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 #include <netinet/in.h>
  92 #ifdef INET
  93 #include <netinet/ip.h>
  94 #endif
  95 #ifdef INET6
  96 #include <netinet/ip6.h>
  97 #include <netinet6/ip6_var.h>
  98 #endif
  99 #include <netinet/udp.h>
 100 #include <netinet/tcp.h>
 101 #include <net/bpf.h>
 102 #include <net/if_tap.h>
 103 #include <net/if_tun.h>
 104
 105 #include <dev/virtio/network/virtio_net.h>
 106
 107 #include <sys/queue.h>
 108 #include <sys/condvar.h>
 109 #include <security/mac/mac_framework.h>
 110
 111 struct tuntap_driver;
 112
 113 /*
 114  * tun_list is protected by global tunmtx.  Other mutable fields are
 115  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
 116  * static for the duration of a tunnel interface.
 117  */
 118 struct tuntap_softc {
 119         TAILQ_ENTRY(tuntap_softc)        tun_list;
 120         struct cdev                     *tun_alias;
 121         struct cdev                     *tun_dev;
 122         u_short                          tun_flags;     /* misc flags */
 123 #define TUN_OPEN        0x0001
 124 #define TUN_INITED      0x0002
 125 #define TUN_UNUSED1     0x0008
 126 #define TUN_DSTADDR     0x0010
 127 #define TUN_LMODE       0x0020
 128 #define TUN_RWAIT       0x0040
 129 #define TUN_ASYNC       0x0080
 130 #define TUN_IFHEAD      0x0100
 131 #define TUN_DYING       0x0200
 132 #define TUN_L2          0x0400
 133 #define TUN_VMNET       0x0800
 134
 135 #define TUN_DRIVER_IDENT_MASK   (TUN_L2 | TUN_VMNET)
 136 #define TUN_READY               (TUN_OPEN | TUN_INITED)
 137
 138         pid_t                    tun_pid;       /* owning pid */
 139         struct ifnet            *tun_ifp;       /* the interface */
 140         struct sigio            *tun_sigio;     /* async I/O info */
 141         struct tuntap_driver    *tun_drv;       /* appropriate driver */
 142         struct selinfo           tun_rsel;      /* read select */
 143         struct mtx               tun_mtx;       /* softc field mutex */
 144         struct cv                tun_cv;        /* for ref'd dev destroy */
 145         struct ether_addr        tun_ether;     /* remote address */
 146         int                      tun_busy;      /* busy count */
 147         int                      tun_vhdrlen;   /* virtio-net header length */
 148 };
 149 #define TUN2IFP(sc)     ((sc)->tun_ifp)
 150
 151 #define TUNDEBUG        if (tundebug) if_printf
 152
 153 #define TUN_LOCK(tp)            mtx_lock(&(tp)->tun_mtx)
 154 #define TUN_UNLOCK(tp)          mtx_unlock(&(tp)->tun_mtx)
 155 #define TUN_LOCK_ASSERT(tp)     mtx_assert(&(tp)->tun_mtx, MA_OWNED);
 156
 157 #define TUN_VMIO_FLAG_MASK      0x0fff
 158
 159 /*
 160  * Interface capabilities of a tap device that supports the virtio-net
 161  * header.
 162  */
 163 #define TAP_VNET_HDR_CAPS       (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6       \
 164                                 | IFCAP_VLAN_HWCSUM                     \
 165                                 | IFCAP_TSO | IFCAP_LRO                 \
 166                                 | IFCAP_VLAN_HWTSO)
 167
 168 #define TAP_ALL_OFFLOAD         (CSUM_TSO | CSUM_TCP | CSUM_UDP |\
 169                                     CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
 170
 171 /*
 172  * All mutable global variables in if_tun are locked using tunmtx, with
 173  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
 174  * which are static after setup.
 175  */
 176 static struct mtx tunmtx;
 177 static eventhandler_tag arrival_tag;
 178 static eventhandler_tag clone_tag;
 179 static const char tunname[] = "tun";
 180 static const char tapname[] = "tap";
 181 static const char vmnetname[] = "vmnet";
 182 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
 183 static int tundebug = 0;
 184 static int tundclone = 1;
 185 static int tap_allow_uopen = 0; /* allow user devfs cloning */
 186 static int tapuponopen = 0;     /* IFF_UP on open() */
 187 static int tapdclone = 1;       /* enable devfs cloning */
 188
 189 static TAILQ_HEAD(,tuntap_softc)        tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
 190 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
 191
 192 static struct sx tun_ioctl_sx;
 193 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
 194
 195 SYSCTL_DECL(_net_link);
 196 /* tun */
 197 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 198     "IP tunnel software network interface");
 199 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
 200     "Enable legacy devfs interface creation");
 201
 202 /* tap */
 203 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 204     "Ethernet tunnel software network interface");
 205 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
 206     "Enable legacy devfs interface creation for all users");
 207 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
 208     "Bring interface up when /dev/tap is opened");
 209 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
 210     "Enable legacy devfs interface creation");
 211 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
 212
 213 static int      tun_create_device(struct tuntap_driver *drv, int unit,
 214     struct ucred *cr, struct cdev **dev, const char *name);
 215 static int      tun_busy_locked(struct tuntap_softc *tp);
 216 static void     tun_unbusy_locked(struct tuntap_softc *tp);
 217 static int      tun_busy(struct tuntap_softc *tp);
 218 static void     tun_unbusy(struct tuntap_softc *tp);
 219
 220 static int      tuntap_name2info(const char *name, int *unit, int *flags);
 221 static void     tunclone(void *arg, struct ucred *cred, char *name,
 222                     int namelen, struct cdev **dev);
 223 static void     tuncreate(struct cdev *dev);
 224 static void     tundtor(void *data);
 225 static void     tunrename(void *arg, struct ifnet *ifp);
 226 static int      tunifioctl(struct ifnet *, u_long, caddr_t);
 227 static void     tuninit(struct ifnet *);
 228 static void     tunifinit(void *xtp);
 229 static int      tuntapmodevent(module_t, int, void *);
 230 static int      tunoutput(struct ifnet *, struct mbuf *,
 231                     const struct sockaddr *, struct route *ro);
 232 static void     tunstart(struct ifnet *);
 233 static void     tunstart_l2(struct ifnet *);
 234
 235 static int      tun_clone_match(struct if_clone *ifc, const char *name);
 236 static int      tap_clone_match(struct if_clone *ifc, const char *name);
 237 static int      vmnet_clone_match(struct if_clone *ifc, const char *name);
 238 static int      tun_clone_create(struct if_clone *, char *, size_t, caddr_t);
 239 static int      tun_clone_destroy(struct if_clone *, struct ifnet *);
 240 static void     tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
 241
 242 static d_open_t         tunopen;
 243 static d_read_t         tunread;
 244 static d_write_t        tunwrite;
 245 static d_ioctl_t        tunioctl;
 246 static d_poll_t         tunpoll;
 247 static d_kqfilter_t     tunkqfilter;
 248
 249 static int              tunkqread(struct knote *, long);
 250 static int              tunkqwrite(struct knote *, long);
 251 static void             tunkqdetach(struct knote *);
 252
 253 static struct filterops tun_read_filterops = {
 254         .f_isfd =       1,
 255         .f_attach =     NULL,
 256         .f_detach =     tunkqdetach,
 257         .f_event =      tunkqread,
 258 };
 259
 260 static struct filterops tun_write_filterops = {
 261         .f_isfd =       1,
 262         .f_attach =     NULL,
 263         .f_detach =     tunkqdetach,
 264         .f_event =      tunkqwrite,
 265 };
 266
 267 static struct tuntap_driver {
 268         struct cdevsw            cdevsw;
 269         int                      ident_flags;
 270         struct unrhdr           *unrhdr;
 271         struct clonedevs        *clones;
 272         ifc_match_t             *clone_match_fn;
 273         ifc_create_t            *clone_create_fn;
 274         ifc_destroy_t           *clone_destroy_fn;
