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_private.h>
  87 #include <net/if_types.h>
  88 #include <net/if_vlan_var.h>
  89 #include <net/netisr.h>
  90 #include <net/route.h>
  91 #include <net/vnet.h>
  92 #include <netinet/in.h>
  93 #ifdef INET
  94 #include <netinet/ip.h>
  95 #endif
  96 #ifdef INET6
  97 #include <netinet/ip6.h>
  98 #include <netinet6/ip6_var.h>
  99 #endif
 100 #include <netinet/udp.h>
 101 #include <netinet/tcp.h>
 102 #include <net/bpf.h>
 103 #include <net/if_tap.h>
 104 #include <net/if_tun.h>
 105
 106 #include <dev/virtio/network/virtio_net.h>
 107
 108 #include <sys/queue.h>
 109 #include <sys/condvar.h>
 110 #include <security/mac/mac_framework.h>
 111
 112 struct tuntap_driver;
 113
 114 /*
 115  * tun_list is protected by global tunmtx.  Other mutable fields are
 116  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
 117  * static for the duration of a tunnel interface.
 118  */
 119 struct tuntap_softc {
 120         TAILQ_ENTRY(tuntap_softc)        tun_list;
 121         struct cdev                     *tun_alias;
 122         struct cdev                     *tun_dev;
 123         u_short                          tun_flags;     /* misc flags */
 124 #define TUN_OPEN        0x0001
 125 #define TUN_INITED      0x0002
 126 #define TUN_UNUSED1     0x0008
 127 #define TUN_UNUSED2     0x0010
 128 #define TUN_LMODE       0x0020
 129 #define TUN_RWAIT       0x0040
 130 #define TUN_ASYNC       0x0080
 131 #define TUN_IFHEAD      0x0100
 132 #define TUN_DYING       0x0200
 133 #define TUN_L2          0x0400
 134 #define TUN_VMNET       0x0800
 135
 136 #define TUN_DRIVER_IDENT_MASK   (TUN_L2 | TUN_VMNET)
 137 #define TUN_READY               (TUN_OPEN | TUN_INITED)
 138
 139         pid_t                    tun_pid;       /* owning pid */
 140         struct ifnet            *tun_ifp;       /* the interface */
 141         struct sigio            *tun_sigio;     /* async I/O info */
 142         struct tuntap_driver    *tun_drv;       /* appropriate driver */
 143         struct selinfo           tun_rsel;      /* read select */
 144         struct mtx               tun_mtx;       /* softc field mutex */
 145         struct cv                tun_cv;        /* for ref'd dev destroy */
 146         struct ether_addr        tun_ether;     /* remote address */
 147         int                      tun_busy;      /* busy count */
 148         int                      tun_vhdrlen;   /* virtio-net header length */
 149 };
 150 #define TUN2IFP(sc)     ((sc)->tun_ifp)
 151
 152 #define TUNDEBUG        if (tundebug) if_printf
 153
 154 #define TUN_LOCK(tp)            mtx_lock(&(tp)->tun_mtx)
 155 #define TUN_UNLOCK(tp)          mtx_unlock(&(tp)->tun_mtx)
 156 #define TUN_LOCK_ASSERT(tp)     mtx_assert(&(tp)->tun_mtx, MA_OWNED);
 157
 158 #define TUN_VMIO_FLAG_MASK      0x0fff
 159
 160 /*
 161  * Interface capabilities of a tap device that supports the virtio-net
 162  * header.
 163  */
 164 #define TAP_VNET_HDR_CAPS       (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6       \
 165                                 | IFCAP_VLAN_HWCSUM                     \
 166                                 | IFCAP_TSO | IFCAP_LRO                 \
 167                                 | IFCAP_VLAN_HWTSO)
 168
 169 #define TAP_ALL_OFFLOAD         (CSUM_TSO | CSUM_TCP | CSUM_UDP |\
 170                                     CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
 171
 172 /*
 173  * All mutable global variables in if_tun are locked using tunmtx, with
 174  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
 175  * which are static after setup.
 176  */
 177 static struct mtx tunmtx;
 178 static eventhandler_tag arrival_tag;
 179 static eventhandler_tag clone_tag;
 180 static const char tunname[] = "tun";
 181 static const char tapname[] = "tap";
 182 static const char vmnetname[] = "vmnet";
 183 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
 184 static int tundebug = 0;
 185 static int tundclone = 1;
 186 static int tap_allow_uopen = 0; /* allow user devfs cloning */
 187 static int tapuponopen = 0;     /* IFF_UP on open() */
 188 static int tapdclone = 1;       /* enable devfs cloning */
 189
 190 static TAILQ_HEAD(,tuntap_softc)        tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
 191 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
 192
 193 static struct sx tun_ioctl_sx;
 194 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
 195
 196 SYSCTL_DECL(_net_link);
 197 /* tun */
 198 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 199     "IP tunnel software network interface");
 200 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
 201     "Enable legacy devfs interface creation");
 202
 203 /* tap */
 204 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 205     "Ethernet tunnel software network interface");
 206 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
 207     "Enable legacy devfs interface creation for all users");
 208 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
 209     "Bring interface up when /dev/tap is opened");
 210 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
 211     "Enable legacy devfs interface creation");
 212 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
 213
 214 static int      tun_create_device(struct tuntap_driver *drv, int unit,
 215     struct ucred *cr, struct cdev **dev, const char *name);
 216 static int      tun_busy_locked(struct tuntap_softc *tp);
 217 static void     tun_unbusy_locked(struct tuntap_softc *tp);
 218 static int      tun_busy(struct tuntap_softc *tp);
 219 static void     tun_unbusy(struct tuntap_softc *tp);
 220
 221 static int      tuntap_name2info(const char *name, int *unit, int *flags);
 222 static void     tunclone(void *arg, struct ucred *cred, char *name,
 223                     int namelen, struct cdev **dev);
 224 static void     tuncreate(struct cdev *dev);
 225 static void     tundtor(void *data);
 226 static void     tunrename(void *arg, struct ifnet *ifp);
 227 static int      tunifioctl(struct ifnet *, u_long, caddr_t);
 228 static void     tuninit(struct ifnet *);
 229 static void     tunifinit(void *xtp);
 230 static int      tuntapmodevent(module_t, int, void *);
 231 static int      tunoutput(struct ifnet *, struct mbuf *,
 232                     const struct sockaddr *, struct route *ro);
 233 static void     tunstart(struct ifnet *);
 234 static void     tunstart_l2(struct ifnet *);
 235
 236 static int      tun_clone_match(struct if_clone *ifc, const char *name);
 237 static int      tap_clone_match(struct if_clone *ifc, const char *name);
 238 static int      vmnet_clone_match(struct if_clone *ifc, const char *name);
 239 static int      tun_clone_create(struct if_clone *, char *, size_t,
 240                     struct ifc_data *, struct ifnet **);
 241 static int      tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
 242 static void     tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
 243
 244 static d_open_t         tunopen;
 245 static d_read_t         tunread;
 246 static d_write_t        tunwrite;
 247 static d_ioctl_t        tunioctl;
 248 static d_poll_t         tunpoll;
 249 static d_kqfilter_t     tunkqfilter;
 250
 251 static int              tunkqread(struct knote *, long);
 252 static int              tunkqwrite(struct knote *, long);
 253 static void             tunkqdetach(struct knote *);
 254
 255 static struct filterops tun_read_filterops = {
 256         .f_isfd =       1,
 257         .f_attach =     NULL,
