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 /*
 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, caddr_t);
 240 static int      tun_clone_destroy(struct if_clone *, struct ifnet *);
 241 static void     tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
 242
 243 static d_open_t         tunopen;
 244 static d_read_t         tunread;
 245 static d_write_t        tunwrite;
 246 static d_ioctl_t        tunioctl;
 247 static d_poll_t         tunpoll;
 248 static d_kqfilter_t     tunkqfilter;
 249
 250 static int              tunkqread(struct knote *, long);
 251 static int              tunkqwrite(struct knote *, long);
 252 static void             tunkqdetach(struct knote *);
 253
 254 static struct filterops tun_read_filterops = {
 255         .f_isfd =       1,
 256         .f_attach =     NULL,
 257         .f_detach =     tunkqdetach,
 258         .f_event =      tunkqread,
 259 };
 260
 261 static struct filterops tun_write_filterops = {
 262         .f_isfd =       1,
 263         .f_attach =     NULL,
 264         .f_detach =     tunkqdetach,
 265         .f_event =      tunkqwrite,
 266 };
 267
 268 static struct tuntap_driver {
 269         struct cdevsw            cdevsw;
 270         int                      ident_flags;
 271         struct unrhdr           *unrhdr;
 272         struct clonedevs        *clones;
 273         ifc_match_t             *clone_match_fn;
 274         ifc_create_t            *clone_create_fn;
 275         ifc_destroy_t           *clone_destroy_fn;
 276 } tuntap_drivers[] = {
 277         {
 278                 .ident_flags =  0,
 279                 .cdevsw =       {
 280                     .d_version =        D_VERSION,
 281                     .d_flags =          D_NEEDMINOR,
 282                     .d_open =           tunopen,
 283                     .d_read =           tunread,
 284                     .d_write =          tunwrite,
 285                     .d_ioctl =          tunioctl,
 286                     .d_poll =           tunpoll,
 287                     .d_kqfilter =       tunkqfilter,
 288                     .d_name =           tunname,
 289                 },
 290                 .clone_match_fn =       tun_clone_match,
 291                 .clone_create_fn =      tun_clone_create,
 292                 .clone_destroy_fn =     tun_clone_destroy,
 293         },
 294         {
 295                 .ident_flags =  TUN_L2,
 296                 .cdevsw =       {
 297                     .d_version =        D_VERSION,
 298                     .d_flags =          D_NEEDMINOR,
 299                     .d_open =           tunopen,
 300                     .d_read =           tunread,
 301                     .d_write =          tunwrite,
 302                     .d_ioctl =          tunioctl,
 303                     .d_poll =           tunpoll,
 304                     .d_kqfilter =       tunkqfilter,
 305                     .d_name =           tapname,
 306                 },
 307                 .clone_match_fn =       tap_clone_match,
 308                 .clone_create_fn =      tun_clone_create,
 309                 .clone_destroy_fn =     tun_clone_destroy,
 310         },
 311         {
 312                 .ident_flags =  TUN_L2 | TUN_VMNET,
 313                 .cdevsw =       {
 314                     .d_version =        D_VERSION,
 315                     .d_flags =          D_NEEDMINOR,
 316                     .d_open =           tunopen,
 317                     .d_read =           tunread,
 318                     .d_write =          tunwrite,
 319                     .d_ioctl =          tunioctl,
 320                     .d_poll =           tunpoll,
 321                     .d_kqfilter =       tunkqfilter,
 322                     .d_name =           vmnetname,
 323                 },
 324                 .clone_match_fn =       vmnet_clone_match,
 325                 .clone_create_fn =      tun_clone_create,
 326                 .clone_destroy_fn =     tun_clone_destroy,
 327         },
 328 };
 329
 330 struct tuntap_driver_cloner {
 331         SLIST_ENTRY(tuntap_driver_cloner)        link;
 332         struct tuntap_driver                    *drv;
 333         struct if_clone                         *cloner;
 334 };
 335
 336 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
 337     SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
 338
 339 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners)
 340
 341 /*
 342  * Mechanism for marking a tunnel device as busy so that we can safely do some
 343  * orthogonal operations (such as operations on devices) without racing against
 344  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
 345  * open, to be woken up when the condition is alleviated.