 275 } tuntap_drivers[] = {
 276         {
 277                 .ident_flags =  0,
 278                 .cdevsw =       {
 279                     .d_version =        D_VERSION,
 280                     .d_flags =          D_NEEDMINOR,
 281                     .d_open =           tunopen,
 282                     .d_read =           tunread,
 283                     .d_write =          tunwrite,
 284                     .d_ioctl =          tunioctl,
 285                     .d_poll =           tunpoll,
 286                     .d_kqfilter =       tunkqfilter,
 287                     .d_name =           tunname,
 288                 },
 289                 .clone_match_fn =       tun_clone_match,
 290                 .clone_create_fn =      tun_clone_create,
 291                 .clone_destroy_fn =     tun_clone_destroy,
 292         },
 293         {
 294                 .ident_flags =  TUN_L2,
 295                 .cdevsw =       {
 296                     .d_version =        D_VERSION,
 297                     .d_flags =          D_NEEDMINOR,
 298                     .d_open =           tunopen,
 299                     .d_read =           tunread,
 300                     .d_write =          tunwrite,
 301                     .d_ioctl =          tunioctl,
 302                     .d_poll =           tunpoll,
 303                     .d_kqfilter =       tunkqfilter,
 304                     .d_name =           tapname,
 305                 },
 306                 .clone_match_fn =       tap_clone_match,
 307                 .clone_create_fn =      tun_clone_create,
 308                 .clone_destroy_fn =     tun_clone_destroy,
 309         },
 310         {
 311                 .ident_flags =  TUN_L2 | TUN_VMNET,
 312                 .cdevsw =       {
 313                     .d_version =        D_VERSION,
 314                     .d_flags =          D_NEEDMINOR,
 315                     .d_open =           tunopen,
 316                     .d_read =           tunread,
 317                     .d_write =          tunwrite,
 318                     .d_ioctl =          tunioctl,
 319                     .d_poll =           tunpoll,
 320                     .d_kqfilter =       tunkqfilter,
 321                     .d_name =           vmnetname,
 322                 },
 323                 .clone_match_fn =       vmnet_clone_match,
 324                 .clone_create_fn =      tun_clone_create,
 325                 .clone_destroy_fn =     tun_clone_destroy,
 326         },
 327 };
 328
 329 struct tuntap_driver_cloner {
 330         SLIST_ENTRY(tuntap_driver_cloner)        link;
 331         struct tuntap_driver                    *drv;
 332         struct if_clone                         *cloner;
 333 };
 334
 335 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
 336     SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
 337
 338 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
 339
 340 /*
 341  * Mechanism for marking a tunnel device as busy so that we can safely do some
 342  * orthogonal operations (such as operations on devices) without racing against
 343  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
 344  * open, to be woken up when the condition is alleviated.
 345  */
 346 static int
 347 tun_busy_locked(struct tuntap_softc *tp)
 348 {
 349
 350         TUN_LOCK_ASSERT(tp);
 351         if ((tp->tun_flags & TUN_DYING) != 0) {
 352                 /*
 353                  * Perhaps unintuitive, but the device is busy going away.
 354                  * Other interpretations of EBUSY from tun_busy make little
 355                  * sense, since making a busy device even more busy doesn't
 356                  * sound like a problem.
 357                  */
 358                 return (EBUSY);
 359         }
 360
 361         ++tp->tun_busy;
 362         return (0);
 363 }
 364
 365 static void
 366 tun_unbusy_locked(struct tuntap_softc *tp)
 367 {
 368
 369         TUN_LOCK_ASSERT(tp);
 370         KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
 371
 372         --tp->tun_busy;
 373         /* Wake up anything that may be waiting on our busy tunnel. */
 374         if (tp->tun_busy == 0)
 375                 cv_broadcast(&tp->tun_cv);
 376 }
 377
 378 static int
 379 tun_busy(struct tuntap_softc *tp)
 380 {
 381         int ret;
 382
 383         TUN_LOCK(tp);
 384         ret = tun_busy_locked(tp);
 385         TUN_UNLOCK(tp);
 386         return (ret);
 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 static int
 478 tun_clone_match(struct if_clone *ifc, const char *name)
 479 {
 480         int tunflags;
 481
 482         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 483                 if ((tunflags & TUN_L2) == 0)
 484                         return (1);
 485         }
 486
 487         return (0);
 488 }
 489
 490 static int
 491 tap_clone_match(struct if_clone *ifc, const char *name)
 492 {
 493         int tunflags;
 494
 495         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 496                 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
 497                         return (1);
 498         }
 499
 500         return (0);
 501 }
 502
 503 static int
 504 vmnet_clone_match(struct if_clone *ifc, const char *name)
 505 {
 506         int tunflags;
 507
 508         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 509                 if ((tunflags & TUN_VMNET) != 0)
 510                         return (1);
 511         }
 512
 513         return (0);
 514 }
 515
 516 static int
 517 tun_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
 518 {
 519         struct tuntap_driver *drv;
 520         struct cdev *dev;
 521         int err, i, tunflags, unit;
 522
 523         tunflags = 0;
 524         /* The name here tells us exactly what we're creating */
 525         err = tuntap_name2info(name, &unit, &tunflags);
 526         if (err != 0)
 527                 return (err);
 528
 529         drv = tuntap_driver_from_flags(tunflags);
 530         if (drv == NULL)
 531                 return (ENXIO);
 532
 533         if (unit != -1) {
 534                 /* If this unit number is still available that's okay. */
 535                 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
 536                         return (EEXIST);
 537         } else {
 538                 unit = alloc_unr(drv->unrhdr);
 539         }
 540
 541         snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
 542
 543         /* find any existing device, or allocate new unit number */
 544         dev = NULL;
 545         i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
 546         /* No preexisting struct cdev *, create one */
 547         if (i != 0)
 548                 i = tun_create_device(drv, unit, NULL, &dev, name);
 549         if (i == 0)
 550                 tuncreate(dev);
 551
 552         return (i);
 553 }
 554
 555 static void
 556 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
 557     struct cdev **dev)
 558 {
 559         char devname[SPECNAMELEN + 1];
 560         struct tuntap_driver *drv;
 561         int append_unit, i, u, tunflags;
 562         bool mayclone;
 563
 564         if (*dev != NULL)
 565                 return;
 566
 567         tunflags = 0;
 568         CURVNET_SET(CRED_TO_VNET(cred));
 569         if (tuntap_name2info(name, &u, &tunflags) != 0)
 570                 goto out;       /* Not recognized */
 571
 572         if (u != -1 && u > IF_MAXUNIT)
 573                 goto out;       /* Unit number too high */
 574
 575         mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
 576         if ((tunflags & TUN_L2) != 0) {
 577                 /* tap/vmnet allow user open with a sysctl */
 578                 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
 579         } else {
 580                 mayclone = mayclone && tundclone;
 581         }
 582
 583         /*
 584          * If tun cloning is enabled, only the superuser can create an
 585          * interface.