 258         .f_detach =     tunkqdetach,
 259         .f_event =      tunkqread,
 260 };
 261
 262 static struct filterops tun_write_filterops = {
 263         .f_isfd =       1,
 264         .f_attach =     NULL,
 265         .f_detach =     tunkqdetach,
 266         .f_event =      tunkqwrite,
 267 };
 268
 269 static struct tuntap_driver {
 270         struct cdevsw            cdevsw;
 271         int                      ident_flags;
 272         struct unrhdr           *unrhdr;
 273         struct clonedevs        *clones;
 274         ifc_match_f             *clone_match_fn;
 275         ifc_create_f            *clone_create_fn;
 276         ifc_destroy_f           *clone_destroy_fn;
 277 } tuntap_drivers[] = {
 278         {
 279                 .ident_flags =  0,
 280                 .cdevsw =       {
 281                     .d_version =        D_VERSION,
 282                     .d_flags =          D_NEEDMINOR,
 283                     .d_open =           tunopen,
 284                     .d_read =           tunread,
 285                     .d_write =          tunwrite,
 286                     .d_ioctl =          tunioctl,
 287                     .d_poll =           tunpoll,
 288                     .d_kqfilter =       tunkqfilter,
 289                     .d_name =           tunname,
 290                 },
 291                 .clone_match_fn =       tun_clone_match,
 292                 .clone_create_fn =      tun_clone_create,
 293                 .clone_destroy_fn =     tun_clone_destroy,
 294         },
 295         {
 296                 .ident_flags =  TUN_L2,
 297                 .cdevsw =       {
 298                     .d_version =        D_VERSION,
 299                     .d_flags =          D_NEEDMINOR,
 300                     .d_open =           tunopen,
 301                     .d_read =           tunread,
 302                     .d_write =          tunwrite,
 303                     .d_ioctl =          tunioctl,
 304                     .d_poll =           tunpoll,
 305                     .d_kqfilter =       tunkqfilter,
 306                     .d_name =           tapname,
 307                 },
 308                 .clone_match_fn =       tap_clone_match,
 309                 .clone_create_fn =      tun_clone_create,
 310                 .clone_destroy_fn =     tun_clone_destroy,
 311         },
 312         {
 313                 .ident_flags =  TUN_L2 | TUN_VMNET,
 314                 .cdevsw =       {
 315                     .d_version =        D_VERSION,
 316                     .d_flags =          D_NEEDMINOR,
 317                     .d_open =           tunopen,
 318                     .d_read =           tunread,
 319                     .d_write =          tunwrite,
 320                     .d_ioctl =          tunioctl,
 321                     .d_poll =           tunpoll,
 322                     .d_kqfilter =       tunkqfilter,
 323                     .d_name =           vmnetname,
 324                 },
 325                 .clone_match_fn =       vmnet_clone_match,
 326                 .clone_create_fn =      tun_clone_create,
 327                 .clone_destroy_fn =     tun_clone_destroy,
 328         },
 329 };
 330
 331 struct tuntap_driver_cloner {
 332         SLIST_ENTRY(tuntap_driver_cloner)        link;
 333         struct tuntap_driver                    *drv;
 334         struct if_clone                         *cloner;
 335 };
 336
 337 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
 338     SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
 339
 340 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
 341
 342 /*
 343  * Mechanism for marking a tunnel device as busy so that we can safely do some
 344  * orthogonal operations (such as operations on devices) without racing against
 345  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
 346  * open, to be woken up when the condition is alleviated.
 347  */
 348 static int
 349 tun_busy_locked(struct tuntap_softc *tp)
 350 {
 351
 352         TUN_LOCK_ASSERT(tp);
 353         if ((tp->tun_flags & TUN_DYING) != 0) {
 354                 /*
 355                  * Perhaps unintuitive, but the device is busy going away.
 356                  * Other interpretations of EBUSY from tun_busy make little
 357                  * sense, since making a busy device even more busy doesn't
 358                  * sound like a problem.
 359                  */
 360                 return (EBUSY);
 361         }
 362
 363         ++tp->tun_busy;
 364         return (0);
 365 }
 366
 367 static void
 368 tun_unbusy_locked(struct tuntap_softc *tp)
 369 {
 370
 371         TUN_LOCK_ASSERT(tp);
 372         KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
 373
 374         --tp->tun_busy;
 375         /* Wake up anything that may be waiting on our busy tunnel. */
 376         if (tp->tun_busy == 0)
 377                 cv_broadcast(&tp->tun_cv);
 378 }
 379
 380 static int
 381 tun_busy(struct tuntap_softc *tp)
 382 {
 383         int ret;
 384
 385         TUN_LOCK(tp);
 386         ret = tun_busy_locked(tp);
 387         TUN_UNLOCK(tp);
 388         return (ret);
 389 }
 390
 391 static void
 392 tun_unbusy(struct tuntap_softc *tp)
 393 {
 394
 395         TUN_LOCK(tp);
 396         tun_unbusy_locked(tp);
 397         TUN_UNLOCK(tp);
 398 }
 399
 400 /*
 401  * Sets unit and/or flags given the device name.  Must be called with correct
 402  * vnet context.
 403  */
 404 static int
 405 tuntap_name2info(const char *name, int *outunit, int *outflags)
 406 {
 407         struct tuntap_driver *drv;
 408         struct tuntap_driver_cloner *drvc;
 409         char *dname;
 410         int flags, unit;
 411         bool found;
 412
 413         if (name == NULL)
 414                 return (EINVAL);
 415
 416         /*
 417          * Needed for dev_stdclone, but dev_stdclone will not modify, it just
 418          * wants to be able to pass back a char * through the second param. We
 419          * will always set that as NULL here, so we'll fake it.
 420          */
 421         dname = __DECONST(char *, name);
 422         found = false;
 423
 424         KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
 425             ("tuntap_driver_cloners failed to initialize"));
 426         SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
 427                 KASSERT(drvc->drv != NULL,
 428                     ("tuntap_driver_cloners entry not properly initialized"));
 429                 drv = drvc->drv;
 430
 431                 if (strcmp(name, drv->cdevsw.d_name) == 0) {
 432                         found = true;
 433                         unit = -1;
 434                         flags = drv->ident_flags;
 435                         break;
 436                 }
 437
 438                 if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
 439                         found = true;
 440                         flags = drv->ident_flags;
 441                         break;
 442                 }
 443         }
 444
 445         if (!found)
 446                 return (ENXIO);
 447
 448         if (outunit != NULL)
 449                 *outunit = unit;
 450         if (outflags != NULL)
 451                 *outflags = flags;
 452         return (0);
 453 }
 454
 455 /*
 456  * Get driver information from a set of flags specified.  Masks the identifying
 457  * part of the flags and compares it against all of the available
 458  * tuntap_drivers. Must be called with correct vnet context.
