctl =          tunioctl,
 301                     .d_poll =           tunpoll,
 302                     .d_kqfilter =       tunkqfilter,
 303                     .d_name =           tapname,
 304                 },
 305                 .clone_match_fn =       tap_clone_match,
 306                 .clone_create_fn =      tun_clone_create,
 307                 .clone_destroy_fn =     tun_clone_destroy,
 308         },
 309         {
 310                 .ident_flags =  TUN_L2 | TUN_VMNET,
 311                 .cdevsw =       {
 312                     .d_version =        D_VERSION,
 313                     .d_flags =          D_NEEDMINOR,
 314                     .d_open =           tunopen,
 315                     .d_read =           tunread,
 316                     .d_write =          tunwrite,
 317                     .d_ioctl =          tunioctl,
 318                     .d_poll =           tunpoll,
 319                     .d_kqfilter =       tunkqfilter,
 320                     .d_name =           vmnetname,
 321                 },
 322                 .clone_match_fn =       vmnet_clone_match,
 323                 .clone_create_fn =      tun_clone_create,
 324                 .clone_destroy_fn =     tun_clone_destroy,
 325         },
 326 };
 327
 328 struct tuntap_driver_cloner {
 329         SLIST_ENTRY(tuntap_driver_cloner)        link;
 330         struct tuntap_driver                    *drv;
 331         struct if_clone                         *cloner;
 332 };
 333
 334 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
 335     SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
 336
 337 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
 338
 339 /*
 340  * Mechanism for marking a tunnel device as busy so that we can safely do some
 341  * orthogonal operations (such as operations on devices) without racing against
 342  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
 343  * open, to be woken up when the condition is alleviated.
 344  */
 345 static int
 346 tun_busy_locked(struct tuntap_softc *tp)
 347 {
 348
 349         TUN_LOCK_ASSERT(tp);
 350         if ((tp->tun_flags & TUN_DYING) != 0) {
 351                 /*
 352                  * Perhaps unintuitive, but the device is busy going away.
 353                  * Other interpretations of EBUSY from tun_busy make little
 354                  * sense, since making a busy device even more busy doesn't
 355                  * sound like a problem.
 356                  */
 357                 return (EBUSY);
 358         }
 359
 360         ++tp->tun_busy;
 361         return (0);
 362 }
 363
 364 static void
 365 tun_unbusy_locked(struct tuntap_softc *tp)
 366 {
 367
 368         TUN_LOCK_ASSERT(tp);
 369         KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
 370
 371         --tp->tun_busy;
 372         /* Wake up anything that may be waiting on our busy tunnel. */
 373         if (tp->tun_busy == 0)
 374                 cv_broadcast(&tp->tun_cv);
 375 }
 376
 377 static int
 378 tun_busy(struct tuntap_softc *tp)
 379 {
 380         int ret;
 381
 382         TUN_LOCK(tp);
 383         ret = tun_busy_locked(tp);
 384         TUN_UNLOCK(tp);
 385         return (ret);
 386 }
 387
 388
 389 static void
 390 tun_unbusy(struct tuntap_softc *tp)
 391 {
 392
 393         TUN_LOCK(tp);
 394         tun_unbusy_locked(tp);
 395         TUN_UNLOCK(tp);
 396 }
 397
 398 /*
 399  * Sets unit and/or flags given the device name.  Must be called with correct
 400  * vnet context.
 401  */
 402 static int
 403 tuntap_name2info(const char *name, int *outunit, int *outflags)
 404 {
 405         struct tuntap_driver *drv;
 406         struct tuntap_driver_cloner *drvc;
 407         char *dname;
 408         int flags, unit;
 409         bool found;
 410
 411         if (name == NULL)
 412                 return (EINVAL);
 413
 414         /*
 415          * Needed for dev_stdclone, but dev_stdclone will not modify, it just
 416          * wants to be able to pass back a char * through the second param. We
 417          * will always set that as NULL here, so we'll fake it.
 418          */
 419         dname = __DECONST(char *, name);
 420         found = false;
 421
 422         KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
 423             ("tuntap_driver_cloners failed to initialize"));
 424         SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
 425                 KASSERT(drvc->drv != NULL,
 426                     ("tuntap_driver_cloners entry not properly initialized"));
 427                 drv = drvc->drv;
 428
 429                 if (strcmp(name, drv->cdevsw.d_name) == 0) {
 430                         found = true;
 431                         unit = -1;
 432                         flags = drv->ident_flags;
 433                         break;
 434                 }
 435
 436                 if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
 437                         found = true;
 438                         flags = drv->ident_flags;
 439                         break;
 440                 }
 441         }
 442
 443         if (!found)
 444                 return (ENXIO);
 445
 446         if (outunit != NULL)
 447                 *outunit = unit;
 448         if (outflags != NULL)
 449                 *outflags = flags;
 450         return (0);
 451 }
 452
 453 /*
 454  * Get driver information from a set of flags specified.  Masks the identifying
 455  * part of the flags and compares it against all of the available
 456  * tuntap_drivers. Must be called with correct vnet context.
 457  */
 458 static struct tuntap_driver *
 459 tuntap_driver_from_flags(int tun_flags)
 460 {
 461         struct tuntap_driver *drv;
 462         struct tuntap_driver_cloner *drvc;
 463
 464         KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
 465             ("tuntap_driver_cloners failed to initialize"));
 466         SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
 467                 KASSERT(drvc->drv != NULL,
 468                     ("tuntap_driver_cloners entry not properly initialized"));
 469                 drv = drvc->drv;
 470                 if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->ident_flags)
 471                         return (drv);
 472         }
 473
 474         return (NULL);
 475 }
 476
 477
 478
 479 static int
 480 tun_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_L2) == 0)
 486                         return (1);
 487         }
 488
 489         return (0);
 490 }
 491
 492 static int
 493 tap_clone_match(struct if_clone *ifc, const char *name)
 494 {
 495         int tunflags;
 496
 497         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 498                 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
 499                         return (1);
 500         }
 501
 502         return (0);
 503 }
 504
 505 static int
 506 vmnet_clone_match(struct if_clone *ifc, const char *name)
 507 {
 508         int tunflags;
 509
 510         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 511                 if ((tunflags & TUN_VMNET) != 0)
 512                         return (1);
 513         }
 514
 515         return (0);
 516 }
 517
 518 static int
 519 tun_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
 520 {
 521         struct tuntap_driver *drv;
 522         struct cdev *dev;
 523         int err, i, tunflags, unit;
 524
 525         tunflags = 0;
 526         /* The name here tells us exactly what we're creating */
 527         err = tuntap_name2info(name, &unit, &tunflags);
 528         if (err != 0)
 529                 return (err);
 530
 531         drv = tuntap_driver_from_flags(tunflags);
 532         if (drv == NULL)
 533                 return (ENXIO);
 534
 535         if (unit != -1) {
 536                 /* If this unit number is still available that's okay. */
 537                 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
 538                         return (EEXIST);
 539         } else {
 540                 unit = alloc_unr(drv->unrhdr);
 541         }
 542
 543         snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
 544
 545         /* find any existing device, or allocate new unit number */
 546         dev = NULL;
 547         i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
 548         /* No preexisting struct cdev *, create one */
 549         if (i != 0)
 550                 i = tun_create_device(drv, unit, NULL, &dev, name);
 551         if (i == 0)
 552                 tuncreate(dev);
 553
 554         return (i);
 555 }
 556
 557 static void
 558 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
 559     struct cdev **dev)
 560 {
 561         char devname[SPECNAMELEN + 1];
 562         struct tuntap_driver *drv;
 563         int append_unit, i, u, tunflags;
 564         bool mayclone;
 565
 566         if (*dev != NULL)
 567                 return;
 568
 569         tunflags = 0;
 570         CURVNET_SET(CRED_TO_VNET(cred));
 571         if (tuntap_name2info(name, &u, &tunflags) != 0)
 572                 goto out;       /* Not recognized */
 573
 574         if (u != -1 && u > IF_MAXUNIT)
 575                 goto out;       /* Unit number too high */
 576
 577         mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
 578         if ((tunflags & TUN_L2) != 0) {
 579                 /* tap/vmnet allow user open with a sysctl */
 580                 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
 581         } else {
 582                 mayclone = mayclone && tundclone;
 583         }
 584
 585         /*
 586          * If tun cloning is enabled, only the superuser can create an
 587          * interface.