 586          */
 587         if (!mayclone)
 588                 goto out;
 589
 590         if (u == -1)
 591                 append_unit = 1;
 592         else
 593                 append_unit = 0;
 594
 595         drv = tuntap_driver_from_flags(tunflags);
 596         if (drv == NULL)
 597                 goto out;
 598
 599         /* find any existing device, or allocate new unit number */
 600         i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
 601         if (i) {
 602                 if (append_unit) {
 603                         namelen = snprintf(devname, sizeof(devname), "%s%d",
 604                             name, u);
 605                         name = devname;
 606                 }
 607
 608                 i = tun_create_device(drv, u, cred, dev, name);
 609         }
 610         if (i == 0)
 611                 if_clone_create(name, namelen, NULL);
 612 out:
 613         CURVNET_RESTORE();
 614 }
 615
 616 static void
 617 tun_destroy(struct tuntap_softc *tp)
 618 {
 619
 620         TUN_LOCK(tp);
 621         tp->tun_flags |= TUN_DYING;
 622         if (tp->tun_busy != 0)
 623                 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
 624         else
 625                 TUN_UNLOCK(tp);
 626
 627         CURVNET_SET(TUN2IFP(tp)->if_vnet);
 628
 629         /* destroy_dev will take care of any alias. */
 630         destroy_dev(tp->tun_dev);
 631         seldrain(&tp->tun_rsel);
 632         knlist_clear(&tp->tun_rsel.si_note, 0);
 633         knlist_destroy(&tp->tun_rsel.si_note);
 634         if ((tp->tun_flags & TUN_L2) != 0) {
 635                 ether_ifdetach(TUN2IFP(tp));
 636         } else {
 637                 bpfdetach(TUN2IFP(tp));
 638                 if_detach(TUN2IFP(tp));
 639         }
 640         sx_xlock(&tun_ioctl_sx);
 641         TUN2IFP(tp)->if_softc = NULL;
 642         sx_xunlock(&tun_ioctl_sx);
 643         free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
 644         if_free(TUN2IFP(tp));
 645         mtx_destroy(&tp->tun_mtx);
 646         cv_destroy(&tp->tun_cv);
 647         free(tp, M_TUN);
 648         CURVNET_RESTORE();
 649 }
 650
 651 static int
 652 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp)
 653 {
 654         struct tuntap_softc *tp = ifp->if_softc;
 655
 656         mtx_lock(&tunmtx);
 657         TAILQ_REMOVE(&tunhead, tp, tun_list);
 658         mtx_unlock(&tunmtx);
 659         tun_destroy(tp);
 660
 661         return (0);
 662 }
 663
 664 static void
 665 vnet_tun_init(const void *unused __unused)
 666 {
 667         struct tuntap_driver *drv;
 668         struct tuntap_driver_cloner *drvc;
 669         int i;
 670
 671         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 672                 drv = &tuntap_drivers[i];
 673                 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
 674
 675                 drvc->drv = drv;
 676                 drvc->cloner = if_clone_advanced(drv->cdevsw.d_name, 0,
 677                     drv->clone_match_fn, drv->clone_create_fn,
 678                     drv->clone_destroy_fn);
 679                 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
 680         };
 681 }
 682 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 683                 vnet_tun_init, NULL);
 684
 685 static void
 686 vnet_tun_uninit(const void *unused __unused)
 687 {
 688         struct tuntap_driver_cloner *drvc;
 689
 690         while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
 691                 drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
 692                 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
 693
 694                 if_clone_detach(drvc->cloner);
 695                 free(drvc, M_TUN);
 696         }
 697 }
 698 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 699     vnet_tun_uninit, NULL);
 700
 701 static void
 702 tun_uninit(const void *unused __unused)
 703 {
 704         struct tuntap_driver *drv;
 705         struct tuntap_softc *tp;
 706         int i;
 707
 708         EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
 709         EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
 710         drain_dev_clone_events();
 711
 712         mtx_lock(&tunmtx);
 713         while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
 714                 TAILQ_REMOVE(&tunhead, tp, tun_list);
 715                 mtx_unlock(&tunmtx);
 716                 tun_destroy(tp);
 717                 mtx_lock(&tunmtx);
 718         }
 719         mtx_unlock(&tunmtx);
 720         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 721                 drv = &tuntap_drivers[i];
 722                 delete_unrhdr(drv->unrhdr);
 723                 clone_cleanup(&drv->clones);
 724         }
 725         mtx_destroy(&tunmtx);
 726 }
 727 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
 728
 729 static struct tuntap_driver *
 730 tuntap_driver_from_ifnet(const struct ifnet *ifp)
 731 {
 732         struct tuntap_driver *drv;
 733         int i;
 734
 735         if (ifp == NULL)
 736                 return (NULL);
 737
 738         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 739                 drv = &tuntap_drivers[i];
 740                 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
 741                         return (drv);
 742         }
 743
 744         return (NULL);
 745 }
 746
 747 static int
 748 tuntapmodevent(module_t mod, int type, void *data)
 749 {
 750         struct tuntap_driver *drv;
 751         int i;
 752
 753         switch (type) {
 754         case MOD_LOAD:
 755                 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
 756                 for (i = 0; i < nitems(tuntap_drivers); ++i) {
 757                         drv = &tuntap_drivers[i];
 758                         clone_setup(&drv->clones);
 759                         drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
 760                 }
 761                 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
 762                    tunrename, 0, 1000);
 763                 if (arrival_tag == NULL)
 764                         return (ENOMEM);
 765                 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
 766                 if (clone_tag == NULL)
 767                         return (ENOMEM);
 768                 break;
 769         case MOD_UNLOAD:
 770                 /* See tun_uninit, so it's done after the vnet_sysuninit() */
 771                 break;
 772         default:
 773                 return EOPNOTSUPP;
 774         }
 775         return 0;
 776 }
 777
 778 static moduledata_t tuntap_mod = {
 779         "if_tuntap",
 780         tuntapmodevent,
 781         0
 782 };
 783
 784 /* We'll only ever have these two, so no need for a macro. */
 785 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
 786 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
 787
 788 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 789 MODULE_VERSION(if_tuntap, 1);
 790 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 791 MODULE_VERSION(if_tun, 1);
 792 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 793 MODULE_VERSION(if_tap, 1);
 794
 795 static int
 796 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
 797     struct cdev **dev, const char *name)
 798 {
 799         struct make_dev_args args;
 800         struct tuntap_softc *tp;
 801         int error;
 802
 803         tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
 804         mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
 805         cv_init(&tp->tun_cv, "tun_condvar");
 806         tp->tun_flags = drv->ident_flags;
 807         tp->tun_drv = drv;
 808
 809         make_dev_args_init(&args);
 810         if (cr != NULL)
 811                 args.mda_flags = MAKEDEV_REF;
 812         args.mda_devsw = &drv->cdevsw;
 813         args.mda_cr = cr;
 814         args.mda_uid = UID_UUCP;
 815         args.mda_gid = GID_DIALER;
 816         args.mda_mode = 0600;
 817         args.mda_unit = unit;
 818         args.mda_si_drv1 = tp;
 819         error = make_dev_s(&args, dev, "%s", name);
 820         if (error != 0) {
 821                 free(tp, M_TUN);
 822                 return (error);
 823         }
 824
 825         KASSERT((*dev)->si_drv1 != NULL,
 826             ("Failed to set si_drv1 at %s creation", name));
 827         tp->tun_dev = *dev;
 828         knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
 829         mtx_lock(&tunmtx);
 830         TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
 831         mtx_unlock(&tunmtx);
 832         return (0);
 833 }
 834
 835 static void
 836 tunstart(struct ifnet *ifp)
 837 {
 838         struct tuntap_softc *tp = ifp->if_softc;
 839         struct mbuf *m;
 840
 841         TUNDEBUG(ifp, "starting\n");
 842         if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
 843                 IFQ_LOCK(&ifp->if_snd);
 844                 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
 845                 if (m == NULL) {
 846                         IFQ_UNLOCK(&ifp->if_snd);
 847                         return;
 848                 }
 849                 IFQ_UNLOCK(&ifp->if_snd);
 850         }
 851
 852         TUN_LOCK(tp);
 853         if (tp->tun_flags & TUN_RWAIT) {
 854                 tp->tun_flags &= ~TUN_RWAIT;
 855                 wakeup(tp);
 856         }
 857         selwakeuppri(&tp->tun_rsel, PZERO + 1);
 858         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 859         if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
 860                 TUN_UNLOCK(tp);
 861                 pgsigio(&tp->tun_sigio, SIGIO, 0);
 862         } else
 863                 TUN_UNLOCK(tp);
 864 }
 865
 866 /*
 867  * tunstart_l2
 868  *
 869  * queue packets from higher level ready to put out
 870  */
 871 static void
 872 tunstart_l2(struct ifnet *ifp)
 873 {
 874         struct tuntap_softc     *tp = ifp->if_softc;
 875
 876         TUNDEBUG(ifp, "starting\n");
 877
 878         /*
 879          * do not junk pending output if we are in VMnet mode.