 459  */
 460 static struct tuntap_driver *
 461 tuntap_driver_from_flags(int tun_flags)
 462 {
 463         struct tuntap_driver *drv;
 464         struct tuntap_driver_cloner *drvc;
 465
 466         KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
 467             ("tuntap_driver_cloners failed to initialize"));
 468         SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
 469                 KASSERT(drvc->drv != NULL,
 470                     ("tuntap_driver_cloners entry not properly initialized"));
 471                 drv = drvc->drv;
 472                 if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->ident_flags)
 473                         return (drv);
 474         }
 475
 476         return (NULL);
 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,
 520     struct ifc_data *ifd, struct ifnet **ifpp)
 521 {
 522         struct tuntap_driver *drv;
 523         struct cdev *dev;
 524         int err, i, tunflags, unit;
 525
 526         tunflags = 0;
 527         /* The name here tells us exactly what we're creating */
 528         err = tuntap_name2info(name, &unit, &tunflags);
 529         if (err != 0)
 530                 return (err);
 531
 532         drv = tuntap_driver_from_flags(tunflags);
 533         if (drv == NULL)
 534                 return (ENXIO);
 535
 536         if (unit != -1) {
 537                 /* If this unit number is still available that's okay. */
 538                 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
 539                         return (EEXIST);
 540         } else {
 541                 unit = alloc_unr(drv->unrhdr);
 542         }
 543
 544         snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
 545
 546         /* find any existing device, or allocate new unit number */
 547         dev = NULL;
 548         i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
 549         /* No preexisting struct cdev *, create one */
 550         if (i != 0)
 551                 i = tun_create_device(drv, unit, NULL, &dev, name);
 552         if (i == 0) {
 553                 tuncreate(dev);
 554                 struct tuntap_softc *tp = dev->si_drv1;
 555                 *ifpp = tp->tun_ifp;
 556         }
 557
 558         return (i);
 559 }
 560
 561 static void
 562 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
 563     struct cdev **dev)
 564 {
 565         char devname[SPECNAMELEN + 1];
 566         struct tuntap_driver *drv;
 567         int append_unit, i, u, tunflags;
 568         bool mayclone;
 569
 570         if (*dev != NULL)
 571                 return;
 572
 573         tunflags = 0;
 574         CURVNET_SET(CRED_TO_VNET(cred));
 575         if (tuntap_name2info(name, &u, &tunflags) != 0)
 576                 goto out;       /* Not recognized */
 577
 578         if (u != -1 && u > IF_MAXUNIT)
 579                 goto out;       /* Unit number too high */
 580
 581         mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
 582         if ((tunflags & TUN_L2) != 0) {
 583                 /* tap/vmnet allow user open with a sysctl */
 584                 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
 585         } else {
 586                 mayclone = mayclone && tundclone;
 587         }
 588
 589         /*
 590          * If tun cloning is enabled, only the superuser can create an
 591          * interface.
 592          */
 593         if (!mayclone)
 594                 goto out;
 595
 596         if (u == -1)
 597                 append_unit = 1;
 598         else
 599                 append_unit = 0;
 600
 601         drv = tuntap_driver_from_flags(tunflags);
 602         if (drv == NULL)
 603                 goto out;
 604
 605         /* find any existing device, or allocate new unit number */
 606         i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
 607         if (i) {
 608                 if (append_unit) {
 609                         namelen = snprintf(devname, sizeof(devname), "%s%d",
 610                             name, u);
 611                         name = devname;
 612                 }
 613
 614                 i = tun_create_device(drv, u, cred, dev, name);
 615         }
 616         if (i == 0)
 617                 if_clone_create(name, namelen, NULL);
 618 out:
 619         CURVNET_RESTORE();
 620 }
 621
 622 static void
 623 tun_destroy(struct tuntap_softc *tp)
 624 {
 625
 626         TUN_LOCK(tp);
 627         tp->tun_flags |= TUN_DYING;
 628         if (tp->tun_busy != 0)
 629                 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
 630         else
 631                 TUN_UNLOCK(tp);
 632
 633         CURVNET_SET(TUN2IFP(tp)->if_vnet);
 634
 635         /* destroy_dev will take care of any alias. */
 636         destroy_dev(tp->tun_dev);
 637         seldrain(&tp->tun_rsel);
 638         knlist_clear(&tp->tun_rsel.si_note, 0);
 639         knlist_destroy(&tp->tun_rsel.si_note);
 640         if ((tp->tun_flags & TUN_L2) != 0) {
 641                 ether_ifdetach(TUN2IFP(tp));
 642         } else {
 643                 bpfdetach(TUN2IFP(tp));
 644                 if_detach(TUN2IFP(tp));
 645         }
 646         sx_xlock(&tun_ioctl_sx);
 647         TUN2IFP(tp)->if_softc = NULL;
 648         sx_xunlock(&tun_ioctl_sx);
 649         free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
 650         if_free(TUN2IFP(tp));
 651         mtx_destroy(&tp->tun_mtx);
 652         cv_destroy(&tp->tun_cv);
 653         free(tp, M_TUN);
 654         CURVNET_RESTORE();
 655 }
 656
 657 static int
 658 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
 659 {
 660         struct tuntap_softc *tp = ifp->if_softc;
 661
 662         mtx_lock(&tunmtx);
 663         TAILQ_REMOVE(&tunhead, tp, tun_list);
 664         mtx_unlock(&tunmtx);
 665         tun_destroy(tp);
 666
 667         return (0);
 668 }
 669
 670 static void
 671 vnet_tun_init(const void *unused __unused)
 672 {
 673         struct tuntap_driver *drv;
 674         struct tuntap_driver_cloner *drvc;
 675         int i;
 676
 677         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 678                 drv = &tuntap_drivers[i];
 679                 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
 680
 681                 drvc->drv = drv;
 682                 struct if_clone_addreq req = {
 683                         .match_f = drv->clone_match_fn,
 684                         .create_f = drv->clone_create_fn,
 685                         .destroy_f = drv->clone_destroy_fn,
 686                 };
 687                 drvc->cloner = ifc_attach_cloner(drv->cdevsw.d_name, &req);
 688                 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
 689         };
 690 }
 691 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 692                 vnet_tun_init, NULL);
 693
 694 static void
 695 vnet_tun_uninit(const void *unused __unused)
 696 {
 697         struct tuntap_driver_cloner *drvc;
 698
 699         while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
 700                 drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
 701                 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
 702
 703                 if_clone_detach(drvc->cloner);
 704                 free(drvc, M_TUN);
 705         }
 706 }
 707 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 708     vnet_tun_uninit, NULL);
 709
 710 static void
 711 tun_uninit(const void *unused __unused)
 712 {
 713         struct tuntap_driver *drv;
 714         struct tuntap_softc *tp;
 715         int i;
 716
 717         EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
 718         EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
 719
 720         mtx_lock(&tunmtx);
 721         while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
 722                 TAILQ_REMOVE(&tunhead, tp, tun_list);
 723                 mtx_unlock(&tunmtx);
 724                 tun_destroy(tp);
 725                 mtx_lock(&tunmtx);
 726         }
 727         mtx_unlock(&tunmtx);
 728         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 729                 drv = &tuntap_drivers[i];
 730                 delete_unrhdr(drv->unrhdr);
 731                 clone_cleanup(&drv->clones);
 732         }
 733         mtx_destroy(&tunmtx);
 734 }
 735 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
 736
 737 static struct tuntap_driver *
 738 tuntap_driver_from_ifnet(const struct ifnet *ifp)
 739 {
 740         struct tuntap_driver *drv;
 741         int i;
 742
 743         if (ifp == NULL)
 744                 return (NULL);
 745
 746         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 747                 drv = &tuntap_drivers[i];
 748                 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
 749                         return (drv);
 750         }
 751
 752         return (NULL);
 753 }
 754
 755 static int
 756 tuntapmodevent(module_t mod, int type, void *data)
 757 {
 758         struct tuntap_driver *drv;
 759         int i;
 760
 761         switch (type) {
 762         case MOD_LOAD:
 763                 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
 764                 for (i = 0; i < nitems(tuntap_drivers); ++i) {
 765                         drv = &tuntap_drivers[i];
 766                         clone_setup(&drv->clones);
 767                         drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
 768                 }
 769                 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
 770                    tunrename, 0, 1000);
 771                 if (arrival_tag == NULL)
 772                         return (ENOMEM);
 773                 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