 346  */
 347 static int
 348 tun_busy_locked(struct tuntap_softc *tp)
 349 {
 350
 351         TUN_LOCK_ASSERT(tp);
 352         if ((tp->tun_flags & TUN_DYING) != 0) {
 353                 /*
 354                  * Perhaps unintuitive, but the device is busy going away.
 355                  * Other interpretations of EBUSY from tun_busy make little
 356                  * sense, since making a busy device even more busy doesn't
 357                  * sound like a problem.
 358                  */
 359                 return (EBUSY);
 360         }
 361
 362         ++tp->tun_busy;
 363         return (0);
 364 }
 365
 366 static void
 367 tun_unbusy_locked(struct tuntap_softc *tp)
 368 {
 369
 370         TUN_LOCK_ASSERT(tp);
 371         KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
 372
 373         --tp->tun_busy;
 374         /* Wake up anything that may be waiting on our busy tunnel. */
 375         if (tp->tun_busy == 0)
 376                 cv_broadcast(&tp->tun_cv);
 377 }
 378
 379 static int
 380 tun_busy(struct tuntap_softc *tp)
 381 {
 382         int ret;
 383
 384         TUN_LOCK(tp);
 385         ret = tun_busy_locked(tp);
 386         TUN_UNLOCK(tp);
 387         return (ret);
 388 }
 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
 480
 481 static int
 482 tun_clone_match(struct if_clone *ifc, const char *name)
 483 {
 484         int tunflags;
 485
 486         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 487                 if ((tunflags & TUN_L2) == 0)
 488                         return (1);
 489         }
 490
 491         return (0);
 492 }
 493
 494 static int
 495 tap_clone_match(struct if_clone *ifc, const char *name)
 496 {
 497         int tunflags;
 498
 499         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 500                 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
 501                         return (1);
 502         }
 503
 504         return (0);
 505 }
 506
 507 static int
 508 vmnet_clone_match(struct if_clone *ifc, const char *name)
 509 {
 510         int tunflags;
 511
 512         if (tuntap_name2info(name, NULL, &tunflags) == 0) {
 513                 if ((tunflags & TUN_VMNET) != 0)
 514                         return (1);
 515         }
 516
 517         return (0);
 518 }
 519
 520 static int
 521 tun_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
 522 {
 523         struct tuntap_driver *drv;
 524         struct cdev *dev;
 525         int err, i, tunflags, unit;
 526
 527         tunflags = 0;
 528         /* The name here tells us exactly what we're creating */
 529         err = tuntap_name2info(name, &unit, &tunflags);
 530         if (err != 0)
 531                 return (err);
 532
 533         drv = tuntap_driver_from_flags(tunflags);
 534         if (drv == NULL)
 535                 return (ENXIO);
 536
 537         if (unit != -1) {
 538                 /* If this unit number is still available that's okay. */
 539                 if (alloc_unr_specific(drv->unrhdr, unit) == -1)
 540                         return (EEXIST);
 541         } else {
 542                 unit = alloc_unr(drv->unrhdr);
 543         }
 544
 545         snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
 546
 547         /* find any existing device, or allocate new unit number */
 548         dev = NULL;
 549         i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
 550         /* No preexisting struct cdev *, create one */
 551         if (i != 0)
 552                 i = tun_create_device(drv, unit, NULL, &dev, name);
 553         if (i == 0)
 554                 tuncreate(dev);
 555
 556         return (i);
 557 }
 558
 559 static void
 560 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
 561     struct cdev **dev)
 562 {
 563         char devname[SPECNAMELEN + 1];
 564         struct tuntap_driver *drv;
 565         int append_unit, i, u, tunflags;
 566         bool mayclone;
 567
 568         if (*dev != NULL)
 569                 return;
 570
 571         tunflags = 0;
 572         CURVNET_SET(CRED_TO_VNET(cred));
 573         if (tuntap_name2info(name, &u, &tunflags) != 0)
 574                 goto out;       /* Not recognized */
 575
 576         if (u != -1 && u > IF_MAXUNIT)
 577                 goto out;       /* Unit number too high */
 578
 579         mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
 580         if ((tunflags & TUN_L2) != 0) {
 581                 /* tap/vmnet allow user open with a sysctl */
 582                 mayclone = (mayclone || tap_allow_uopen) && tapdclone;
 583         } else {
 584                 mayclone = mayclone && tundclone;
 585         }
 586
 587         /*
 588          * If tun cloning is enabled, only the superuser can create an
 589          * interface.