 588          */
 589         if (!mayclone)
 590                 goto out;
 591
 592         if (u == -1)
 593                 append_unit = 1;
 594         else
 595                 append_unit = 0;
 596
 597         drv = tuntap_driver_from_flags(tunflags);
 598         if (drv == NULL)
 599                 goto out;
 600
 601         /* find any existing device, or allocate new unit number */
 602         i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
 603         if (i) {
 604                 if (append_unit) {
 605                         namelen = snprintf(devname, sizeof(devname), "%s%d",
 606                             name, u);
 607                         name = devname;
 608                 }
 609
 610                 i = tun_create_device(drv, u, cred, dev, name);
 611         }
 612         if (i == 0)
 613                 if_clone_create(name, namelen, NULL);
 614 out:
 615         CURVNET_RESTORE();
 616 }
 617
 618 static void
 619 tun_destroy(struct tuntap_softc *tp)
 620 {
 621
 622         TUN_LOCK(tp);
 623         tp->tun_flags |= TUN_DYING;
 624         if (tp->tun_busy != 0)
 625                 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
 626         else
 627                 TUN_UNLOCK(tp);
 628
 629         CURVNET_SET(TUN2IFP(tp)->if_vnet);
 630
 631         /* destroy_dev will take care of any alias. */
 632         destroy_dev(tp->tun_dev);
 633         seldrain(&tp->tun_rsel);
 634         knlist_clear(&tp->tun_rsel.si_note, 0);
 635         knlist_destroy(&tp->tun_rsel.si_note);
 636         if ((tp->tun_flags & TUN_L2) != 0) {
 637                 ether_ifdetach(TUN2IFP(tp));
 638         } else {
 639                 bpfdetach(TUN2IFP(tp));
 640                 if_detach(TUN2IFP(tp));
 641         }
 642         sx_xlock(&tun_ioctl_sx);
 643         TUN2IFP(tp)->if_softc = NULL;
 644         sx_xunlock(&tun_ioctl_sx);
 645         free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
 646         if_free(TUN2IFP(tp));
 647         mtx_destroy(&tp->tun_mtx);
 648         cv_destroy(&tp->tun_cv);
 649         free(tp, M_TUN);
 650         CURVNET_RESTORE();
 651 }
 652
 653 static int
 654 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp)
 655 {
 656         struct tuntap_softc *tp = ifp->if_softc;
 657
 658         mtx_lock(&tunmtx);
 659         TAILQ_REMOVE(&tunhead, tp, tun_list);
 660         mtx_unlock(&tunmtx);
 661         tun_destroy(tp);
 662
 663         return (0);
 664 }
 665
 666 static void
 667 vnet_tun_init(const void *unused __unused)
 668 {
 669         struct tuntap_driver *drv;
 670         struct tuntap_driver_cloner *drvc;
 671         int i;
 672
 673         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 674                 drv = &tuntap_drivers[i];
 675                 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
 676
 677                 drvc->drv = drv;
 678                 drvc->cloner = if_clone_advanced(drv->cdevsw.d_name, 0,
 679                     drv->clone_match_fn, drv->clone_create_fn,
 680                     drv->clone_destroy_fn);
 681                 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
 682         };
 683 }
 684 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 685                 vnet_tun_init, NULL);
 686
 687 static void
 688 vnet_tun_uninit(const void *unused __unused)
 689 {
 690         struct tuntap_driver_cloner *drvc;
 691
 692         while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
 693                 drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
 694                 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
 695
 696                 if_clone_detach(drvc->cloner);
 697                 free(drvc, M_TUN);
 698         }
 699 }
 700 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 701     vnet_tun_uninit, NULL);
 702
 703 static void
 704 tun_uninit(const void *unused __unused)
 705 {
 706         struct tuntap_driver *drv;
 707         struct tuntap_softc *tp;
 708         int i;
 709
 710         EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
 711         EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
 712         drain_dev_clone_events();
 713
 714         mtx_lock(&tunmtx);
 715         while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
 716                 TAILQ_REMOVE(&tunhead, tp, tun_list);
 717                 mtx_unlock(&tunmtx);
 718                 tun_destroy(tp);
 719                 mtx_lock(&tunmtx);
 720         }
 721         mtx_unlock(&tunmtx);
 722         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 723                 drv = &tuntap_drivers[i];
 724                 delete_unrhdr(drv->unrhdr);
 725                 clone_cleanup(&drv->clones);
 726         }
 727         mtx_destroy(&tunmtx);
 728 }
 729 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
 730
 731 static struct tuntap_driver *
 732 tuntap_driver_from_ifnet(const struct ifnet *ifp)
 733 {
 734         struct tuntap_driver *drv;
 735         int i;
 736
 737         if (ifp == NULL)
 738                 return (NULL);
 739
 740         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 741                 drv = &tuntap_drivers[i];
 742                 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
 743                         return (drv);
 744         }
 745
 746         return (NULL);
 747 }
 748
 749 static int
 750 tuntapmodevent(module_t mod, int type, void *data)
 751 {
 752         struct tuntap_driver *drv;
 753         int i;
 754
 755         switch (type) {
 756         case MOD_LOAD:
 757                 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
 758                 for (i = 0; i < nitems(tuntap_drivers); ++i) {
 759                         drv = &tuntap_drivers[i];
 760                         clone_setup(&drv->clones);
 761                         drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
 762                 }
 763                 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
 764                    tunrename, 0, 1000);
 765                 if (arrival_tag == NULL)
 766                         return (ENOMEM);
 767                 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
 768                 if (clone_tag == NULL)
 769                         return (ENOMEM);
 770                 break;
 771         case MOD_UNLOAD:
 772                 /* See tun_uninit, so it's done after the vnet_sysuninit() */
 773                 break;
 774         default:
 775                 return EOPNOTSUPP;
 776         }
 777         return 0;
 778 }
 779
 780 static moduledata_t tuntap_mod = {
 781         "if_tuntap",
 782         tuntapmodevent,
 783         0
 784 };
 785
 786 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 787 MODULE_VERSION(if_tuntap, 1);
 788 MODULE_VERSION(if_tun, 1);
 789 MODULE_VERSION(if_tap, 1);
 790
 791 static int
 792 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
 793     struct cdev **dev, const char *name)
 794 {
 795         struct make_dev_args args;
 796         struct tuntap_softc *tp;
 797         int error;
 798
 799         tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