 774                 if (clone_tag == NULL)
 775                         return (ENOMEM);
 776                 break;
 777         case MOD_UNLOAD:
 778                 /* See tun_uninit, so it's done after the vnet_sysuninit() */
 779                 break;
 780         default:
 781                 return EOPNOTSUPP;
 782         }
 783         return 0;
 784 }
 785
 786 static moduledata_t tuntap_mod = {
 787         "if_tuntap",
 788         tuntapmodevent,
 789         0
 790 };
 791
 792 /* We'll only ever have these two, so no need for a macro. */
 793 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
 794 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
 795
 796 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 797 MODULE_VERSION(if_tuntap, 1);
 798 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 799 MODULE_VERSION(if_tun, 1);
 800 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 801 MODULE_VERSION(if_tap, 1);
 802
 803 static int
 804 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
 805     struct cdev **dev, const char *name)
 806 {
 807         struct make_dev_args args;
 808         struct tuntap_softc *tp;
 809         int error;
 810
 811         tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
 812         mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
 813         cv_init(&tp->tun_cv, "tun_condvar");
 814         tp->tun_flags = drv->ident_flags;
 815         tp->tun_drv = drv;
 816
 817         make_dev_args_init(&args);
 818         if (cr != NULL)
 819                 args.mda_flags = MAKEDEV_REF;
 820         args.mda_devsw = &drv->cdevsw;
 821         args.mda_cr = cr;
 822         args.mda_uid = UID_UUCP;
 823         args.mda_gid = GID_DIALER;
 824         args.mda_mode = 0600;
 825         args.mda_unit = unit;
 826         args.mda_si_drv1 = tp;
 827         error = make_dev_s(&args, dev, "%s", name);
 828         if (error != 0) {
 829                 free(tp, M_TUN);
 830                 return (error);
 831         }
 832
 833         KASSERT((*dev)->si_drv1 != NULL,
 834             ("Failed to set si_drv1 at %s creation", name));
 835         tp->tun_dev = *dev;
 836         knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
 837         mtx_lock(&tunmtx);
 838         TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
 839         mtx_unlock(&tunmtx);
 840         return (0);
 841 }
 842
 843 static void
 844 tunstart(struct ifnet *ifp)
 845 {
 846         struct tuntap_softc *tp = ifp->if_softc;
 847         struct mbuf *m;
 848
 849         TUNDEBUG(ifp, "starting\n");
 850         if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
 851                 IFQ_LOCK(&ifp->if_snd);
 852                 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
 853                 if (m == NULL) {
 854                         IFQ_UNLOCK(&ifp->if_snd);
 855                         return;
 856                 }
 857                 IFQ_UNLOCK(&ifp->if_snd);
 858         }
 859
 860         TUN_LOCK(tp);
 861         if (tp->tun_flags & TUN_RWAIT) {
 862                 tp->tun_flags &= ~TUN_RWAIT;
 863                 wakeup(tp);
 864         }
 865         selwakeuppri(&tp->tun_rsel, PZERO + 1);
 866         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 867         if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
 868                 TUN_UNLOCK(tp);
 869                 pgsigio(&tp->tun_sigio, SIGIO, 0);
 870         } else
 871                 TUN_UNLOCK(tp);
 872 }
 873
 874 /*
 875  * tunstart_l2
 876  *
 877  * queue packets from higher level ready to put out
 878  */
 879 static void
 880 tunstart_l2(struct ifnet *ifp)
 881 {
 882         struct tuntap_softc     *tp = ifp->if_softc;
 883
 884         TUNDEBUG(ifp, "starting\n");
 885
 886         /*
 887          * do not junk pending output if we are in VMnet mode.
 888          * XXX: can this do any harm because of queue overflow?
 889          */
 890
 891         TUN_LOCK(tp);
 892         if (((tp->tun_flags & TUN_VMNET) == 0) &&
 893             ((tp->tun_flags & TUN_READY) != TUN_READY)) {
 894                 struct mbuf *m;
 895
 896                 /* Unlocked read. */
 897                 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
 898
 899                 for (;;) {
 900                         IF_DEQUEUE(&ifp->if_snd, m);
 901                         if (m != NULL) {
 902                                 m_freem(m);
 903                                 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 904                         } else
 905                                 break;
 906                 }
 907                 TUN_UNLOCK(tp);
 908
 909                 return;
 910         }
 911
 912         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
 913
 914         if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
 915                 if (tp->tun_flags & TUN_RWAIT) {
 916                         tp->tun_flags &= ~TUN_RWAIT;
 917                         wakeup(tp);
 918                 }
 919
 920                 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
 921                         TUN_UNLOCK(tp);
 922                         pgsigio(&tp->tun_sigio, SIGIO, 0);
 923                         TUN_LOCK(tp);
 924                 }
 925
 926                 selwakeuppri(&tp->tun_rsel, PZERO+1);
 927                 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 928                 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
 929         }
 930
 931         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
 932         TUN_UNLOCK(tp);
 933 } /* tunstart_l2 */
 934
 935 /* XXX: should return an error code so it can fail. */
 936 static void
 937 tuncreate(struct cdev *dev)
 938 {
 939         struct tuntap_driver *drv;
 940         struct tuntap_softc *tp;
 941         struct ifnet *ifp;
 942         struct ether_addr eaddr;
 943         int iflags;
 944         u_char type;
 945
 946         tp = dev->si_drv1;
 947         KASSERT(tp != NULL,
 948             ("si_drv1 should have been initialized at creation"));
 949
 950         drv = tp->tun_drv;
 951         iflags = IFF_MULTICAST;
 952         if ((tp->tun_flags & TUN_L2) != 0) {
 953                 type = IFT_ETHER;
 954                 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
 955         } else {
 956                 type = IFT_PPP;
 957                 iflags |= IFF_POINTOPOINT;
 958         }
 959         ifp = tp->tun_ifp = if_alloc(type);
 960         if (ifp == NULL)
 961                 panic("%s%d: failed to if_alloc() interface.\n",
 962                     drv->cdevsw.d_name, dev2unit(dev));
 963         ifp->if_softc = tp;
 964         if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
 965         ifp->if_ioctl = tunifioctl;
 966         ifp->if_flags = iflags;
 967         IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
 968         ifp->if_capabilities |= IFCAP_LINKSTATE;
 969         ifp->if_capenable |= IFCAP_LINKSTATE;
 970
 971         if ((tp->tun_flags & TUN_L2) != 0) {
 972                 ifp->if_init = tunifinit;
 973                 ifp->if_start = tunstart_l2;
 974
 975                 ether_gen_addr(ifp, &eaddr);
 976                 ether_ifattach(ifp, eaddr.octet);
 977         } else {
 978                 ifp->if_mtu = TUNMTU;
 979                 ifp->if_start = tunstart;
 980                 ifp->if_output = tunoutput;
 981
 982                 ifp->if_snd.ifq_drv_maxlen = 0;
 983                 IFQ_SET_READY(&ifp->if_snd);
 984
 985                 if_attach(ifp);
 986                 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
 987         }
 988
 989         TUN_LOCK(tp);
 990         tp->tun_flags |= TUN_INITED;
 991         TUN_UNLOCK(tp);
 992
 993         TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
 994             ifp->if_xname, dev2unit(dev));
 995 }
 996
 997 static void
 998 tunrename(void *arg __unused, struct ifnet *ifp)
 999 {
1000         struct tuntap_softc *tp;
1001         int error;
1002
1003         if ((ifp->if_flags & IFF_RENAMING) == 0)
1004                 return;
1005
1006         if (tuntap_driver_from_ifnet(ifp) == NULL)
1007                 return;
1008
1009         /*
1010          * We need to grab the ioctl sx long enough to make sure the softc is
1011          * still there.  If it is, we can safely try to busy the tun device.
1012          * The busy may fail if the device is currently dying, in which case
1013          * we do nothing.  If it doesn't fail, the busy count stops the device
1014          * from dying until we've created the alias (that will then be
1015          * subsequently destroyed).
1016          */
1017         sx_xlock(&tun_ioctl_sx);
1018         tp = ifp->if_softc;
1019         if (tp == NULL) {
1020                 sx_xunlock(&tun_ioctl_sx);
1021                 return;
1022         }
1023         error = tun_busy(tp);
1024         sx_xunlock(&tun_ioctl_sx);
1025         if (error != 0)
1026                 return;
1027         if (tp->tun_alias != NULL) {
1028                 destroy_dev(tp->tun_alias);
1029                 tp->tun_alias = NULL;
1030         }
1031
1032         if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1033                 goto out;
1034
1035         /*
1036          * Failure's ok, aliases are created on a best effort basis.  If a
1037          * tun user/consumer decides to rename the interface to conflict with
1038          * another device (non-ifnet) on the system, we will assume they know
1039          * what they are doing.  make_dev_alias_p won't touch tun_alias on
1040          * failure, so we use it but ignore the return value.