 880          * XXX: can this do any harm because of queue overflow?
 881          */
 882
 883         TUN_LOCK(tp);
 884         if (((tp->tun_flags & TUN_VMNET) == 0) &&
 885             ((tp->tun_flags & TUN_READY) != TUN_READY)) {
 886                 struct mbuf *m;
 887
 888                 /* Unlocked read. */
 889                 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
 890
 891                 for (;;) {
 892                         IF_DEQUEUE(&ifp->if_snd, m);
 893                         if (m != NULL) {
 894                                 m_freem(m);
 895                                 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 896                         } else
 897                                 break;
 898                 }
 899                 TUN_UNLOCK(tp);
 900
 901                 return;
 902         }
 903
 904         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
 905
 906         if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
 907                 if (tp->tun_flags & TUN_RWAIT) {
 908                         tp->tun_flags &= ~TUN_RWAIT;
 909                         wakeup(tp);
 910                 }
 911
 912                 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
 913                         TUN_UNLOCK(tp);
 914                         pgsigio(&tp->tun_sigio, SIGIO, 0);
 915                         TUN_LOCK(tp);
 916                 }
 917
 918                 selwakeuppri(&tp->tun_rsel, PZERO+1);
 919                 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 920                 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
 921         }
 922
 923         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
 924         TUN_UNLOCK(tp);
 925 } /* tunstart_l2 */
 926
 927 /* XXX: should return an error code so it can fail. */
 928 static void
 929 tuncreate(struct cdev *dev)
 930 {
 931         struct tuntap_driver *drv;
 932         struct tuntap_softc *tp;
 933         struct ifnet *ifp;
 934         struct ether_addr eaddr;
 935         int iflags;
 936         u_char type;
 937
 938         tp = dev->si_drv1;
 939         KASSERT(tp != NULL,
 940             ("si_drv1 should have been initialized at creation"));
 941
 942         drv = tp->tun_drv;
 943         iflags = IFF_MULTICAST;
 944         if ((tp->tun_flags & TUN_L2) != 0) {
 945                 type = IFT_ETHER;
 946                 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
 947         } else {
 948                 type = IFT_PPP;
 949                 iflags |= IFF_POINTOPOINT;
 950         }
 951         ifp = tp->tun_ifp = if_alloc(type);
 952         if (ifp == NULL)
 953                 panic("%s%d: failed to if_alloc() interface.\n",
 954                     drv->cdevsw.d_name, dev2unit(dev));
 955         ifp->if_softc = tp;
 956         if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
 957         ifp->if_ioctl = tunifioctl;
 958         ifp->if_flags = iflags;
 959         IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
 960         ifp->if_capabilities |= IFCAP_LINKSTATE;
 961         ifp->if_capenable |= IFCAP_LINKSTATE;
 962
 963         if ((tp->tun_flags & TUN_L2) != 0) {
 964                 ifp->if_init = tunifinit;
 965                 ifp->if_start = tunstart_l2;
 966
 967                 ether_gen_addr(ifp, &eaddr);
 968                 ether_ifattach(ifp, eaddr.octet);
 969         } else {
 970                 ifp->if_mtu = TUNMTU;
 971                 ifp->if_start = tunstart;
 972                 ifp->if_output = tunoutput;
 973
 974                 ifp->if_snd.ifq_drv_maxlen = 0;
 975                 IFQ_SET_READY(&ifp->if_snd);
 976
 977                 if_attach(ifp);
 978                 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
 979         }
 980
 981         TUN_LOCK(tp);
 982         tp->tun_flags |= TUN_INITED;
 983         TUN_UNLOCK(tp);
 984
 985         TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
 986             ifp->if_xname, dev2unit(dev));
 987 }
 988
 989 static void
 990 tunrename(void *arg __unused, struct ifnet *ifp)
 991 {
 992         struct tuntap_softc *tp;
 993         int error;
 994
 995         if ((ifp->if_flags & IFF_RENAMING) == 0)
 996                 return;
 997
 998         if (tuntap_driver_from_ifnet(ifp) == NULL)
 999                 return;
1000
1001         /*
1002          * We need to grab the ioctl sx long enough to make sure the softc is
1003          * still there.  If it is, we can safely try to busy the tun device.
1004          * The busy may fail if the device is currently dying, in which case
1005          * we do nothing.  If it doesn't fail, the busy count stops the device
1006          * from dying until we've created the alias (that will then be
1007          * subsequently destroyed).
1008          */
1009         sx_xlock(&tun_ioctl_sx);
1010         tp = ifp->if_softc;
1011         if (tp == NULL) {
1012                 sx_xunlock(&tun_ioctl_sx);
1013                 return;
1014         }
1015         error = tun_busy(tp);
1016         sx_xunlock(&tun_ioctl_sx);
1017         if (error != 0)
1018                 return;
1019         if (tp->tun_alias != NULL) {
1020                 destroy_dev(tp->tun_alias);
1021                 tp->tun_alias = NULL;
1022         }
1023
1024         if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1025                 goto out;
1026
1027         /*
1028          * Failure's ok, aliases are created on a best effort basis.  If a
1029          * tun user/consumer decides to rename the interface to conflict with
1030          * another device (non-ifnet) on the system, we will assume they know
1031          * what they are doing.  make_dev_alias_p won't touch tun_alias on
1032          * failure, so we use it but ignore the return value.
1033          */
1034         make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1035             ifp->if_xname);
1036 out:
1037         tun_unbusy(tp);
1038 }
1039
1040 static int
1041 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1042 {
1043         struct ifnet    *ifp;
1044         struct tuntap_softc *tp;
1045         int error, tunflags;
1046
1047         tunflags = 0;
1048         CURVNET_SET(TD_TO_VNET(td));
1049         error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1050         if (error != 0) {
1051                 CURVNET_RESTORE();
1052                 return (error); /* Shouldn't happen */
1053         }
1054
1055         tp = dev->si_drv1;
1056         KASSERT(tp != NULL,
1057             ("si_drv1 should have been initialized at creation"));
1058
1059         TUN_LOCK(tp);
1060         if ((tp->tun_flags & TUN_INITED) == 0) {
1061                 TUN_UNLOCK(tp);
1062                 CURVNET_RESTORE();
1063                 return (ENXIO);
1064         }
1065         if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1066                 TUN_UNLOCK(tp);
1067                 CURVNET_RESTORE();
1068                 return (EBUSY);
1069         }
1070
1071         error = tun_busy_locked(tp);
1072         KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1073         ifp = TUN2IFP(tp);
1074
1075         if ((tp->tun_flags & TUN_L2) != 0) {
1076                 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1077                     sizeof(tp->tun_ether.octet));
1078
1079                 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1080                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1081
1082                 if (tapuponopen)
1083                         ifp->if_flags |= IFF_UP;
1084         }
1085
1086         tp->tun_pid = td->td_proc->p_pid;
1087         tp->tun_flags |= TUN_OPEN;
1088
1089         if_link_state_change(ifp, LINK_STATE_UP);
1090         TUNDEBUG(ifp, "open\n");
1091         TUN_UNLOCK(tp);
1092
1093         /*
1094          * This can fail with either ENOENT or EBUSY.  This is in the middle of
1095          * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1096          * the only cdevpriv dtor being set will be tundtor and the softc being
1097          * passed is constant for a given cdev.  We ignore the possible error
1098          * because of this as either "unlikely" or "not actually a problem."