 590          */
 591         if (!mayclone)
 592                 goto out;
 593
 594         if (u == -1)
 595                 append_unit = 1;
 596         else
 597                 append_unit = 0;
 598
 599         drv = tuntap_driver_from_flags(tunflags);
 600         if (drv == NULL)
 601                 goto out;
 602
 603         /* find any existing device, or allocate new unit number */
 604         i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
 605         if (i) {
 606                 if (append_unit) {
 607                         namelen = snprintf(devname, sizeof(devname), "%s%d",
 608                             name, u);
 609                         name = devname;
 610                 }
 611
 612                 i = tun_create_device(drv, u, cred, dev, name);
 613         }
 614         if (i == 0)
 615                 if_clone_create(name, namelen, NULL);
 616 out:
 617         CURVNET_RESTORE();
 618 }
 619
 620 static void
 621 tun_destroy(struct tuntap_softc *tp)
 622 {
 623
 624         TUN_LOCK(tp);
 625         tp->tun_flags |= TUN_DYING;
 626         if (tp->tun_busy != 0)
 627                 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
 628         else
 629                 TUN_UNLOCK(tp);
 630
 631         CURVNET_SET(TUN2IFP(tp)->if_vnet);
 632
 633         /* destroy_dev will take care of any alias. */
 634         destroy_dev(tp->tun_dev);
 635         seldrain(&tp->tun_rsel);
 636         knlist_clear(&tp->tun_rsel.si_note, 0);
 637         knlist_destroy(&tp->tun_rsel.si_note);
 638         if ((tp->tun_flags & TUN_L2) != 0) {
 639                 ether_ifdetach(TUN2IFP(tp));
 640         } else {
 641                 bpfdetach(TUN2IFP(tp));
 642                 if_detach(TUN2IFP(tp));
 643         }
 644         sx_xlock(&tun_ioctl_sx);
 645         TUN2IFP(tp)->if_softc = NULL;
 646         sx_xunlock(&tun_ioctl_sx);
 647         free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
 648         if_free(TUN2IFP(tp));
 649         mtx_destroy(&tp->tun_mtx);
 650         cv_destroy(&tp->tun_cv);
 651         free(tp, M_TUN);
 652         CURVNET_RESTORE();
 653 }
 654
 655 static int
 656 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp)
 657 {
 658         struct tuntap_softc *tp = ifp->if_softc;
 659
 660         mtx_lock(&tunmtx);
 661         TAILQ_REMOVE(&tunhead, tp, tun_list);
 662         mtx_unlock(&tunmtx);
 663         tun_destroy(tp);
 664
 665         return (0);
 666 }
 667
 668 static void
 669 vnet_tun_init(const void *unused __unused)
 670 {
 671         struct tuntap_driver *drv;
 672         struct tuntap_driver_cloner *drvc;
 673         int i;
 674
 675         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 676                 drv = &tuntap_drivers[i];
 677                 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
 678
 679                 drvc->drv = drv;
 680                 drvc->cloner = if_clone_advanced(drv->cdevsw.d_name, 0,
 681                     drv->clone_match_fn, drv->clone_create_fn,
 682                     drv->clone_destroy_fn);
 683                 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
 684         };
 685 }
 686 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 687                 vnet_tun_init, NULL);
 688
 689 static void
 690 vnet_tun_uninit(const void *unused __unused)
 691 {
 692         struct tuntap_driver_cloner *drvc;
 693
 694         while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
 695                 drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
 696                 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
 697
 698                 if_clone_detach(drvc->cloner);
 699                 free(drvc, M_TUN);
 700         }
 701 }
 702 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
 703     vnet_tun_uninit, NULL);
 704
 705 static void
 706 tun_uninit(const void *unused __unused)
 707 {
 708         struct tuntap_driver *drv;
 709         struct tuntap_softc *tp;
 710         int i;
 711
 712         EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
 713         EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
 714         drain_dev_clone_events();
 715
 716         mtx_lock(&tunmtx);
 717         while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
 718                 TAILQ_REMOVE(&tunhead, tp, tun_list);
 719                 mtx_unlock(&tunmtx);
 720                 tun_destroy(tp);