1041          */
1042         make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1043             ifp->if_xname);
1044 out:
1045         tun_unbusy(tp);
1046 }
1047
1048 static int
1049 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1050 {
1051         struct ifnet    *ifp;
1052         struct tuntap_softc *tp;
1053         int error __diagused, tunflags;
1054
1055         tunflags = 0;
1056         CURVNET_SET(TD_TO_VNET(td));
1057         error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1058         if (error != 0) {
1059                 CURVNET_RESTORE();
1060                 return (error); /* Shouldn't happen */
1061         }
1062
1063         tp = dev->si_drv1;
1064         KASSERT(tp != NULL,
1065             ("si_drv1 should have been initialized at creation"));
1066
1067         TUN_LOCK(tp);
1068         if ((tp->tun_flags & TUN_INITED) == 0) {
1069                 TUN_UNLOCK(tp);
1070                 CURVNET_RESTORE();
1071                 return (ENXIO);
1072         }
1073         if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1074                 TUN_UNLOCK(tp);
1075                 CURVNET_RESTORE();
1076                 return (EBUSY);
1077         }
1078
1079         error = tun_busy_locked(tp);
1080         KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1081         ifp = TUN2IFP(tp);
1082
1083         if ((tp->tun_flags & TUN_L2) != 0) {
1084                 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1085                     sizeof(tp->tun_ether.octet));
1086
1087                 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1088                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1089
1090                 if (tapuponopen)
1091                         ifp->if_flags |= IFF_UP;
1092         }
1093
1094         tp->tun_pid = td->td_proc->p_pid;
1095         tp->tun_flags |= TUN_OPEN;
1096
1097         if_link_state_change(ifp, LINK_STATE_UP);
1098         TUNDEBUG(ifp, "open\n");
1099         TUN_UNLOCK(tp);
1100
1101         /*
1102          * This can fail with either ENOENT or EBUSY.  This is in the middle of
1103          * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1104          * the only cdevpriv dtor being set will be tundtor and the softc being
1105          * passed is constant for a given cdev.  We ignore the possible error
1106          * because of this as either "unlikely" or "not actually a problem."
1107          */
1108         (void)devfs_set_cdevpriv(tp, tundtor);
1109         CURVNET_RESTORE();
1110         return (0);
1111 }
1112
1113 /*
1114  * tundtor - tear down the device - mark i/f down & delete
1115  * routing info
1116  */
1117 static void
1118 tundtor(void *data)
1119 {
1120         struct proc *p;
1121         struct tuntap_softc *tp;
1122         struct ifnet *ifp;
1123         bool l2tun;
1124
1125         tp = data;
1126         p = curproc;
1127         ifp = TUN2IFP(tp);
1128
1129         TUN_LOCK(tp);
1130
1131         /*
1132          * Realistically, we can't be obstinate here.  This only means that the
1133          * tuntap device was closed out of order, and the last closer wasn't the
1134          * controller.  These are still good to know about, though, as software
1135          * should avoid multiple processes with a tuntap device open and
1136          * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1137          * parent).
1138          */
1139         if (p->p_pid != tp->tun_pid) {
1140                 log(LOG_INFO,
1141                     "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1142                     p->p_pid, p->p_comm, tp->tun_dev->si_name);
1143         }
1144
1145         /*
1146          * junk all pending output
1147          */
1148         CURVNET_SET(ifp->if_vnet);
1149
1150         l2tun = false;
1151         if ((tp->tun_flags & TUN_L2) != 0) {
1152                 l2tun = true;
1153                 IF_DRAIN(&ifp->if_snd);
1154         } else {
1155                 IFQ_PURGE(&ifp->if_snd);
1156         }
1157
1158         /* For vmnet, we won't do most of the address/route bits */
1159         if ((tp->tun_flags & TUN_VMNET) != 0 ||
1160             (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1161                 goto out;
1162
1163         if (ifp->if_flags & IFF_UP) {
1164                 TUN_UNLOCK(tp);
1165                 if_down(ifp);
1166                 TUN_LOCK(tp);
1167         }
1168
1169         /* Delete all addresses and routes which reference this interface. */
1170         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1171                 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1172                 TUN_UNLOCK(tp);
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
1198         TUNDEBUG(ifp, "tuninit\n");
1199
1200         TUN_LOCK(tp);
1201         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1202         if ((tp->tun_flags & TUN_L2) == 0) {
1203                 ifp->if_flags |= IFF_UP;
1204                 getmicrotime(&ifp->if_lastchange);
1205                 TUN_UNLOCK(tp);
1206         } else {
1207                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1208                 TUN_UNLOCK(tp);
1209                 /* attempt to start output */
1210                 tunstart_l2(ifp);
1211         }
1212
1213 }
1214
1215 /*
1216  * Used only for l2 tunnel.
1217  */
1218 static void
1219 tunifinit(void *xtp)
1220 {
1221         struct tuntap_softc *tp;
1222
1223         tp = (struct tuntap_softc *)xtp;
1224         tuninit(tp->tun_ifp);
1225 }
1226
1227 /*
1228  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1229  * current value of ifp->if_capenable.
1230  */
1231 static void
1232 tun_caps_changed(struct ifnet *ifp)
1233 {
1234         uint64_t hwassist = 0;
1235
1236         TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1237         if (ifp->if_capenable & IFCAP_TXCSUM)
1238                 hwassist |= CSUM_TCP | CSUM_UDP;
1239         if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1240                 hwassist |= CSUM_TCP_IPV6
1241                     | CSUM_UDP_IPV6;
1242         if (ifp->if_capenable & IFCAP_TSO4)
1243                 hwassist |= CSUM_IP_TSO;
1244         if (ifp->if_capenable & IFCAP_TSO6)
1245                 hwassist |= CSUM_IP6_TSO;
1246         ifp->if_hwassist = hwassist;
1247 }
1248
1249 /*
1250  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1251  * if_capabilities and if_capenable as needed.
1252  */
1253 static void
1254 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1255 {
1256         struct tuntap_softc *tp = ifp->if_softc;
1257
1258         TUN_LOCK_ASSERT(tp);
1259
1260         if (tp->tun_vhdrlen == vhdrlen)
1261                 return;
1262
1263         /*
1264          * Update if_capabilities to reflect the
1265          * functionalities offered by the virtio-net
1266          * header.
1267          */
1268         if (vhdrlen != 0)
1269                 ifp->if_capabilities |=
1270                         TAP_VNET_HDR_CAPS;
1271         else
1272                 ifp->if_capabilities &=
1273                         ~TAP_VNET_HDR_CAPS;
1274         /*
1275          * Disable any capabilities that we don't
1276          * support anymore.
1277          */
1278         ifp->if_capenable &= ifp->if_capabilities;
1279         tun_caps_changed(ifp);
1280         tp->tun_vhdrlen = vhdrlen;
1281
1282         TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1283             vhdrlen, ifp->if_capabilities);
1284 }
1285
1286 /*
1287  * Process an ioctl request.