1099          */
1100         (void)devfs_set_cdevpriv(tp, tundtor);
1101         CURVNET_RESTORE();
1102         return (0);
1103 }
1104
1105 /*
1106  * tundtor - tear down the device - mark i/f down & delete
1107  * routing info
1108  */
1109 static void
1110 tundtor(void *data)
1111 {
1112         struct proc *p;
1113         struct tuntap_softc *tp;
1114         struct ifnet *ifp;
1115         bool l2tun;
1116
1117         tp = data;
1118         p = curproc;
1119         ifp = TUN2IFP(tp);
1120
1121         TUN_LOCK(tp);
1122
1123         /*
1124          * Realistically, we can't be obstinate here.  This only means that the
1125          * tuntap device was closed out of order, and the last closer wasn't the
1126          * controller.  These are still good to know about, though, as software
1127          * should avoid multiple processes with a tuntap device open and
1128          * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1129          * parent).
1130          */
1131         if (p->p_pid != tp->tun_pid) {
1132                 log(LOG_INFO,
1133                     "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1134                     p->p_pid, p->p_comm, tp->tun_dev->si_name);
1135         }
1136
1137         /*
1138          * junk all pending output
1139          */
1140         CURVNET_SET(ifp->if_vnet);
1141
1142         l2tun = false;
1143         if ((tp->tun_flags & TUN_L2) != 0) {
1144                 l2tun = true;
1145                 IF_DRAIN(&ifp->if_snd);
1146         } else {
1147                 IFQ_PURGE(&ifp->if_snd);
1148         }
1149
1150         /* For vmnet, we won't do most of the address/route bits */
1151         if ((tp->tun_flags & TUN_VMNET) != 0 ||
1152             (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1153                 goto out;
1154
1155         if (ifp->if_flags & IFF_UP) {
1156                 TUN_UNLOCK(tp);
1157                 if_down(ifp);
1158                 TUN_LOCK(tp);
1159         }
1160
1161         /* Delete all addresses and routes which reference this interface. */
1162         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1163                 struct ifaddr *ifa;
1164
1165                 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1166                 TUN_UNLOCK(tp);
1167                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1168                         /* deal w/IPv4 PtP destination; unlocked read */
1169                         if (!l2tun && ifa->ifa_addr->sa_family == AF_INET) {
1170                                 rtinit(ifa, (int)RTM_DELETE,
1171                                     tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
1172                         } else {
1173                                 rtinit(ifa, (int)RTM_DELETE, 0);
1174                         }
1175                 }
1176                 if_purgeaddrs(ifp);
1177                 TUN_LOCK(tp);
1178         }
1179
1180 out:
1181         if_link_state_change(ifp, LINK_STATE_DOWN);
1182         CURVNET_RESTORE();
1183
1184         funsetown(&tp->tun_sigio);
1185         selwakeuppri(&tp->tun_rsel, PZERO + 1);
1186         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1187         TUNDEBUG (ifp, "closed\n");
1188         tp->tun_flags &= ~TUN_OPEN;
1189         tp->tun_pid = 0;
1190         tun_vnethdr_set(ifp, 0);
1191
1192         tun_unbusy_locked(tp);
1193         TUN_UNLOCK(tp);
1194 }
1195
1196 static void
1197 tuninit(struct ifnet *ifp)
1198 {
1199         struct tuntap_softc *tp = ifp->if_softc;
1200 #ifdef INET
1201         struct epoch_tracker et;
1202         struct ifaddr *ifa;
1203 #endif
1204
1205         TUNDEBUG(ifp, "tuninit\n");
1206
1207         TUN_LOCK(tp);
1208         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1209         if ((tp->tun_flags & TUN_L2) == 0) {
1210                 ifp->if_flags |= IFF_UP;
1211                 getmicrotime(&ifp->if_lastchange);
1212 #ifdef INET
1213                 NET_EPOCH_ENTER(et);
1214                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1215                         if (ifa->ifa_addr->sa_family == AF_INET) {
1216                                 struct sockaddr_in *si;
1217
1218                                 si = (struct sockaddr_in *)ifa->ifa_dstaddr;
1219                                 if (si && si->sin_addr.s_addr) {
1220                                         tp->tun_flags |= TUN_DSTADDR;
1221                                         break;
1222                                 }
1223                         }
1224                 }
1225                 NET_EPOCH_EXIT(et);
1226 #endif
1227                 TUN_UNLOCK(tp);
1228         } else {
1229                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1230                 TUN_UNLOCK(tp);
1231                 /* attempt to start output */
1232                 tunstart_l2(ifp);
1233         }
1234
1235 }
1236
1237 /*
1238  * Used only for l2 tunnel.
1239  */
1240 static void
1241 tunifinit(void *xtp)
1242 {
1243         struct tuntap_softc *tp;
1244
1245         tp = (struct tuntap_softc *)xtp;
1246         tuninit(tp->tun_ifp);
1247 }
1248
1249 /*
1250  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1251  * current value of ifp->if_capenable.
1252  */
1253 static void
1254 tun_caps_changed(struct ifnet *ifp)
1255 {
1256         uint64_t hwassist = 0;
1257
1258         TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1259         if (ifp->if_capenable & IFCAP_TXCSUM)
1260                 hwassist |= CSUM_TCP | CSUM_UDP;
1261         if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1262                 hwassist |= CSUM_TCP_IPV6
1263                     | CSUM_UDP_IPV6;
1264         if (ifp->if_capenable & IFCAP_TSO4)
1265                 hwassist |= CSUM_IP_TSO;
1266         if (ifp->if_capenable & IFCAP_TSO6)
1267                 hwassist |= CSUM_IP6_TSO;
1268         ifp->if_hwassist = hwassist;
1269 }
1270
1271 /*
1272  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1273  * if_capabilities and if_capenable as needed.
1274  */
1275 static void
1276 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1277 {
1278         struct tuntap_softc *tp = ifp->if_softc;
1279
1280         TUN_LOCK_ASSERT(tp);
1281
1282         if (tp->tun_vhdrlen == vhdrlen)
1283                 return;
1284
1285         /*
1286          * Update if_capabilities to reflect the
1287          * functionalities offered by the virtio-net
1288          * header.
1289          */
1290         if (vhdrlen != 0)
1291                 ifp->if_capabilities |=
1292                         TAP_VNET_HDR_CAPS;
1293         else
1294                 ifp->if_capabilities &=
1295                         ~TAP_VNET_HDR_CAPS;
1296         /*
1297          * Disable any capabilities that we don't
1298          * support anymore.
1299          */
1300         ifp->if_capenable &= ifp->if_capabilities;
1301         tun_caps_changed(ifp);
1302         tp->tun_vhdrlen = vhdrlen;
1303
1304         TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1305             vhdrlen, ifp->if_capabilities);
1306 }
1307
1308 /*
1309  * Process an ioctl request.