 800         mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
 801         cv_init(&tp->tun_cv, "tun_condvar");
 802         tp->tun_flags = drv->ident_flags;
 803         tp->tun_drv = drv;
 804
 805         make_dev_args_init(&args);
 806         if (cr != NULL)
 807                 args.mda_flags = MAKEDEV_REF;
 808         args.mda_devsw = &drv->cdevsw;
 809         args.mda_cr = cr;
 810         args.mda_uid = UID_UUCP;
 811         args.mda_gid = GID_DIALER;
 812         args.mda_mode = 0600;
 813         args.mda_unit = unit;
 814         args.mda_si_drv1 = tp;
 815         error = make_dev_s(&args, dev, "%s", name);
 816         if (error != 0) {
 817                 free(tp, M_TUN);
 818                 return (error);
 819         }
 820
 821         KASSERT((*dev)->si_drv1 != NULL,
 822             ("Failed to set si_drv1 at %s creation", name));
 823         tp->tun_dev = *dev;
 824         knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
 825         mtx_lock(&tunmtx);
 826         TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
 827         mtx_unlock(&tunmtx);
 828         return (0);
 829 }
 830
 831 static void
 832 tunstart(struct ifnet *ifp)
 833 {
 834         struct tuntap_softc *tp = ifp->if_softc;
 835         struct mbuf *m;
 836
 837         TUNDEBUG(ifp, "starting\n");
 838         if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
 839                 IFQ_LOCK(&ifp->if_snd);
 840                 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
 841                 if (m == NULL) {
 842                         IFQ_UNLOCK(&ifp->if_snd);
 843                         return;
 844                 }
 845                 IFQ_UNLOCK(&ifp->if_snd);
 846         }
 847
 848         TUN_LOCK(tp);
 849         if (tp->tun_flags & TUN_RWAIT) {
 850                 tp->tun_flags &= ~TUN_RWAIT;
 851                 wakeup(tp);
 852         }
 853         selwakeuppri(&tp->tun_rsel, PZERO + 1);
 854         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 855         if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
 856                 TUN_UNLOCK(tp);
 857                 pgsigio(&tp->tun_sigio, SIGIO, 0);
 858         } else
 859                 TUN_UNLOCK(tp);
 860 }
 861
 862 /*
 863  * tunstart_l2
 864  *
 865  * queue packets from higher level ready to put out
 866  */
 867 static void
 868 tunstart_l2(struct ifnet *ifp)
 869 {
 870         struct tuntap_softc     *tp = ifp->if_softc;
 871
 872         TUNDEBUG(ifp, "starting\n");
 873
 874         /*
 875          * do not junk pending output if we are in VMnet mode.
 876          * XXX: can this do any harm because of queue overflow?
 877          */
 878
 879         TUN_LOCK(tp);
 880         if (((tp->tun_flags & TUN_VMNET) == 0) &&
 881             ((tp->tun_flags & TUN_READY) != TUN_READY)) {
 882                 struct mbuf *m;
 883
 884                 /* Unlocked read. */
 885                 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
 886
 887                 for (;;) {
 888                         IF_DEQUEUE(&ifp->if_snd, m);
 889                         if (m != NULL) {
 890                                 m_freem(m);
 891                                 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 892                         } else
 893                                 break;
 894                 }
 895                 TUN_UNLOCK(tp);
 896
 897                 return;
 898         }
 899
 900         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
 901
 902         if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
 903                 if (tp->tun_flags & TUN_RWAIT) {
 904                         tp->tun_flags &= ~TUN_RWAIT;
 905                         wakeup(tp);
 906                 }
 907
 908                 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
 909                         TUN_UNLOCK(tp);
 910                         pgsigio(&tp->tun_sigio, SIGIO, 0);
 911                         TUN_LOCK(tp);
 912                 }
 913
 914                 selwakeuppri(&tp->tun_rsel, PZERO+1);
 915                 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 916                 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
 917         }
 918
 919         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
 920         TUN_UNLOCK(tp);
 921 } /* tunstart_l2 */
 922
 923 /* XXX: should return an error code so it can fail. */
 924 static void
 925 tuncreate(struct cdev *dev)
 926 {
 927         struct tuntap_driver *drv;
 928         struct tuntap_softc *tp;
 929         struct ifnet *ifp;
 930         struct ether_addr eaddr;
 931         int iflags;
 932         u_char type;
 933
 934         tp = dev->si_drv1;
 935         KASSERT(tp != NULL,
 936             ("si_drv1 should have been initialized at creation"));
 937
 938         drv = tp->tun_drv;
 939         iflags = IFF_MULTICAST;
 940         if ((tp->tun_flags & TUN_L2) != 0) {
 941                 type = IFT_ETHER;
 942                 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
 943         } else {
 944                 type = IFT_PPP;
 945                 iflags |= IFF_POINTOPOINT;
 946         }
 947         ifp = tp->tun_ifp = if_alloc(type);
 948         if (ifp == NULL)
 949                 panic("%s%d: failed to if_alloc() interface.\n",
 950                     drv->cdevsw.d_name, dev2unit(dev));
 951         ifp->if_softc = tp;
 952         if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
 953         ifp->if_ioctl = tunifioctl;
 954         ifp->if_flags = iflags;
 955         IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
 956         ifp->if_capabilities |= IFCAP_LINKSTATE;
 957         ifp->if_capenable |= IFCAP_LINKSTATE;
 958
 959         if ((tp->tun_flags & TUN_L2) != 0) {
 960                 ifp->if_mtu = ETHERMTU;
 961                 ifp->if_init = tunifinit;
 962                 ifp->if_start = tunstart_l2;
 963
 964                 ether_gen_addr(ifp, &eaddr);
 965                 ether_ifattach(ifp, eaddr.octet);
 966         } else {
 967                 ifp->if_mtu = TUNMTU;
 968                 ifp->if_start = tunstart;
 969                 ifp->if_output = tunoutput;
 970
 971                 ifp->if_snd.ifq_drv_maxlen = 0;
 972                 IFQ_SET_READY(&ifp->if_snd);
 973
 974                 if_attach(ifp);
 975                 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
 976         }
 977
 978         TUN_LOCK(tp);
 979         tp->tun_flags |= TUN_INITED;
 980         TUN_UNLOCK(tp);
 981
 982         TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
 983             ifp->if_xname, dev2unit(dev));
 984 }
 985
 986 static void
 987 tunrename(void *arg __unused, struct ifnet *ifp)
 988 {
 989         struct tuntap_softc *tp;
 990         int error;
 991
 992         if ((ifp->if_flags & IFF_RENAMING) == 0)
 993                 return;
 994
 995         if (tuntap_driver_from_ifnet(ifp) == NULL)
 996                 return;
 997
 998         /*
 999          * We need to grab the ioctl sx long enough to make sure the softc is
1000          * still there.  If it is, we can safely try to busy the tun device.