 721                 mtx_lock(&tunmtx);
 722         }
 723         mtx_unlock(&tunmtx);
 724         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 725                 drv = &tuntap_drivers[i];
 726                 delete_unrhdr(drv->unrhdr);
 727                 clone_cleanup(&drv->clones);
 728         }
 729         mtx_destroy(&tunmtx);
 730 }
 731 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
 732
 733 static struct tuntap_driver *
 734 tuntap_driver_from_ifnet(const struct ifnet *ifp)
 735 {
 736         struct tuntap_driver *drv;
 737         int i;
 738
 739         if (ifp == NULL)
 740                 return (NULL);
 741
 742         for (i = 0; i < nitems(tuntap_drivers); ++i) {
 743                 drv = &tuntap_drivers[i];
 744                 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
 745                         return (drv);
 746         }
 747
 748         return (NULL);
 749 }
 750
 751 static int
 752 tuntapmodevent(module_t mod, int type, void *data)
 753 {
 754         struct tuntap_driver *drv;
 755         int i;
 756
 757         switch (type) {
 758         case MOD_LOAD:
 759                 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
 760                 for (i = 0; i < nitems(tuntap_drivers); ++i) {
 761                         drv = &tuntap_drivers[i];
 762                         clone_setup(&drv->clones);
 763                         drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
 764                 }
 765                 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
 766                    tunrename, 0, 1000);
 767                 if (arrival_tag == NULL)
 768                         return (ENOMEM);
 769                 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
 770                 if (clone_tag == NULL)
 771                         return (ENOMEM);
 772                 break;
 773         case MOD_UNLOAD:
 774                 /* See tun_uninit, so it's done after the vnet_sysuninit() */
 775                 break;
 776         default:
 777                 return EOPNOTSUPP;
 778         }
 779         return 0;
 780 }
 781
 782 static moduledata_t tuntap_mod = {
 783         "if_tuntap",
 784         tuntapmodevent,
 785         0
 786 };
 787
 788 /* We'll only ever have these two, so no need for a macro. */
 789 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
 790 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
 791
 792 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 793 MODULE_VERSION(if_tuntap, 1);
 794 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 795 MODULE_VERSION(if_tun, 1);
 796 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
 797 MODULE_VERSION(if_tap, 1);
 798
 799 static int
 800 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
 801     struct cdev **dev, const char *name)
 802 {
 803         struct make_dev_args args;
 804         struct tuntap_softc *tp;
 805         int error;
 806
 807         tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
 808         mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
 809         cv_init(&tp->tun_cv, "tun_condvar");
 810         tp->tun_flags = drv->ident_flags;
 811         tp->tun_drv = drv;
 812
 813         make_dev_args_init(&args);
 814         if (cr != NULL)
 815                 args.mda_flags = MAKEDEV_REF;
 816         args.mda_devsw = &drv->cdevsw;
 817         args.mda_cr = cr;
 818         args.mda_uid = UID_UUCP;
 819         args.mda_gid = GID_DIALER;
 820         args.mda_mode = 0600;
 821         args.mda_unit = unit;
 822         args.mda_si_drv1 = tp;
 823         error = make_dev_s(&args, dev, "%s", name);
 824         if (error != 0) {
 825                 free(tp, M_TUN);
 826                 return (error);
 827         }
 828
 829         KASSERT((*dev)->si_drv1 != NULL,
 830             ("Failed to set si_drv1 at %s creation", name));
 831         tp->tun_dev = *dev;
 832         knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
 833         mtx_lock(&tunmtx);
 834         TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
 835         mtx_unlock(&tunmtx);
 836         return (0);
 837 }
 838
 839 static void
 840 tunstart(struct ifnet *ifp)
 841 {
 842         struct tuntap_softc *tp = ifp->if_softc;
 843         struct mbuf *m;
 844
 845         TUNDEBUG(ifp, "starting\n");
 846         if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