1310  */
1311 static int
1312 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1313 {
1314         struct ifreq *ifr = (struct ifreq *)data;
1315         struct tuntap_softc *tp;
1316         struct ifstat *ifs;
1317         struct ifmediareq       *ifmr;
1318         int             dummy, error = 0;
1319         bool            l2tun;
1320
1321         ifmr = NULL;
1322         sx_xlock(&tun_ioctl_sx);
1323         tp = ifp->if_softc;
1324         if (tp == NULL) {
1325                 error = ENXIO;
1326                 goto bad;
1327         }
1328         l2tun = (tp->tun_flags & TUN_L2) != 0;
1329         switch(cmd) {
1330         case SIOCGIFSTATUS:
1331                 ifs = (struct ifstat *)data;
1332                 TUN_LOCK(tp);
1333                 if (tp->tun_pid)
1334                         snprintf(ifs->ascii, sizeof(ifs->ascii),
1335                             "\tOpened by PID %d\n", tp->tun_pid);
1336                 else
1337                         ifs->ascii[0] = '\0';
1338                 TUN_UNLOCK(tp);
1339                 break;
1340         case SIOCSIFADDR:
1341                 if (l2tun)
1342                         error = ether_ioctl(ifp, cmd, data);
1343                 else
1344                         tuninit(ifp);
1345                 if (error == 0)
1346                     TUNDEBUG(ifp, "address set\n");
1347                 break;
1348         case SIOCSIFMTU:
1349                 ifp->if_mtu = ifr->ifr_mtu;
1350                 TUNDEBUG(ifp, "mtu set\n");
1351                 break;
1352         case SIOCSIFFLAGS:
1353         case SIOCADDMULTI:
1354         case SIOCDELMULTI:
1355                 break;
1356         case SIOCGIFMEDIA:
1357                 if (!l2tun) {
1358                         error = EINVAL;
1359                         break;
1360                 }
1361
1362                 ifmr = (struct ifmediareq *)data;
1363                 dummy = ifmr->ifm_count;
1364                 ifmr->ifm_count = 1;
1365                 ifmr->ifm_status = IFM_AVALID;
1366                 ifmr->ifm_active = IFM_ETHER;
1367                 if (tp->tun_flags & TUN_OPEN)
1368                         ifmr->ifm_status |= IFM_ACTIVE;
1369                 ifmr->ifm_current = ifmr->ifm_active;
1370                 if (dummy >= 1) {
1371                         int media = IFM_ETHER;
1372                         error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1373                 }
1374                 break;
1375         case SIOCSIFCAP:
1376                 TUN_LOCK(tp);
1377                 ifp->if_capenable = ifr->ifr_reqcap;
1378                 tun_caps_changed(ifp);
1379                 TUN_UNLOCK(tp);
1380                 VLAN_CAPABILITIES(ifp);
1381                 break;
1382         default:
1383                 if (l2tun) {
1384                         error = ether_ioctl(ifp, cmd, data);
1385                 } else {
1386                         error = EINVAL;
1387                 }
1388         }
1389 bad:
1390         sx_xunlock(&tun_ioctl_sx);
1391         return (error);
1392 }
1393
1394 /*
1395  * tunoutput - queue packets from higher level ready to put out.
1396  */
1397 static int
1398 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1399     struct route *ro)
1400 {
1401         struct tuntap_softc *tp = ifp->if_softc;
1402         u_short cached_tun_flags;
1403         int error;
1404         u_int32_t af;
1405
1406         TUNDEBUG (ifp, "tunoutput\n");
1407
1408 #ifdef MAC
1409         error = mac_ifnet_check_transmit(ifp, m0);
1410         if (error) {
1411                 m_freem(m0);
1412                 return (error);
1413         }
1414 #endif
1415
1416         /* Could be unlocked read? */
1417         TUN_LOCK(tp);
1418         cached_tun_flags = tp->tun_flags;
1419         TUN_UNLOCK(tp);
1420         if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1421                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1422                 m_freem (m0);
1423                 return (EHOSTDOWN);
1424         }
1425
1426         if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1427                 m_freem (m0);
1428                 return (EHOSTDOWN);
1429         }
1430
1431         /* BPF writes need to be handled specially. */
1432         if (dst->sa_family == AF_UNSPEC)
1433                 bcopy(dst->sa_data, &af, sizeof(af));
1434         else
1435                 af = dst->sa_family;
1436
1437         if (bpf_peers_present(ifp->if_bpf))
1438                 bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1439
1440         /* prepend sockaddr? this may abort if the mbuf allocation fails */
1441         if (cached_tun_flags & TUN_LMODE) {
1442                 /* allocate space for sockaddr */
1443                 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1444
1445                 /* if allocation failed drop packet */
1446                 if (m0 == NULL) {
1447                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1448                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1449                         return (ENOBUFS);
1450                 } else {
1451                         bcopy(dst, m0->m_data, dst->sa_len);
1452                 }
1453         }
1454
1455         if (cached_tun_flags & TUN_IFHEAD) {
1456                 /* Prepend the address family */
1457                 M_PREPEND(m0, 4, M_NOWAIT);
1458
1459                 /* if allocation failed drop packet */
1460                 if (m0 == NULL) {
1461                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1462                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1463                         return (ENOBUFS);
1464                 } else
1465                         *(u_int32_t *)m0->m_data = htonl(af);
1466         } else {
1467 #ifdef INET
1468                 if (af != AF_INET)
1469 #endif
1470                 {
1471                         m_freem(m0);
1472                         return (EAFNOSUPPORT);
1473                 }
1474         }
1475
1476         error = (ifp->if_transmit)(ifp, m0);
1477         if (error)
1478                 return (ENOBUFS);
1479         if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1480         return (0);
1481 }
1482
1483 /*
1484  * the cdevsw interface is now pretty minimal.
1485  */
1486 static  int
1487 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1488     struct thread *td)
1489 {
1490         struct ifreq ifr, *ifrp;
1491         struct tuntap_softc *tp = dev->si_drv1;
1492         struct ifnet *ifp = TUN2IFP(tp);
1493         struct tuninfo *tunp;
1494         int error, iflags, ival;
1495         bool    l2tun;
1496
1497         l2tun = (tp->tun_flags & TUN_L2) != 0;
1498         if (l2tun) {
1499                 /* tap specific ioctls */
1500                 switch(cmd) {
1501                 /* VMware/VMnet port ioctl's */
1502 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1503     defined(COMPAT_FREEBSD4)
1504                 case _IO('V', 0):
1505                         ival = IOCPARM_IVAL(data);
1506                         data = (caddr_t)&ival;
1507                         /* FALLTHROUGH */
1508 #endif
1509                 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1510                         iflags = *(int *)data;
1511                         iflags &= TUN_VMIO_FLAG_MASK;
1512                         iflags &= ~IFF_CANTCHANGE;
1513                         iflags |= IFF_UP;
1514
1515                         TUN_LOCK(tp);
1516                         ifp->if_flags = iflags |
1517                             (ifp->if_flags & IFF_CANTCHANGE);
1518                         TUN_UNLOCK(tp);
1519
1520                         return (0);
1521                 case SIOCGIFADDR:       /* get MAC address of the remote side */
1522                         TUN_LOCK(tp);
1523                         bcopy(&tp->tun_ether.octet, data,
1524                             sizeof(tp->tun_ether.octet));
1525                         TUN_UNLOCK(tp);
1526
1527                         return (0);
1528                 case SIOCSIFADDR:       /* set MAC address of the remote side */
1529                         TUN_LOCK(tp);
1530                         bcopy(data, &tp->tun_ether.octet,
1531                             sizeof(tp->tun_ether.octet));
1532                         TUN_UNLOCK(tp);
1533
1534                         return (0);
1535                 case TAPSVNETHDR:
1536                         ival = *(int *)data;
1537                         if (ival != 0 &&
1538                             ival != sizeof(struct virtio_net_hdr) &&
1539                             ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1540                                 return (EINVAL);
1541                         }
1542                         TUN_LOCK(tp);
1543                         tun_vnethdr_set(ifp, ival);
1544                         TUN_UNLOCK(tp);
1545
1546                         return (0);
1547                 case TAPGVNETHDR:
1548                         TUN_LOCK(tp);
1549                         *(int *)data = tp->tun_vhdrlen;
1550                         TUN_UNLOCK(tp);
1551
1552                         return (0);
1553                 }
1554
1555                 /* Fall through to the common ioctls if unhandled */