1288  */
1289 static int
1290 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1291 {
1292         struct ifreq *ifr = (struct ifreq *)data;
1293         struct tuntap_softc *tp;
1294         struct ifstat *ifs;
1295         struct ifmediareq       *ifmr;
1296         int             dummy, error = 0;
1297         bool            l2tun;
1298
1299         ifmr = NULL;
1300         sx_xlock(&tun_ioctl_sx);
1301         tp = ifp->if_softc;
1302         if (tp == NULL) {
1303                 error = ENXIO;
1304                 goto bad;
1305         }
1306         l2tun = (tp->tun_flags & TUN_L2) != 0;
1307         switch(cmd) {
1308         case SIOCGIFSTATUS:
1309                 ifs = (struct ifstat *)data;
1310                 TUN_LOCK(tp);
1311                 if (tp->tun_pid)
1312                         snprintf(ifs->ascii, sizeof(ifs->ascii),
1313                             "\tOpened by PID %d\n", tp->tun_pid);
1314                 else
1315                         ifs->ascii[0] = '\0';
1316                 TUN_UNLOCK(tp);
1317                 break;
1318         case SIOCSIFADDR:
1319                 if (l2tun)
1320                         error = ether_ioctl(ifp, cmd, data);
1321                 else
1322                         tuninit(ifp);
1323                 if (error == 0)
1324                     TUNDEBUG(ifp, "address set\n");
1325                 break;
1326         case SIOCSIFMTU:
1327                 ifp->if_mtu = ifr->ifr_mtu;
1328                 TUNDEBUG(ifp, "mtu set\n");
1329                 break;
1330         case SIOCSIFFLAGS:
1331         case SIOCADDMULTI:
1332         case SIOCDELMULTI:
1333                 break;
1334         case SIOCGIFMEDIA:
1335                 if (!l2tun) {
1336                         error = EINVAL;
1337                         break;
1338                 }
1339
1340                 ifmr = (struct ifmediareq *)data;
1341                 dummy = ifmr->ifm_count;
1342                 ifmr->ifm_count = 1;
1343                 ifmr->ifm_status = IFM_AVALID;
1344                 ifmr->ifm_active = IFM_ETHER;
1345                 if (tp->tun_flags & TUN_OPEN)
1346                         ifmr->ifm_status |= IFM_ACTIVE;
1347                 ifmr->ifm_current = ifmr->ifm_active;
1348                 if (dummy >= 1) {
1349                         int media = IFM_ETHER;
1350                         error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1351                 }
1352                 break;
1353         case SIOCSIFCAP:
1354                 TUN_LOCK(tp);
1355                 ifp->if_capenable = ifr->ifr_reqcap;
1356                 tun_caps_changed(ifp);
1357                 TUN_UNLOCK(tp);
1358                 VLAN_CAPABILITIES(ifp);
1359                 break;
1360         default:
1361                 if (l2tun) {
1362                         error = ether_ioctl(ifp, cmd, data);
1363                 } else {
1364                         error = EINVAL;
1365                 }
1366         }
1367 bad:
1368         sx_xunlock(&tun_ioctl_sx);
1369         return (error);
1370 }
1371
1372 /*
1373  * tunoutput - queue packets from higher level ready to put out.
1374  */
1375 static int
1376 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1377     struct route *ro)
1378 {
1379         struct tuntap_softc *tp = ifp->if_softc;
1380         u_short cached_tun_flags;
1381         int error;
1382         u_int32_t af;
1383
1384         TUNDEBUG (ifp, "tunoutput\n");
1385
1386 #ifdef MAC
1387         error = mac_ifnet_check_transmit(ifp, m0);
1388         if (error) {
1389                 m_freem(m0);
1390                 return (error);
1391         }
1392 #endif
1393
1394         /* Could be unlocked read? */
1395         TUN_LOCK(tp);
1396         cached_tun_flags = tp->tun_flags;
1397         TUN_UNLOCK(tp);
1398         if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1399                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1400                 m_freem (m0);
1401                 return (EHOSTDOWN);
1402         }
1403
1404         if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1405                 m_freem (m0);
1406                 return (EHOSTDOWN);
1407         }
1408
1409         /* BPF writes need to be handled specially. */
1410         if (dst->sa_family == AF_UNSPEC)
1411                 bcopy(dst->sa_data, &af, sizeof(af));
1412         else
1413                 af = RO_GET_FAMILY(ro, dst);
1414
1415         if (bpf_peers_present(ifp->if_bpf))
1416                 bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1417
1418         /* prepend sockaddr? this may abort if the mbuf allocation fails */
1419         if (cached_tun_flags & TUN_LMODE) {
1420                 /* allocate space for sockaddr */
1421                 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1422
1423                 /* if allocation failed drop packet */
1424                 if (m0 == NULL) {
1425                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1426                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1427                         return (ENOBUFS);
1428                 } else {
1429                         bcopy(dst, m0->m_data, dst->sa_len);
1430                 }
1431         }
1432
1433         if (cached_tun_flags & TUN_IFHEAD) {
1434                 /* Prepend the address family */
1435                 M_PREPEND(m0, 4, M_NOWAIT);
1436
1437                 /* if allocation failed drop packet */
1438                 if (m0 == NULL) {
1439                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1440                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1441                         return (ENOBUFS);
1442                 } else
1443                         *(u_int32_t *)m0->m_data = htonl(af);
1444         } else {
1445 #ifdef INET
1446                 if (af != AF_INET)
1447 #endif
1448                 {
1449                         m_freem(m0);
1450                         return (EAFNOSUPPORT);
1451                 }
1452         }
1453
1454         error = (ifp->if_transmit)(ifp, m0);
1455         if (error)
1456                 return (ENOBUFS);
1457         if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1458         return (0);
1459 }
1460
1461 /*
1462  * the cdevsw interface is now pretty minimal.
1463  */
1464 static  int
1465 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1466     struct thread *td)
1467 {
1468         struct ifreq ifr, *ifrp;
1469         struct tuntap_softc *tp = dev->si_drv1;
1470         struct ifnet *ifp = TUN2IFP(tp);
1471         struct tuninfo *tunp;
1472         int error, iflags, ival;
1473         bool    l2tun;
1474
1475         l2tun = (tp->tun_flags & TUN_L2) != 0;
1476         if (l2tun) {
1477                 /* tap specific ioctls */
1478                 switch(cmd) {
1479                 /* VMware/VMnet port ioctl's */
1480 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1481     defined(COMPAT_FREEBSD4)
1482                 case _IO('V', 0):
1483                         ival = IOCPARM_IVAL(data);
1484                         data = (caddr_t)&ival;
1485                         /* FALLTHROUGH */
1486 #endif
1487                 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1488                         iflags = *(int *)data;
1489                         iflags &= TUN_VMIO_FLAG_MASK;
1490                         iflags &= ~IFF_CANTCHANGE;
1491                         iflags |= IFF_UP;
1492
1493                         TUN_LOCK(tp);
1494                         ifp->if_flags = iflags |
1495                             (ifp->if_flags & IFF_CANTCHANGE);
1496                         TUN_UNLOCK(tp);
1497
1498                         return (0);
1499                 case SIOCGIFADDR:       /* get MAC address of the remote side */
1500                         TUN_LOCK(tp);
1501                         bcopy(&tp->tun_ether.octet, data,
1502                             sizeof(tp->tun_ether.octet));
1503                         TUN_UNLOCK(tp);
1504
1505                         return (0);
1506                 case SIOCSIFADDR:       /* set MAC address of the remote side */
1507                         TUN_LOCK(tp);
1508                         bcopy(data, &tp->tun_ether.octet,
1509                             sizeof(tp->tun_ether.octet));
1510                         TUN_UNLOCK(tp);
1511
1512                         return (0);
1513                 case TAPSVNETHDR:
1514                         ival = *(int *)data;
1515                         if (ival != 0 &&
1516                             ival != sizeof(struct virtio_net_hdr) &&
1517                             ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1518                                 return (EINVAL);
1519                         }
1520                         TUN_LOCK(tp);
1521                         tun_vnethdr_set(ifp, ival);
1522                         TUN_UNLOCK(tp);
1523
1524                         return (0);
1525                 case TAPGVNETHDR:
1526                         TUN_LOCK(tp);
1527                         *(int *)data = tp->tun_vhdrlen;
1528                         TUN_UNLOCK(tp);
1529
1530                         return (0);
1531                 }
1532