 847                 IFQ_LOCK(&ifp->if_snd);
 848                 IFQ_POLL_NOLOCK(&ifp->if_snd, m);
 849                 if (m == NULL) {
 850                         IFQ_UNLOCK(&ifp->if_snd);
 851                         return;
 852                 }
 853                 IFQ_UNLOCK(&ifp->if_snd);
 854         }
 855
 856         TUN_LOCK(tp);
 857         if (tp->tun_flags & TUN_RWAIT) {
 858                 tp->tun_flags &= ~TUN_RWAIT;
 859                 wakeup(tp);
 860         }
 861         selwakeuppri(&tp->tun_rsel, PZERO + 1);
 862         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 863         if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
 864                 TUN_UNLOCK(tp);
 865                 pgsigio(&tp->tun_sigio, SIGIO, 0);
 866         } else
 867                 TUN_UNLOCK(tp);
 868 }
 869
 870 /*
 871  * tunstart_l2
 872  *
 873  * queue packets from higher level ready to put out
 874  */
 875 static void
 876 tunstart_l2(struct ifnet *ifp)
 877 {
 878         struct tuntap_softc     *tp = ifp->if_softc;
 879
 880         TUNDEBUG(ifp, "starting\n");
 881
 882         /*
 883          * do not junk pending output if we are in VMnet mode.
 884          * XXX: can this do any harm because of queue overflow?
 885          */
 886
 887         TUN_LOCK(tp);
 888         if (((tp->tun_flags & TUN_VMNET) == 0) &&
 889             ((tp->tun_flags & TUN_READY) != TUN_READY)) {
 890                 struct mbuf *m;
 891
 892                 /* Unlocked read. */
 893                 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
 894
 895                 for (;;) {
 896                         IF_DEQUEUE(&ifp->if_snd, m);
 897                         if (m != NULL) {
 898                                 m_freem(m);
 899                                 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 900                         } else
 901                                 break;
 902                 }
 903                 TUN_UNLOCK(tp);
 904
 905                 return;
 906         }
 907
 908         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
 909
 910         if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
 911                 if (tp->tun_flags & TUN_RWAIT) {
 912                         tp->tun_flags &= ~TUN_RWAIT;
 913                         wakeup(tp);
 914                 }
 915
 916                 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
 917                         TUN_UNLOCK(tp);
 918                         pgsigio(&tp->tun_sigio, SIGIO, 0);
 919                         TUN_LOCK(tp);
 920                 }
 921
 922                 selwakeuppri(&tp->tun_rsel, PZERO+1);
 923                 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
 924                 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
 925         }
 926
 927         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
 928         TUN_UNLOCK(tp);
 929 } /* tunstart_l2 */
 930
 931 /* XXX: should return an error code so it can fail. */
 932 static void
 933 tuncreate(struct cdev *dev)
 934 {
 935         struct tuntap_driver *drv;
 936         struct tuntap_softc *tp;
 937         struct ifnet *ifp;
 938         struct ether_addr eaddr;
 939         int iflags;
 940         u_char type;
 941
 942         tp = dev->si_drv1;
 943         KASSERT(tp != NULL,
 944             ("si_drv1 should have been initialized at creation"));
 945
 946         drv = tp->tun_drv;
 947         iflags = IFF_MULTICAST;
 948         if ((tp->tun_flags & TUN_L2) != 0) {
 949                 type = IFT_ETHER;
 950                 iflags |= IFF_BROADCAST | IFF_SIMPLEX;
 951         } else {
 952                 type = IFT_PPP;
 953                 iflags |= IFF_POINTOPOINT;
 954         }
 955         ifp = tp->tun_ifp = if_alloc(type);
 956         if (ifp == NULL)
 957                 panic("%s%d: failed to if_alloc() interface.\n",
 958                     drv->cdevsw.d_name, dev2unit(dev));
 959         ifp->if_softc = tp;
 960         if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
 961         ifp->if_ioctl = tunifioctl;
 962         ifp->if_flags = iflags;
 963         IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
 964         ifp->if_capabilities |= IFCAP_LINKSTATE;
 965         ifp->if_capenable |= IFCAP_LINKSTATE;
 966
 967         if ((tp->tun_flags & TUN_L2) != 0) {