1556         } else {
1557                 switch (cmd) {
1558                 case TUNSLMODE:
1559                         TUN_LOCK(tp);
1560                         if (*(int *)data) {
1561                                 tp->tun_flags |= TUN_LMODE;
1562                                 tp->tun_flags &= ~TUN_IFHEAD;
1563                         } else
1564                                 tp->tun_flags &= ~TUN_LMODE;
1565                         TUN_UNLOCK(tp);
1566
1567                         return (0);
1568                 case TUNSIFHEAD:
1569                         TUN_LOCK(tp);
1570                         if (*(int *)data) {
1571                                 tp->tun_flags |= TUN_IFHEAD;
1572                                 tp->tun_flags &= ~TUN_LMODE;
1573                         } else
1574                                 tp->tun_flags &= ~TUN_IFHEAD;
1575                         TUN_UNLOCK(tp);
1576
1577                         return (0);
1578                 case TUNGIFHEAD:
1579                         TUN_LOCK(tp);
1580                         *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1581                         TUN_UNLOCK(tp);
1582
1583                         return (0);
1584                 case TUNSIFMODE:
1585                         /* deny this if UP */
1586                         if (TUN2IFP(tp)->if_flags & IFF_UP)
1587                                 return (EBUSY);
1588
1589                         switch (*(int *)data & ~IFF_MULTICAST) {
1590                         case IFF_POINTOPOINT:
1591                         case IFF_BROADCAST:
1592                                 TUN_LOCK(tp);
1593                                 TUN2IFP(tp)->if_flags &=
1594                                     ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1595                                 TUN2IFP(tp)->if_flags |= *(int *)data;
1596                                 TUN_UNLOCK(tp);
1597
1598                                 break;
1599                         default:
1600                                 return (EINVAL);
1601                         }
1602
1603                         return (0);
1604                 case TUNSIFPID:
1605                         TUN_LOCK(tp);
1606                         tp->tun_pid = curthread->td_proc->p_pid;
1607                         TUN_UNLOCK(tp);
1608
1609                         return (0);
1610                 }
1611                 /* Fall through to the common ioctls if unhandled */
1612         }
1613
1614         switch (cmd) {
1615         case TUNGIFNAME:
1616                 ifrp = (struct ifreq *)data;
1617                 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1618
1619                 return (0);
1620         case TUNSIFINFO:
1621                 tunp = (struct tuninfo *)data;
1622                 if (TUN2IFP(tp)->if_type != tunp->type)
1623                         return (EPROTOTYPE);
1624                 TUN_LOCK(tp);
1625                 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1626                         strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1627                         ifr.ifr_mtu = tunp->mtu;
1628                         CURVNET_SET(TUN2IFP(tp)->if_vnet);
1629                         error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1630                             (caddr_t)&ifr, td);
1631                         CURVNET_RESTORE();
1632                         if (error) {
1633                                 TUN_UNLOCK(tp);
1634                                 return (error);
1635                         }
1636                 }
1637                 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1638                 TUN_UNLOCK(tp);
1639                 break;
1640         case TUNGIFINFO:
1641                 tunp = (struct tuninfo *)data;
1642                 TUN_LOCK(tp);
1643                 tunp->mtu = TUN2IFP(tp)->if_mtu;
1644                 tunp->type = TUN2IFP(tp)->if_type;
1645                 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1646                 TUN_UNLOCK(tp);
1647                 break;
1648         case TUNSDEBUG:
1649                 tundebug = *(int *)data;
1650                 break;
1651         case TUNGDEBUG:
1652                 *(int *)data = tundebug;
1653                 break;
1654         case FIONBIO:
1655                 break;
1656         case FIOASYNC:
1657                 TUN_LOCK(tp);
1658                 if (*(int *)data)
1659                         tp->tun_flags |= TUN_ASYNC;
1660                 else
1661                         tp->tun_flags &= ~TUN_ASYNC;
1662                 TUN_UNLOCK(tp);
1663                 break;
1664         case FIONREAD:
1665                 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1666                         struct mbuf *mb;
1667                         IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1668                         IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1669                         for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1670                                 *(int *)data += mb->m_len;
1671                         IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1672                 } else
1673                         *(int *)data = 0;
1674                 break;
1675         case FIOSETOWN:
1676                 return (fsetown(*(int *)data, &tp->tun_sigio));
1677
1678         case FIOGETOWN:
1679                 *(int *)data = fgetown(&tp->tun_sigio);
1680                 return (0);
1681
1682         /* This is deprecated, FIOSETOWN should be used instead. */
1683         case TIOCSPGRP:
1684                 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1685
1686         /* This is deprecated, FIOGETOWN should be used instead. */
1687         case TIOCGPGRP:
1688                 *(int *)data = -fgetown(&tp->tun_sigio);
1689                 return (0);
1690
1691         default:
1692                 return (ENOTTY);
1693         }
1694         return (0);
1695 }
1696
1697 /*
1698  * The cdevsw read interface - reads a packet at a time, or at
1699  * least as much of a packet as can be read.
1700  */
1701 static  int
1702 tunread(struct cdev *dev, struct uio *uio, int flag)
1703 {
1704         struct tuntap_softc *tp = dev->si_drv1;
1705         struct ifnet    *ifp = TUN2IFP(tp);
1706         struct mbuf     *m;
1707         size_t          len;
1708         int             error = 0;
1709
1710         TUNDEBUG (ifp, "read\n");
1711         TUN_LOCK(tp);
1712         if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1713                 TUN_UNLOCK(tp);
1714                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1715                 return (EHOSTDOWN);
1716         }
1717
1718         tp->tun_flags &= ~TUN_RWAIT;
1719
1720         for (;;) {
1721                 IFQ_DEQUEUE(&ifp->if_snd, m);
1722                 if (m != NULL)
1723                         break;
1724                 if (flag & O_NONBLOCK) {
1725                         TUN_UNLOCK(tp);
1726                         return (EWOULDBLOCK);
1727                 }
1728                 tp->tun_flags |= TUN_RWAIT;
1729                 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1730                     "tunread", 0);
1731                 if (error != 0) {
1732                         TUN_UNLOCK(tp);
1733                         return (error);
1734                 }
1735         }
1736         TUN_UNLOCK(tp);
1737
1738         if ((tp->tun_flags & TUN_L2) != 0)
1739                 BPF_MTAP(ifp, m);
1740
1741         len = min(tp->tun_vhdrlen, uio->uio_resid);
1742         if (len > 0) {
1743                 struct virtio_net_hdr_mrg_rxbuf vhdr;
1744
1745                 bzero(&vhdr, sizeof(vhdr));
1746                 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1747                         m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1748                 }
1749
1750                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1751                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1752                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1753                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1754                     vhdr.hdr.csum_offset);
1755                 error = uiomove(&vhdr, len, uio);
1756         }
1757
1758         while (m && uio->uio_resid > 0 && error == 0) {
1759                 len = min(uio->uio_resid, m->m_len);
1760                 if (len != 0)
1761                         error = uiomove(mtod(m, void *), len, uio);
1762                 m = m_free(m);
1763         }
1764
1765         if (m) {
1766                 TUNDEBUG(ifp, "Dropping mbuf\n");
1767                 m_freem(m);
1768         }
1769         return (error);
1770 }
1771
1772 static int
1773 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1774             struct virtio_net_hdr_mrg_rxbuf *vhdr)
1775 {
1776         struct epoch_tracker et;
1777         struct ether_header *eh;
1778         struct ifnet *ifp;
1779
1780         ifp = TUN2IFP(tp);
1781
1782         /*
1783          * Only pass a unicast frame to ether_input(), if it would
1784          * actually have been received by non-virtual hardware.