1533                 /* Fall through to the common ioctls if unhandled */
1534         } else {
1535                 switch (cmd) {
1536                 case TUNSLMODE:
1537                         TUN_LOCK(tp);
1538                         if (*(int *)data) {
1539                                 tp->tun_flags |= TUN_LMODE;
1540                                 tp->tun_flags &= ~TUN_IFHEAD;
1541                         } else
1542                                 tp->tun_flags &= ~TUN_LMODE;
1543                         TUN_UNLOCK(tp);
1544
1545                         return (0);
1546                 case TUNSIFHEAD:
1547                         TUN_LOCK(tp);
1548                         if (*(int *)data) {
1549                                 tp->tun_flags |= TUN_IFHEAD;
1550                                 tp->tun_flags &= ~TUN_LMODE;
1551                         } else
1552                                 tp->tun_flags &= ~TUN_IFHEAD;
1553                         TUN_UNLOCK(tp);
1554
1555                         return (0);
1556                 case TUNGIFHEAD:
1557                         TUN_LOCK(tp);
1558                         *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1559                         TUN_UNLOCK(tp);
1560
1561                         return (0);
1562                 case TUNSIFMODE:
1563                         /* deny this if UP */
1564                         if (TUN2IFP(tp)->if_flags & IFF_UP)
1565                                 return (EBUSY);
1566
1567                         switch (*(int *)data & ~IFF_MULTICAST) {
1568                         case IFF_POINTOPOINT:
1569                         case IFF_BROADCAST:
1570                                 TUN_LOCK(tp);
1571                                 TUN2IFP(tp)->if_flags &=
1572                                     ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1573                                 TUN2IFP(tp)->if_flags |= *(int *)data;
1574                                 TUN_UNLOCK(tp);
1575
1576                                 break;
1577                         default:
1578                                 return (EINVAL);
1579                         }
1580
1581                         return (0);
1582                 case TUNSIFPID:
1583                         TUN_LOCK(tp);
1584                         tp->tun_pid = curthread->td_proc->p_pid;
1585                         TUN_UNLOCK(tp);
1586
1587                         return (0);
1588                 }
1589                 /* Fall through to the common ioctls if unhandled */
1590         }
1591
1592         switch (cmd) {
1593         case TUNGIFNAME:
1594                 ifrp = (struct ifreq *)data;
1595                 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1596
1597                 return (0);
1598         case TUNSIFINFO:
1599                 tunp = (struct tuninfo *)data;
1600                 if (TUN2IFP(tp)->if_type != tunp->type)
1601                         return (EPROTOTYPE);
1602                 TUN_LOCK(tp);
1603                 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1604                         strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1605                         ifr.ifr_mtu = tunp->mtu;
1606                         CURVNET_SET(TUN2IFP(tp)->if_vnet);
1607                         error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1608                             (caddr_t)&ifr, td);
1609                         CURVNET_RESTORE();
1610                         if (error) {
1611                                 TUN_UNLOCK(tp);
1612                                 return (error);
1613                         }
1614                 }
1615                 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1616                 TUN_UNLOCK(tp);
1617                 break;
1618         case TUNGIFINFO:
1619                 tunp = (struct tuninfo *)data;
1620                 TUN_LOCK(tp);
1621                 tunp->mtu = TUN2IFP(tp)->if_mtu;
1622                 tunp->type = TUN2IFP(tp)->if_type;
1623                 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1624                 TUN_UNLOCK(tp);
1625                 break;
1626         case TUNSDEBUG:
1627                 tundebug = *(int *)data;
1628                 break;
1629         case TUNGDEBUG:
1630                 *(int *)data = tundebug;
1631                 break;
1632         case FIONBIO:
1633                 break;
1634         case FIOASYNC:
1635                 TUN_LOCK(tp);
1636                 if (*(int *)data)
1637                         tp->tun_flags |= TUN_ASYNC;
1638                 else
1639                         tp->tun_flags &= ~TUN_ASYNC;
1640                 TUN_UNLOCK(tp);
1641                 break;
1642         case FIONREAD:
1643                 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1644                         struct mbuf *mb;
1645                         IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1646                         IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1647                         for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1648                                 *(int *)data += mb->m_len;
1649                         IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1650                 } else
1651                         *(int *)data = 0;
1652                 break;
1653         case FIOSETOWN:
1654                 return (fsetown(*(int *)data, &tp->tun_sigio));
1655
1656         case FIOGETOWN:
1657                 *(int *)data = fgetown(&tp->tun_sigio);
1658                 return (0);
1659
1660         /* This is deprecated, FIOSETOWN should be used instead. */
1661         case TIOCSPGRP:
1662                 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1663
1664         /* This is deprecated, FIOGETOWN should be used instead. */
1665         case TIOCGPGRP:
1666                 *(int *)data = -fgetown(&tp->tun_sigio);
1667                 return (0);
1668
1669         default:
1670                 return (ENOTTY);
1671         }
1672         return (0);
1673 }
1674
1675 /*
1676  * The cdevsw read interface - reads a packet at a time, or at
1677  * least as much of a packet as can be read.
1678  */
1679 static  int
1680 tunread(struct cdev *dev, struct uio *uio, int flag)
1681 {
1682         struct tuntap_softc *tp = dev->si_drv1;
1683         struct ifnet    *ifp = TUN2IFP(tp);
1684         struct mbuf     *m;
1685         size_t          len;
1686         int             error = 0;
1687
1688         TUNDEBUG (ifp, "read\n");
1689         TUN_LOCK(tp);
1690         if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1691                 TUN_UNLOCK(tp);
1692                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1693                 return (EHOSTDOWN);
1694         }
1695
1696         tp->tun_flags &= ~TUN_RWAIT;
1697
1698         for (;;) {
1699                 IFQ_DEQUEUE(&ifp->if_snd, m);
1700                 if (m != NULL)
1701                         break;
1702                 if (flag & O_NONBLOCK) {
1703                         TUN_UNLOCK(tp);
1704                         return (EWOULDBLOCK);
1705                 }
1706                 tp->tun_flags |= TUN_RWAIT;
1707                 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1708                     "tunread", 0);
1709                 if (error != 0) {
1710                         TUN_UNLOCK(tp);
1711                         return (error);
1712                 }
1713         }
1714         TUN_UNLOCK(tp);
1715
1716         if ((tp->tun_flags & TUN_L2) != 0)
1717                 BPF_MTAP(ifp, m);
1718
1719         len = min(tp->tun_vhdrlen, uio->uio_resid);
1720         if (len > 0) {
1721                 struct virtio_net_hdr_mrg_rxbuf vhdr;
1722
1723                 bzero(&vhdr, sizeof(vhdr));
1724                 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1725                         m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1726                 }
1727
1728                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1729                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1730                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1731                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1732                     vhdr.hdr.csum_offset);
1733                 error = uiomove(&vhdr, len, uio);
1734         }
1735
1736         while (m && uio->uio_resid > 0 && error == 0) {
1737                 len = min(uio->uio_resid, m->m_len);
1738                 if (len != 0)
1739                         error = uiomove(mtod(m, void *), len, uio);
1740                 m = m_free(m);
1741         }
1742
1743         if (m) {
1744                 TUNDEBUG(ifp, "Dropping mbuf\n");
1745                 m_freem(m);
1746         }
1747         return (error);
1748 }
1749
1750 static int
1751 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1752             struct virtio_net_hdr_mrg_rxbuf *vhdr)
1753 {
1754         struct epoch_tracker et;
1755         struct ether_header *eh;
1756         struct ifnet *ifp;
1757
1758         ifp = TUN2IFP(tp);
1759
1760         /*
1761          * Only pass a unicast frame to ether_input(), if it would
1762          * actually have been received by non-virtual hardware.