1001          * The busy may fail if the device is currently dying, in which case
1002          * we do nothing.  If it doesn't fail, the busy count stops the device
1003          * from dying until we've created the alias (that will then be
1004          * subsequently destroyed).
1005          */
1006         sx_xlock(&tun_ioctl_sx);
1007         tp = ifp->if_softc;
1008         if (tp == NULL) {
1009                 sx_xunlock(&tun_ioctl_sx);
1010                 return;
1011         }
1012         error = tun_busy(tp);
1013         sx_xunlock(&tun_ioctl_sx);
1014         if (error != 0)
1015                 return;
1016         if (tp->tun_alias != NULL) {
1017                 destroy_dev(tp->tun_alias);
1018                 tp->tun_alias = NULL;
1019         }
1020
1021         if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1022                 goto out;
1023
1024         /*
1025          * Failure's ok, aliases are created on a best effort basis.  If a
1026          * tun user/consumer decides to rename the interface to conflict with
1027          * another device (non-ifnet) on the system, we will assume they know
1028          * what they are doing.  make_dev_alias_p won't touch tun_alias on
1029          * failure, so we use it but ignore the return value.
1030          */
1031         make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1032             ifp->if_xname);
1033 out:
1034         tun_unbusy(tp);
1035 }
1036
1037 static int
1038 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1039 {
1040         struct ifnet    *ifp;
1041         struct tuntap_softc *tp;
1042         int error, tunflags;
1043
1044         tunflags = 0;
1045         CURVNET_SET(TD_TO_VNET(td));
1046         error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1047         if (error != 0) {
1048                 CURVNET_RESTORE();
1049                 return (error); /* Shouldn't happen */
1050         }
1051
1052         tp = dev->si_drv1;
1053         KASSERT(tp != NULL,
1054             ("si_drv1 should have been initialized at creation"));
1055
1056         TUN_LOCK(tp);
1057         if ((tp->tun_flags & TUN_INITED) == 0) {
1058                 TUN_UNLOCK(tp);
1059                 CURVNET_RESTORE();
1060                 return (ENXIO);
1061         }
1062         if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1063                 TUN_UNLOCK(tp);
1064                 CURVNET_RESTORE();
1065                 return (EBUSY);
1066         }
1067
1068         error = tun_busy_locked(tp);
1069         KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1070         ifp = TUN2IFP(tp);
1071
1072         if ((tp->tun_flags & TUN_L2) != 0) {
1073                 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1074                     sizeof(tp->tun_ether.octet));
1075
1076                 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1077                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1078
1079                 if (tapuponopen)
1080                         ifp->if_flags |= IFF_UP;
1081         }
1082
1083         tp->tun_pid = td->td_proc->p_pid;
1084         tp->tun_flags |= TUN_OPEN;
1085
1086         if_link_state_change(ifp, LINK_STATE_UP);
1087         TUNDEBUG(ifp, "open\n");
1088         TUN_UNLOCK(tp);
1089
1090         /*
1091          * This can fail with either ENOENT or EBUSY.  This is in the middle of
1092          * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1093          * the only cdevpriv dtor being set will be tundtor and the softc being
1094          * passed is constant for a given cdev.  We ignore the possible error
1095          * because of this as either "unlikely" or "not actually a problem."
1096          */
1097         (void)devfs_set_cdevpriv(tp, tundtor);
1098         CURVNET_RESTORE();
1099         return (0);
1100 }
1101
1102 /*
1103  * tundtor - tear down the device - mark i/f down & delete
1104  * routing info
1105  */
1106 static void
1107 tundtor(void *data)
1108 {
1109         struct proc *p;
1110         struct tuntap_softc *tp;
1111         struct ifnet *ifp;
1112         bool l2tun;
1113
1114         tp = data;
1115         p = curproc;
1116         ifp = TUN2IFP(tp);
1117
1118         TUN_LOCK(tp);
1119
1120         /*
1121          * Realistically, we can't be obstinate here.  This only means that the
1122          * tuntap device was closed out of order, and the last closer wasn't the
1123          * controller.  These are still good to know about, though, as software
1124          * should avoid multiple processes with a tuntap device open and
1125          * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1126          * parent).
1127          */
1128         if (p->p_pid != tp->tun_pid) {
1129                 log(LOG_INFO,
1130                     "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1131                     p->p_pid, p->p_comm, tp->tun_dev->si_name);
1132         }
1133
1134         /*
1135          * junk all pending output
1136          */
1137         CURVNET_SET(ifp->if_vnet);
1138
1139         l2tun = false;
1140         if ((tp->tun_flags & TUN_L2) != 0) {
1141                 l2tun = true;
1142                 IF_DRAIN(&ifp->if_snd);
1143         } else {
1144                 IFQ_PURGE(&ifp->if_snd);
1145         }
1146
1147         /* For vmnet, we won't do most of the address/route bits */
1148         if ((tp->tun_flags & TUN_VMNET) != 0 ||
1149             (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1150                 goto out;
1151
1152         if (ifp->if_flags & IFF_UP) {
1153                 TUN_UNLOCK(tp);
1154                 if_down(ifp);
1155                 TUN_LOCK(tp);
1156         }
1157
1158         /* Delete all addresses and routes which reference this interface. */
1159         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1160                 struct ifaddr *ifa;
1161
1162                 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1163                 TUN_UNLOCK(tp);
1164                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1165                         /* deal w/IPv4 PtP destination; unlocked read */
1166                         if (!l2tun && ifa->ifa_addr->sa_family == AF_INET) {
1167                                 rtinit(ifa, (int)RTM_DELETE,
1168                                     tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
1169                         } else {
1170                                 rtinit(ifa, (int)RTM_DELETE, 0);
1171                         }
1172                 }
1173                 if_purgeaddrs(ifp);
1174                 TUN_LOCK(tp);
1175         }
1176
1177 out:
1178         if_link_state_change(ifp, LINK_STATE_DOWN);
1179         CURVNET_RESTORE();
1180
1181         funsetown(&tp->tun_sigio);
1182         selwakeuppri(&tp->tun_rsel, PZERO + 1);
1183         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1184         TUNDEBUG (ifp, "closed\n");
1185         tp->tun_flags &= ~TUN_OPEN;
1186         tp->tun_pid = 0;
1187         tun_vnethdr_set(ifp, 0);
1188
1189         tun_unbusy_locked(tp);
1190         TUN_UNLOCK(tp);
1191 }
1192
1193 static void
1194 tuninit(struct ifnet *ifp)
1195 {
1196         struct tuntap_softc *tp = ifp->if_softc;
1197 #ifdef INET
1198         struct epoch_tracker et;
1199         struct ifaddr *ifa;
1200 #endif
1201
1202         TUNDEBUG(ifp, "tuninit\n");
1203
1204         TUN_LOCK(tp);
1205         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1206         if ((tp->tun_flags & TUN_L2) == 0) {
1207                 ifp->if_flags |= IFF_UP;
1208                 getmicrotime(&ifp->if_lastchange);
1209 #ifdef INET
1210                 NET_EPOCH_ENTER(et);
1211                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1212                         if (ifa->ifa_addr->sa_family == AF_INET) {
1213                                 struct sockaddr_in *si;
1214
1215                                 si = (struct sockaddr_in *)ifa->ifa_dstaddr;
1216                                 if (si && si->sin_addr.s_addr) {
1217                                         tp->tun_flags |= TUN_DSTADDR;
1218                                         break;
1219                                 }
1220                         }
1221                 }
1222                 NET_EPOCH_EXIT(et);
1223 #endif
1224                 TUN_UNLOCK(tp);
1225         } else {
1226                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1227                 TUN_UNLOCK(tp);
1228                 /* attempt to start output */
1229                 tunstart_l2(ifp);
1230         }
1231
1232 }
1233
1234 /*
1235  * Used only for l2 tunnel.