1785          */
1786         if (m->m_len < sizeof(struct ether_header)) {
1787                 m_freem(m);
1788                 return (0);
1789         }
1790
1791         eh = mtod(m, struct ether_header *);
1792
1793         if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1794             !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1795             bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1796                 m_freem(m);
1797                 return (0);
1798         }
1799
1800         if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1801                 m_freem(m);
1802                 return (0);
1803         }
1804
1805         /* Pass packet up to parent. */
1806         CURVNET_SET(ifp->if_vnet);
1807         NET_EPOCH_ENTER(et);
1808         (*ifp->if_input)(ifp, m);
1809         NET_EPOCH_EXIT(et);
1810         CURVNET_RESTORE();
1811         /* ibytes are counted in parent */
1812         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1813         return (0);
1814 }
1815
1816 static int
1817 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1818 {
1819         struct epoch_tracker et;
1820         struct ifnet *ifp;
1821         int family, isr;
1822
1823         ifp = TUN2IFP(tp);
1824         /* Could be unlocked read? */
1825         TUN_LOCK(tp);
1826         if (tp->tun_flags & TUN_IFHEAD) {
1827                 TUN_UNLOCK(tp);
1828                 if (m->m_len < sizeof(family) &&
1829                 (m = m_pullup(m, sizeof(family))) == NULL)
1830                         return (ENOBUFS);
1831                 family = ntohl(*mtod(m, u_int32_t *));
1832                 m_adj(m, sizeof(family));
1833         } else {
1834                 TUN_UNLOCK(tp);
1835                 family = AF_INET;
1836         }
1837
1838         BPF_MTAP2(ifp, &family, sizeof(family), m);
1839
1840         switch (family) {
1841 #ifdef INET
1842         case AF_INET:
1843                 isr = NETISR_IP;
1844                 break;
1845 #endif
1846 #ifdef INET6
1847         case AF_INET6:
1848                 isr = NETISR_IPV6;
1849                 break;
1850 #endif
1851         default:
1852                 m_freem(m);
1853                 return (EAFNOSUPPORT);
1854         }
1855         random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1856         if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1857         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1858         CURVNET_SET(ifp->if_vnet);
1859         M_SETFIB(m, ifp->if_fib);
1860         NET_EPOCH_ENTER(et);
1861         netisr_dispatch(isr, m);
1862         NET_EPOCH_EXIT(et);
1863         CURVNET_RESTORE();
1864         return (0);
1865 }
1866
1867 /*
1868  * the cdevsw write interface - an atomic write is a packet - or else!
1869  */
1870 static  int
1871 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1872 {
1873         struct virtio_net_hdr_mrg_rxbuf vhdr;
1874         struct tuntap_softc *tp;
1875         struct ifnet    *ifp;
1876         struct mbuf     *m;
1877         uint32_t        mru;
1878         int             align, vhdrlen, error;
1879         bool            l2tun;
1880
1881         tp = dev->si_drv1;
1882         ifp = TUN2IFP(tp);
1883         TUNDEBUG(ifp, "tunwrite\n");
1884         if ((ifp->if_flags & IFF_UP) != IFF_UP)
1885                 /* ignore silently */
1886                 return (0);
1887
1888         if (uio->uio_resid == 0)
1889                 return (0);
1890
1891         l2tun = (tp->tun_flags & TUN_L2) != 0;
1892         mru = l2tun ? TAPMRU : TUNMRU;
1893         vhdrlen = tp->tun_vhdrlen;
1894         align = 0;
1895         if (l2tun) {
1896                 align = ETHER_ALIGN;
1897                 mru += vhdrlen;
1898         } else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1899                 mru += sizeof(uint32_t);        /* family */
1900         if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1901                 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1902                 return (EIO);
1903         }
1904
1905         if (vhdrlen > 0) {
1906                 error = uiomove(&vhdr, vhdrlen, uio);
1907                 if (error != 0)
1908                         return (error);
1909                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1910                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1911                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1912                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1913                     vhdr.hdr.csum_offset);
1914         }
1915
1916         if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1917                 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1918                 return (ENOBUFS);
1919         }
1920
1921         m->m_pkthdr.rcvif = ifp;
1922 #ifdef MAC
1923         mac_ifnet_create_mbuf(ifp, m);
1924 #endif
1925
1926         if (l2tun)
1927                 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1928
1929         return (tunwrite_l3(tp, m));
1930 }
1931
1932 /*
1933  * tunpoll - the poll interface, this is only useful on reads
1934  * really. The write detect always returns true, write never blocks
1935  * anyway, it either accepts the packet or drops it.
1936  */
1937 static  int
1938 tunpoll(struct cdev *dev, int events, struct thread *td)
1939 {
1940         struct tuntap_softc *tp = dev->si_drv1;
1941         struct ifnet    *ifp = TUN2IFP(tp);
1942         int             revents = 0;
1943
1944         TUNDEBUG(ifp, "tunpoll\n");
1945
1946         if (events & (POLLIN | POLLRDNORM)) {
1947                 IFQ_LOCK(&ifp->if_snd);
1948                 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1949                         TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1950                         revents |= events & (POLLIN | POLLRDNORM);
1951                 } else {
1952                         TUNDEBUG(ifp, "tunpoll waiting\n");
1953                         selrecord(td, &tp->tun_rsel);
1954                 }
1955                 IFQ_UNLOCK(&ifp->if_snd);
1956         }
1957         revents |= events & (POLLOUT | POLLWRNORM);
1958
1959         return (revents);
1960 }
1961
1962 /*
1963  * tunkqfilter - support for the kevent() system call.
1964  */
1965 static int
1966 tunkqfilter(struct cdev *dev, struct knote *kn)
1967 {
1968         struct tuntap_softc     *tp = dev->si_drv1;
1969         struct ifnet    *ifp = TUN2IFP(tp);
1970
1971         switch(kn->kn_filter) {
1972         case EVFILT_READ:
1973                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1974                     ifp->if_xname, dev2unit(dev));
1975                 kn->kn_fop = &tun_read_filterops;
1976                 break;
1977
1978         case EVFILT_WRITE:
1979                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1980                     ifp->if_xname, dev2unit(dev));
1981                 kn->kn_fop = &tun_write_filterops;
1982                 break;
1983
1984         default:
1985                 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1986                     ifp->if_xname, dev2unit(dev));
1987                 return(EINVAL);
1988         }
1989
1990         kn->kn_hook = tp;
1991         knlist_add(&tp->tun_rsel.si_note, kn, 0);
1992
1993         return (0);
1994 }
1995
1996 /*
1997  * Return true of there is data in the interface queue.
1998  */
1999 static int
2000 tunkqread(struct knote *kn, long hint)
2001 {
2002         int                     ret;
2003         struct tuntap_softc     *tp = kn->kn_hook;
2004         struct cdev             *dev = tp->tun_dev;
2005         struct ifnet    *ifp = TUN2IFP(tp);
2006
2007         if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2008                 TUNDEBUG(ifp,
2009                     "%s have data in the queue.  Len = %d, minor = %#x\n",
2010                     ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2011                 ret = 1;
2012         } else {
2013                 TUNDEBUG(ifp,
2014                     "%s waiting for data, minor = %#x\n", ifp->if_xname,
2015                     dev2unit(dev));
2016                 ret = 0;
2017         }
2018
2019         return (ret);
2020 }
2021
2022 /*
2023  * Always can write, always return MTU in kn->data.
2024  */
2025 static int
2026 tunkqwrite(struct knote *kn, long hint)
2027 {
2028         struct tuntap_softc     *tp = kn->kn_hook;
2029         struct ifnet    *ifp = TUN2IFP(tp);
2030
2031         kn->kn_data = ifp->if_mtu;
2032
2033         return (1);
2034 }
2035
2036 static void
2037 tunkqdetach(struct knote *kn)
2038 {
2039         struct tuntap_softc     *tp = kn->kn_hook;
2040
2041         knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2042 }