1763          */
1764         if (m->m_len < sizeof(struct ether_header)) {
1765                 m_freem(m);
1766                 return (0);
1767         }
1768
1769         eh = mtod(m, struct ether_header *);
1770
1771         if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1772             !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1773             bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1774                 m_freem(m);
1775                 return (0);
1776         }
1777
1778         if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1779                 m_freem(m);
1780                 return (0);
1781         }
1782
1783         /* Pass packet up to parent. */
1784         CURVNET_SET(ifp->if_vnet);
1785         NET_EPOCH_ENTER(et);
1786         (*ifp->if_input)(ifp, m);
1787         NET_EPOCH_EXIT(et);
1788         CURVNET_RESTORE();
1789         /* ibytes are counted in parent */
1790         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1791         return (0);
1792 }
1793
1794 static int
1795 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1796 {
1797         struct epoch_tracker et;
1798         struct ifnet *ifp;
1799         int family, isr;
1800
1801         ifp = TUN2IFP(tp);
1802         /* Could be unlocked read? */
1803         TUN_LOCK(tp);
1804         if (tp->tun_flags & TUN_IFHEAD) {
1805                 TUN_UNLOCK(tp);
1806                 if (m->m_len < sizeof(family) &&
1807                 (m = m_pullup(m, sizeof(family))) == NULL)
1808                         return (ENOBUFS);
1809                 family = ntohl(*mtod(m, u_int32_t *));
1810                 m_adj(m, sizeof(family));
1811         } else {
1812                 TUN_UNLOCK(tp);
1813                 family = AF_INET;
1814         }
1815
1816         BPF_MTAP2(ifp, &family, sizeof(family), m);
1817
1818         switch (family) {
1819 #ifdef INET
1820         case AF_INET:
1821                 isr = NETISR_IP;
1822                 break;
1823 #endif
1824 #ifdef INET6
1825         case AF_INET6:
1826                 isr = NETISR_IPV6;
1827                 break;
1828 #endif
1829         default:
1830                 m_freem(m);
1831                 return (EAFNOSUPPORT);
1832         }
1833         random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1834         if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1835         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1836         CURVNET_SET(ifp->if_vnet);
1837         M_SETFIB(m, ifp->if_fib);
1838         NET_EPOCH_ENTER(et);
1839         netisr_dispatch(isr, m);
1840         NET_EPOCH_EXIT(et);
1841         CURVNET_RESTORE();
1842         return (0);
1843 }
1844
1845 /*
1846  * the cdevsw write interface - an atomic write is a packet - or else!
1847  */
1848 static  int
1849 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1850 {
1851         struct virtio_net_hdr_mrg_rxbuf vhdr;
1852         struct tuntap_softc *tp;
1853         struct ifnet    *ifp;
1854         struct mbuf     *m;
1855         uint32_t        mru;
1856         int             align, vhdrlen, error;
1857         bool            l2tun;
1858
1859         tp = dev->si_drv1;
1860         ifp = TUN2IFP(tp);
1861         TUNDEBUG(ifp, "tunwrite\n");
1862         if ((ifp->if_flags & IFF_UP) != IFF_UP)
1863                 /* ignore silently */
1864                 return (0);
1865
1866         if (uio->uio_resid == 0)
1867                 return (0);
1868
1869         l2tun = (tp->tun_flags & TUN_L2) != 0;
1870         mru = l2tun ? TAPMRU : TUNMRU;
1871         vhdrlen = tp->tun_vhdrlen;
1872         align = 0;
1873         if (l2tun) {
1874                 align = ETHER_ALIGN;
1875                 mru += vhdrlen;
1876         } else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1877                 mru += sizeof(uint32_t);        /* family */
1878         if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1879                 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1880                 return (EIO);
1881         }
1882
1883         if (vhdrlen > 0) {
1884                 error = uiomove(&vhdr, vhdrlen, uio);
1885                 if (error != 0)
1886                         return (error);
1887                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1888                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1889                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1890                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1891                     vhdr.hdr.csum_offset);
1892         }
1893
1894         if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1895                 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1896                 return (ENOBUFS);
1897         }
1898
1899         m->m_pkthdr.rcvif = ifp;
1900 #ifdef MAC
1901         mac_ifnet_create_mbuf(ifp, m);
1902 #endif
1903
1904         if (l2tun)
1905                 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1906
1907         return (tunwrite_l3(tp, m));
1908 }
1909
1910 /*
1911  * tunpoll - the poll interface, this is only useful on reads
1912  * really. The write detect always returns true, write never blocks
1913  * anyway, it either accepts the packet or drops it.
1914  */
1915 static  int
1916 tunpoll(struct cdev *dev, int events, struct thread *td)
1917 {
1918         struct tuntap_softc *tp = dev->si_drv1;
1919         struct ifnet    *ifp = TUN2IFP(tp);
1920         int             revents = 0;
1921
1922         TUNDEBUG(ifp, "tunpoll\n");
1923
1924         if (events & (POLLIN | POLLRDNORM)) {
1925                 IFQ_LOCK(&ifp->if_snd);
1926                 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1927                         TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1928                         revents |= events & (POLLIN | POLLRDNORM);
1929                 } else {
1930                         TUNDEBUG(ifp, "tunpoll waiting\n");
1931                         selrecord(td, &tp->tun_rsel);
1932                 }
1933                 IFQ_UNLOCK(&ifp->if_snd);
1934         }
1935         revents |= events & (POLLOUT | POLLWRNORM);
1936
1937         return (revents);
1938 }
1939
1940 /*
1941  * tunkqfilter - support for the kevent() system call.
1942  */
1943 static int
1944 tunkqfilter(struct cdev *dev, struct knote *kn)
1945 {
1946         struct tuntap_softc     *tp = dev->si_drv1;
1947         struct ifnet    *ifp = TUN2IFP(tp);
1948
1949         switch(kn->kn_filter) {
1950         case EVFILT_READ:
1951                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1952                     ifp->if_xname, dev2unit(dev));
1953                 kn->kn_fop = &tun_read_filterops;
1954                 break;
1955
1956         case EVFILT_WRITE:
1957                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1958                     ifp->if_xname, dev2unit(dev));
1959                 kn->kn_fop = &tun_write_filterops;
1960                 break;
1961
1962         default:
1963                 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1964                     ifp->if_xname, dev2unit(dev));
1965                 return(EINVAL);
1966         }
1967
1968         kn->kn_hook = tp;
1969         knlist_add(&tp->tun_rsel.si_note, kn, 0);
1970
1971         return (0);
1972 }
1973
1974 /*
1975  * Return true of there is data in the interface queue.
1976  */
1977 static int
1978 tunkqread(struct knote *kn, long hint)
1979 {
1980         int                     ret;
1981         struct tuntap_softc     *tp = kn->kn_hook;
1982         struct cdev             *dev = tp->tun_dev;
1983         struct ifnet    *ifp = TUN2IFP(tp);
1984
1985         if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
1986                 TUNDEBUG(ifp,
1987                     "%s have data in the queue.  Len = %d, minor = %#x\n",
1988                     ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
1989                 ret = 1;
1990         } else {
1991                 TUNDEBUG(ifp,
1992                     "%s waiting for data, minor = %#x\n", ifp->if_xname,
1993                     dev2unit(dev));
1994                 ret = 0;
1995         }
1996
1997         return (ret);
1998 }
1999
2000 /*
2001  * Always can write, always return MTU in kn->data.
2002  */
2003 static int
2004 tunkqwrite(struct knote *kn, long hint)
2005 {
2006         struct tuntap_softc     *tp = kn->kn_hook;
2007         struct ifnet    *ifp = TUN2IFP(tp);
2008
2009         kn->kn_data = ifp->if_mtu;
2010
2011         return (1);
2012 }
2013
2014 static void
2015 tunkqdetach(struct knote *kn)
2016 {
2017         struct tuntap_softc     *tp = kn->kn_hook;
2018
2019         knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2020 }