1236  */
1237 static void
1238 tunifinit(void *xtp)
1239 {
1240         struct tuntap_softc *tp;
1241
1242         tp = (struct tuntap_softc *)xtp;
1243         tuninit(tp->tun_ifp);
1244 }
1245
1246 /*
1247  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1248  * current value of ifp->if_capenable.
1249  */
1250 static void
1251 tun_caps_changed(struct ifnet *ifp)
1252 {
1253         uint64_t hwassist = 0;
1254
1255         TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1256         if (ifp->if_capenable & IFCAP_TXCSUM)
1257                 hwassist |= CSUM_TCP | CSUM_UDP;
1258         if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1259                 hwassist |= CSUM_TCP_IPV6
1260                     | CSUM_UDP_IPV6;
1261         if (ifp->if_capenable & IFCAP_TSO4)
1262                 hwassist |= CSUM_IP_TSO;
1263         if (ifp->if_capenable & IFCAP_TSO6)
1264                 hwassist |= CSUM_IP6_TSO;
1265         ifp->if_hwassist = hwassist;
1266 }
1267
1268 /*
1269  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1270  * if_capabilities and if_capenable as needed.
1271  */
1272 static void
1273 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1274 {
1275         struct tuntap_softc *tp = ifp->if_softc;
1276
1277         TUN_LOCK_ASSERT(tp);
1278
1279         if (tp->tun_vhdrlen == vhdrlen)
1280                 return;
1281
1282         /*
1283          * Update if_capabilities to reflect the
1284          * functionalities offered by the virtio-net
1285          * header.
1286          */
1287         if (vhdrlen != 0)
1288                 ifp->if_capabilities |=
1289                         TAP_VNET_HDR_CAPS;
1290         else
1291                 ifp->if_capabilities &=
1292                         ~TAP_VNET_HDR_CAPS;
1293         /*
1294          * Disable any capabilities that we don't
1295          * support anymore.
1296          */
1297         ifp->if_capenable &= ifp->if_capabilities;
1298         tun_caps_changed(ifp);
1299         tp->tun_vhdrlen = vhdrlen;
1300
1301         TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1302             vhdrlen, ifp->if_capabilities);
1303 }
1304
1305 /*
1306  * Process an ioctl request.
1307  */
1308 static int
1309 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1310 {
1311         struct ifreq *ifr = (struct ifreq *)data;
1312         struct tuntap_softc *tp;
1313         struct ifstat *ifs;
1314         struct ifmediareq       *ifmr;
1315         int             dummy, error = 0;
1316         bool            l2tun;
1317
1318         ifmr = NULL;
1319         sx_xlock(&tun_ioctl_sx);
1320         tp = ifp->if_softc;
1321         if (tp == NULL) {
1322                 error = ENXIO;
1323                 goto bad;
1324         }
1325         l2tun = (tp->tun_flags & TUN_L2) != 0;
1326         switch(cmd) {
1327         case SIOCGIFSTATUS:
1328                 ifs = (struct ifstat *)data;
1329                 TUN_LOCK(tp);
1330                 if (tp->tun_pid)
1331                         snprintf(ifs->ascii, sizeof(ifs->ascii),
1332                             "\tOpened by PID %d\n", tp->tun_pid);
1333                 else
1334                         ifs->ascii[0] = '\0';
1335                 TUN_UNLOCK(tp);
1336                 break;
1337         case SIOCSIFADDR:
1338                 if (l2tun)
1339                         error = ether_ioctl(ifp, cmd, data);
1340                 else
1341                         tuninit(ifp);
1342                 if (error == 0)
1343                     TUNDEBUG(ifp, "address set\n");
1344                 break;
1345         case SIOCSIFMTU:
1346                 ifp->if_mtu = ifr->ifr_mtu;
1347                 TUNDEBUG(ifp, "mtu set\n");
1348                 break;
1349         case SIOCSIFFLAGS:
1350         case SIOCADDMULTI:
1351         case SIOCDELMULTI:
1352                 break;
1353         case SIOCGIFMEDIA:
1354                 if (!l2tun) {
1355                         error = EINVAL;
1356                         break;
1357                 }
1358
1359                 ifmr = (struct ifmediareq *)data;
1360                 dummy = ifmr->ifm_count;
1361                 ifmr->ifm_count = 1;
1362                 ifmr->ifm_status = IFM_AVALID;
1363                 ifmr->ifm_active = IFM_ETHER;
1364                 if (tp->tun_flags & TUN_OPEN)
1365                         ifmr->ifm_status |= IFM_ACTIVE;
1366                 ifmr->ifm_current = ifmr->ifm_active;
1367                 if (dummy >= 1) {
1368                         int media = IFM_ETHER;
1369                         error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1370                 }
1371                 break;
1372         case SIOCSIFCAP:
1373                 TUN_LOCK(tp);
1374                 ifp->if_capenable = ifr->ifr_reqcap;
1375                 tun_caps_changed(ifp);
1376                 TUN_UNLOCK(tp);
1377                 VLAN_CAPABILITIES(ifp);
1378                 break;
1379         default:
1380                 if (l2tun) {
1381                         error = ether_ioctl(ifp, cmd, data);
1382                 } else {
1383                         error = EINVAL;
1384                 }
1385         }
1386 bad:
1387         sx_xunlock(&tun_ioctl_sx);
1388         return (error);
1389 }
1390
1391 /*
1392  * tunoutput - queue packets from higher level ready to put out.
1393  */
1394 static int
1395 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1396     struct route *ro)
1397 {
1398         struct tuntap_softc *tp = ifp->if_softc;
1399         u_short cached_tun_flags;
1400         int error;
1401         u_int32_t af;
1402
1403         TUNDEBUG (ifp, "tunoutput\n");
1404
1405 #ifdef MAC
1406         error = mac_ifnet_check_transmit(ifp, m0);
1407         if (error) {
1408                 m_freem(m0);
1409                 return (error);
1410         }
1411 #endif
1412
1413         /* Could be unlocked read? */
1414         TUN_LOCK(tp);
1415         cached_tun_flags = tp->tun_flags;
1416         TUN_UNLOCK(tp);
1417         if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1418                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1419                 m_freem (m0);
1420                 return (EHOSTDOWN);
1421         }
1422
1423         if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1424                 m_freem (m0);
1425                 return (EHOSTDOWN);
1426         }
1427
1428         /* BPF writes need to be handled specially. */
1429         if (dst->sa_family == AF_UNSPEC)
1430                 bcopy(dst->sa_data, &af, sizeof(af));
1431         else
1432                 af = dst->sa_family;
1433
1434         if (bpf_peers_present(ifp->if_bpf))
1435                 bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1436
1437         /* prepend sockaddr? this may abort if the mbuf allocation fails */
1438         if (cached_tun_flags & TUN_LMODE) {
1439                 /* allocate space for sockaddr */
1440                 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1441
1442                 /* if allocation failed drop packet */
1443                 if (m0 == NULL) {
1444                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1445                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1446                         return (ENOBUFS);
1447                 } else {
1448                         bcopy(dst, m0->m_data, dst->sa_len);
1449                 }
1450         }
1451
1452         if (cached_tun_flags & TUN_IFHEAD) {
1453                 /* Prepend the address family */
1454                 M_PREPEND(m0, 4, M_NOWAIT);
1455
1456                 /* if allocation failed drop packet */
1457                 if (m0 == NULL) {
1458                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1459                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1460                         return (ENOBUFS);
1461                 } else
1462                         *(u_int32_t *)m0->m_data = htonl(af);
1463         } else {
1464 #ifdef INET
1465                 if (af != AF_INET)
1466 #endif
1467                 {
1468                         m_freem(m0);
1469                         return (EAFNOSUPPORT);
1470                 }
1471         }
1472
1473         error = (ifp->if_transmit)(ifp, m0);
1474         if (error)
1475                 return (ENOBUFS);
1476         if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1477         return (0);
1478 }
1479
1480 /*
1481  * the cdevsw interface is now pretty minimal.
1482  */
1483 static  int
1484 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1485     struct thread *td)
1486 {
1487         struct ifreq ifr, *ifrp;
1488         struct tuntap_softc *tp = dev->si_drv1;
1489         struct ifnet *ifp = TUN2IFP(tp);
1490         struct tuninfo *tunp;
1491         int error, iflags, ival;
1492         bool    l2tun;
1493
1494         l2tun = (tp->tun_flags & TUN_L2) != 0;
1495         if (l2tun) {
1496                 /* tap specific ioctls */
1497                 switch(cmd) {
1498                 /* VMware/VMnet port ioctl's */
1499 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1500     defined(COMPAT_FREEBSD4)
1501                 case _IO('V', 0):
1502                         ival = IOCPARM_IVAL(data);
1503                         data = (caddr_t)&ival;
1504                         /* FALLTHROUGH */
1505 #endif
1506                 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1507                         iflags = *(int *)data;
1508                         iflags &= TUN_VMIO_FLAG_MASK;
1509                         iflags &= ~IFF_CANTCHANGE;
1510                         iflags |= IFF_UP;
1511
1512                         TUN_LOCK(tp);
1513                         ifp->if_flags = iflags |
1514                             (ifp->if_flags & IFF_CANTCHANGE);
1515                         TUN_UNLOCK(tp);
1516
1517                         return (0);
1518                 case SIOCGIFADDR:       /* get MAC address of the remote side */
1519                         TUN_LOCK(tp);
1520                         bcopy(&tp->tun_ether.octet, data,
1521                             sizeof(tp->tun_ether.octet));
1522                         TUN_UNLOCK(tp);
1523
1524                         return (0);
1525                 case SIOCSIFADDR:       /* set MAC address of the remote side */
1526                         TUN_LOCK(tp);
1527                         bcopy(data, &tp->tun_ether.octet,
1528                             sizeof(tp->tun_ether.octet));
1529                         TUN_UNLOCK(tp);
1530
1531                         return (0);
1532                 case TAPSVNETHDR:
1533                         ival = *(int *)data;
1534                         if (ival != 0 &&
1535                             ival != sizeof(struct virtio_net_hdr) &&
1536                             ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1537                                 return (EINVAL);
1538                         }
1539                         TUN_LOCK(tp);
1540                         tun_vnethdr_set(ifp, ival);
1541                         TUN_UNLOCK(tp);
1542
1543                         return (0);
1544                 case TAPGVNETHDR:
1545                         TUN_LOCK(tp);
1546                         *(int *)data = tp->tun_vhdrlen;
1547                         TUN_UNLOCK(tp);
1548
1549                         return (0);
1550                 }
1551
1552                 /* Fall through to the common ioctls if unhandled */
1553         } else {
1554                 switch (cmd) {
1555                 case TUNSLMODE:
1556                         TUN_LOCK(tp);
1557                         if (*(int *)data) {
1558                                 tp->tun_flags |= TUN_LMODE;
1559                                 tp->tun_flags &= ~TUN_IFHEAD;
1560                         } else
1561                                 tp->tun_flags &= ~TUN_LMODE;
1562                         TUN_UNLOCK(tp);
1563
1564                         return (0);
1565                 case TUNSIFHEAD:
1566                         TUN_LOCK(tp);
1567                         if (*(int *)data) {
1568                                 tp->tun_flags |= TUN_IFHEAD;
1569                                 tp->tun_flags &= ~TUN_LMODE;
1570                         } else
1571                                 tp->tun_flags &= ~TUN_IFHEAD;
1572                         TUN_UNLOCK(tp);
1573
1574                         return (0);
1575                 case TUNGIFHEAD:
1576                         TUN_LOCK(tp);
1577                         *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1578                         TUN_UNLOCK(tp);
1579
1580                         return (0);
1581                 case TUNSIFMODE:
1582                         /* deny this if UP */
1583                         if (TUN2IFP(tp)->if_flags & IFF_UP)
1584                                 return (EBUSY);
1585
1586                         switch (*(int *)data & ~IFF_MULTICAST) {
1587                         case IFF_POINTOPOINT:
1588                         case IFF_BROADCAST:
1589                                 TUN_LOCK(tp);
1590                                 TUN2IFP(tp)->if_flags &=
1591                                     ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1592                                 TUN2IFP(tp)->if_flags |= *(int *)data;
1593                                 TUN_UNLOCK(tp);
1594
1595                                 break;
1596                         default:
1597                                 return (EINVAL);
1598                         }
1599
1600                         return (0);
1601                 case TUNSIFPID:
1602                         TUN_LOCK(tp);
1603                         tp->tun_pid = curthread->td_proc->p_pid;
1604                         TUN_UNLOCK(tp);
1605
1606                         return (0);
1607                 }
1608                 /* Fall through to the common ioctls if unhandled */
1609         }
1610
1611         switch (cmd) {
1612         case TUNGIFNAME:
1613                 ifrp = (struct ifreq *)data;
1614                 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1615
1616                 return (0);
1617         case TUNSIFINFO:
1618                 tunp = (struct tuninfo *)data;
1619                 if (TUN2IFP(tp)->if_type != tunp->type)
1620                         return (EPROTOTYPE);
1621                 TUN_LOCK(tp);
1622                 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1623                         strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1624                         ifr.ifr_mtu = tunp->mtu;
1625                         CURVNET_SET(TUN2IFP(tp)->if_vnet);
1626                         error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1627                             (caddr_t)&ifr, td);
1628                         CURVNET_RESTORE();
1629                         if (error) {
1630                                 TUN_UNLOCK(tp);
1631                                 return (error);
1632                         }
1633                 }
1634                 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1635                 TUN_UNLOCK(tp);
1636                 break;
1637         case TUNGIFINFO:
1638                 tunp = (struct tuninfo *)data;
1639                 TUN_LOCK(tp);
1640                 tunp->mtu = TUN2IFP(tp)->if_mtu;
1641                 tunp->type = TUN2IFP(tp)->if_type;
1642                 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1643                 TUN_UNLOCK(tp);
1644                 break;
1645         case TUNSDEBUG:
1646                 tundebug = *(int *)data;
1647                 break;
1648         case TUNGDEBUG:
1649                 *(int *)data = tundebug;
1650                 break;
1651         case FIONBIO:
1652                 break;
1653         case FIOASYNC:
1654                 TUN_LOCK(tp);
1655                 if (*(int *)data)
1656                         tp->tun_flags |= TUN_ASYNC;
1657                 else
1658                         tp->tun_flags &= ~TUN_ASYNC;
1659                 TUN_UNLOCK(tp);
1660                 break;
1661         case FIONREAD:
1662                 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1663                         struct mbuf *mb;
1664                         IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1665                         IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1666                         for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1667                                 *(int *)data += mb->m_len;
1668                         IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1669                 } else
1670                         *(int *)data = 0;
1671                 break;
1672         case FIOSETOWN:
1673                 return (fsetown(*(int *)data, &tp->tun_sigio));
1674
1675         case FIOGETOWN:
1676                 *(int *)data = fgetown(&tp->tun_sigio);
1677                 return (0);
1678
1679         /* This is deprecated, FIOSETOWN should be used instead. */
1680         case TIOCSPGRP:
1681                 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1682
1683         /* This is deprecated, FIOGETOWN should be used instead. */
1684         case TIOCGPGRP:
1685                 *(int *)data = -fgetown(&tp->tun_sigio);
1686                 return (0);
1687
1688         default:
1689                 return (ENOTTY);
1690         }
1691         return (0);
1692 }
1693
1694 /*
1695  * The cdevsw read interface - reads a packet at a time, or at
1696  * least as much of a packet as can be read.
1697  */
1698 static  int
1699 tunread(struct cdev *dev, struct uio *uio, int flag)
1700 {
1701         struct tuntap_softc *tp = dev->si_drv1;
1702         struct ifnet    *ifp = TUN2IFP(tp);
1703         struct mbuf     *m;
1704         size_t          len;
1705         int             error = 0;
1706
1707         TUNDEBUG (ifp, "read\n");
1708         TUN_LOCK(tp);
1709         if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1710                 TUN_UNLOCK(tp);
1711                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1712                 return (EHOSTDOWN);
1713         }
1714
1715         tp->tun_flags &= ~TUN_RWAIT;
1716
1717         for (;;) {
1718                 IFQ_DEQUEUE(&ifp->if_snd, m);
1719                 if (m != NULL)
1720                         break;
1721                 if (flag & O_NONBLOCK) {
1722                         TUN_UNLOCK(tp);
1723                         return (EWOULDBLOCK);
1724                 }
1725                 tp->tun_flags |= TUN_RWAIT;
1726                 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1727                     "tunread", 0);
1728                 if (error != 0) {
1729                         TUN_UNLOCK(tp);
1730                         return (error);
1731                 }
1732         }
1733         TUN_UNLOCK(tp);
1734
1735         if ((tp->tun_flags & TUN_L2) != 0)
1736                 BPF_MTAP(ifp, m);
1737
1738         len = min(tp->tun_vhdrlen, uio->uio_resid);
1739         if (len > 0) {
1740                 struct virtio_net_hdr_mrg_rxbuf vhdr;
1741
1742                 bzero(&vhdr, sizeof(vhdr));
1743                 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1744                         m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1745                 }
1746
1747                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1748                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1749                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1750                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1751                     vhdr.hdr.csum_offset);
1752                 error = uiomove(&vhdr, len, uio);
1753         }
1754
1755         while (m && uio->uio_resid > 0 && error == 0) {
1756                 len = min(uio->uio_resid, m->m_len);
1757                 if (len != 0)
1758                         error = uiomove(mtod(m, void *), len, uio);
1759                 m = m_free(m);
1760         }
1761
1762         if (m) {
1763                 TUNDEBUG(ifp, "Dropping mbuf\n");
1764                 m_freem(m);
1765         }
1766         return (error);
1767 }
1768
1769 static int
1770 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1771             struct virtio_net_hdr_mrg_rxbuf *vhdr)
1772 {
1773         struct ether_header *eh;
1774         struct ifnet *ifp;
1775
1776         ifp = TUN2IFP(tp);
1777
1778         /*
1779          * Only pass a unicast frame to ether_input(), if it would
1780          * actually have been received by non-virtual hardware.
1781          */
1782         if (m->m_len < sizeof(struct ether_header)) {
1783                 m_freem(m);
1784                 return (0);
1785         }
1786
1787         eh = mtod(m, struct ether_header *);
1788
1789         if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1790             !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1791             bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1792                 m_freem(m);
1793                 return (0);
1794         }
1795
1796         if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1797                 m_freem(m);
1798                 return (0);
1799         }
1800
1801         /* Pass packet up to parent. */
1802         CURVNET_SET(ifp->if_vnet);
1803         (*ifp->if_input)(ifp, m);
1804         CURVNET_RESTORE();
1805         /* ibytes are counted in parent */
1806         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1807         return (0);
1808 }
1809
1810 static int
1811 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1812 {
1813         struct epoch_tracker et;
1814         struct ifnet *ifp;
1815         int family, isr;
1816
1817         ifp = TUN2IFP(tp);
1818         /* Could be unlocked read? */
1819         TUN_LOCK(tp);
1820         if (tp->tun_flags & TUN_IFHEAD) {
1821                 TUN_UNLOCK(tp);
1822                 if (m->m_len < sizeof(family) &&
1823                 (m = m_pullup(m, sizeof(family))) == NULL)
1824                         return (ENOBUFS);
1825                 family = ntohl(*mtod(m, u_int32_t *));
1826                 m_adj(m, sizeof(family));
1827         } else {
1828                 TUN_UNLOCK(tp);
1829                 family = AF_INET;
1830         }
1831
1832         BPF_MTAP2(ifp, &family, sizeof(family), m);
1833
1834         switch (family) {
1835 #ifdef INET
1836         case AF_INET:
1837                 isr = NETISR_IP;
1838                 break;
1839 #endif
1840 #ifdef INET6
1841         case AF_INET6:
1842                 isr = NETISR_IPV6;
1843                 break;
1844 #endif
1845         default:
1846                 m_freem(m);
1847                 return (EAFNOSUPPORT);
1848         }
1849         random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1850         if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1851         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1852         CURVNET_SET(ifp->if_vnet);
1853         M_SETFIB(m, ifp->if_fib);
1854         NET_EPOCH_ENTER(et);
1855         netisr_dispatch(isr, m);
1856         NET_EPOCH_EXIT(et);
1857         CURVNET_RESTORE();
1858         return (0);
1859 }
1860
1861 /*
1862  * the cdevsw write interface - an atomic write is a packet - or else!
1863  */
1864 static  int
1865 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1866 {
1867         struct virtio_net_hdr_mrg_rxbuf vhdr;
1868         struct tuntap_softc *tp;
1869         struct ifnet    *ifp;
1870         struct mbuf     *m;
1871         uint32_t        mru;
1872         int             align, vhdrlen, error;
1873         bool            l2tun;
1874
1875         tp = dev->si_drv1;
1876         ifp = TUN2IFP(tp);
1877         TUNDEBUG(ifp, "tunwrite\n");
1878         if ((ifp->if_flags & IFF_UP) != IFF_UP)
1879                 /* ignore silently */
1880                 return (0);
1881
1882         if (uio->uio_resid == 0)
1883                 return (0);
1884
1885         l2tun = (tp->tun_flags & TUN_L2) != 0;
1886         mru = l2tun ? TAPMRU : TUNMRU;
1887         vhdrlen = tp->tun_vhdrlen;
1888         align = 0;
1889         if (l2tun) {
1890                 align = ETHER_ALIGN;
1891                 mru += vhdrlen;
1892         } else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1893                 mru += sizeof(uint32_t);        /* family */
1894         if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1895                 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1896                 return (EIO);
1897         }
1898
1899         if (vhdrlen > 0) {
1900                 error = uiomove(&vhdr, vhdrlen, uio);
1901                 if (error != 0)
1902                         return (error);
1903                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1904                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1905                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1906                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1907                     vhdr.hdr.csum_offset);
1908         }
1909
1910         if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1911                 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1912                 return (ENOBUFS);
1913         }
1914
1915         m->m_pkthdr.rcvif = ifp;
1916 #ifdef MAC
1917         mac_ifnet_create_mbuf(ifp, m);
1918 #endif
1919
1920         if (l2tun)
1921                 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1922
1923         return (tunwrite_l3(tp, m));
1924 }
1925
1926 /*
1927  * tunpoll - the poll interface, this is only useful on reads
1928  * really. The write detect always returns true, write never blocks
1929  * anyway, it either accepts the packet or drops it.
1930  */
1931 static  int
1932 tunpoll(struct cdev *dev, int events, struct thread *td)
1933 {
1934         struct tuntap_softc *tp = dev->si_drv1;
1935         struct ifnet    *ifp = TUN2IFP(tp);
1936         int             revents = 0;
1937
1938         TUNDEBUG(ifp, "tunpoll\n");
1939
1940         if (events & (POLLIN | POLLRDNORM)) {
1941                 IFQ_LOCK(&ifp->if_snd);
1942                 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1943                         TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1944                         revents |= events & (POLLIN | POLLRDNORM);
1945                 } else {
1946                         TUNDEBUG(ifp, "tunpoll waiting\n");
1947                         selrecord(td, &tp->tun_rsel);
1948                 }
1949                 IFQ_UNLOCK(&ifp->if_snd);
1950         }
1951         revents |= events & (POLLOUT | POLLWRNORM);
1952
1953         return (revents);
1954 }
1955
1956 /*
1957  * tunkqfilter - support for the kevent() system call.
1958  */
1959 static int
1960 tunkqfilter(struct cdev *dev, struct knote *kn)
1961 {
1962         struct tuntap_softc     *tp = dev->si_drv1;
1963         struct ifnet    *ifp = TUN2IFP(tp);
1964
1965         switch(kn->kn_filter) {
1966         case EVFILT_READ:
1967                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1968                     ifp->if_xname, dev2unit(dev));
1969                 kn->kn_fop = &tun_read_filterops;
1970                 break;
1971
1972         case EVFILT_WRITE:
1973                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1974                     ifp->if_xname, dev2unit(dev));
1975                 kn->kn_fop = &tun_write_filterops;
1976                 break;
1977
1978         default:
1979                 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1980                     ifp->if_xname, dev2unit(dev));
1981                 return(EINVAL);
1982         }
1983
1984         kn->kn_hook = tp;
1985         knlist_add(&tp->tun_rsel.si_note, kn, 0);
1986
1987         return (0);
1988 }
1989
1990 /*
1991  * Return true of there is data in the interface queue.
1992  */
1993 static int
1994 tunkqread(struct knote *kn, long hint)
1995 {
1996         int                     ret;
1997         struct tuntap_softc     *tp = kn->kn_hook;
1998         struct cdev             *dev = tp->tun_dev;
1999         struct ifnet    *ifp = TUN2IFP(tp);
2000
2001         if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2002                 TUNDEBUG(ifp,
2003                     "%s have data in the queue.  Len = %d, minor = %#x\n",
2004                     ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2005                 ret = 1;
2006         } else {
2007                 TUNDEBUG(ifp,
2008                     "%s waiting for data, minor = %#x\n", ifp->if_xname,
2009                     dev2unit(dev));
2010                 ret = 0;
2011         }
2012
2013         return (ret);
2014 }
2015
2016 /*
2017  * Always can write, always return MTU in kn->data.
2018  */
2019 static int
2020 tunkqwrite(struct knote *kn, long hint)
2021 {
2022         struct tuntap_softc     *tp = kn->kn_hook;
2023         struct ifnet    *ifp = TUN2IFP(tp);
2024
2025         kn->kn_data = ifp->if_mtu;
2026
2027         return (1);
2028 }
2029
2030 static void
2031 tunkqdetach(struct knote *kn)
2032 {
2033         struct tuntap_softc     *tp = kn->kn_hook;
2034
2035         knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2036 }