 968                 ifp->if_mtu = ETHERMTU;
 969                 ifp->if_init = tunifinit;
 970                 ifp->if_start = tunstart_l2;
 971
 972                 ether_gen_addr(ifp, &eaddr);
 973                 ether_ifattach(ifp, eaddr.octet);
 974         } else {
 975                 ifp->if_mtu = TUNMTU;
 976                 ifp->if_start = tunstart;
 977                 ifp->if_output = tunoutput;
 978
 979                 ifp->if_snd.ifq_drv_maxlen = 0;
 980                 IFQ_SET_READY(&ifp->if_snd);
 981
 982                 if_attach(ifp);
 983                 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
 984         }
 985
 986         TUN_LOCK(tp);
 987         tp->tun_flags |= TUN_INITED;
 988         TUN_UNLOCK(tp);
 989
 990         TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
 991             ifp->if_xname, dev2unit(dev));
 992 }
 993
 994 static void
 995 tunrename(void *arg __unused, struct ifnet *ifp)
 996 {
 997         struct tuntap_softc *tp;
 998         int error;
 999
1000         if ((ifp->if_flags & IFF_RENAMING) == 0)
1001                 return;
1002
1003         if (tuntap_driver_from_ifnet(ifp) == NULL)
1004                 return;
1005
1006         /*
1007          * We need to grab the ioctl sx long enough to make sure the softc is
1008          * still there.  If it is, we can safely try to busy the tun device.
1009          * The busy may fail if the device is currently dying, in which case
1010          * we do nothing.  If it doesn't fail, the busy count stops the device
1011          * from dying until we've created the alias (that will then be
1012          * subsequently destroyed).
1013          */
1014         sx_xlock(&tun_ioctl_sx);
1015         tp = ifp->if_softc;
1016         if (tp == NULL) {
1017                 sx_xunlock(&tun_ioctl_sx);
1018                 return;
1019         }
1020         error = tun_busy(tp);
1021         sx_xunlock(&tun_ioctl_sx);
1022         if (error != 0)
1023                 return;
1024         if (tp->tun_alias != NULL) {
1025                 destroy_dev(tp->tun_alias);
1026                 tp->tun_alias = NULL;
1027         }
1028
1029         if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1030                 goto out;
1031
1032         /*
1033          * Failure's ok, aliases are created on a best effort basis.  If a
1034          * tun user/consumer decides to rename the interface to conflict with
1035          * another device (non-ifnet) on the system, we will assume they know
1036          * what they are doing.  make_dev_alias_p won't touch tun_alias on
1037          * failure, so we use it but ignore the return value.
1038          */
1039         make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1040             ifp->if_xname);
1041 out:
1042         tun_unbusy(tp);
1043 }
1044
1045 static int
1046 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1047 {
1048         struct ifnet    *ifp;
1049         struct tuntap_softc *tp;
1050         int error, tunflags;
1051
1052         tunflags = 0;
1053         CURVNET_SET(TD_TO_VNET(td));
1054         error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1055         if (error != 0) {
1056                 CURVNET_RESTORE();
1057                 return (error); /* Shouldn't happen */
1058         }
1059
1060         tp = dev->si_drv1;
1061         KASSERT(tp != NULL,
1062             ("si_drv1 should have been initialized at creation"));
1063
1064         TUN_LOCK(tp);
1065         if ((tp->tun_flags & TUN_INITED) == 0) {
1066                 TUN_UNLOCK(tp);
1067                 CURVNET_RESTORE();
1068                 return (ENXIO);
1069         }
1070         if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1071                 TUN_UNLOCK(tp);
1072                 CURVNET_RESTORE();
1073                 return (EBUSY);
1074         }
1075
1076         error = tun_busy_locked(tp);
1077         KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1078         ifp = TUN2IFP(tp);
1079
1080         if ((tp->tun_flags & TUN_L2) != 0) {
1081                 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1082                     sizeof(tp->tun_ether.octet));
1083
1084                 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1085                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1086
1087                 if (tapuponopen)
1088                         ifp->if_flags |= IFF_UP;
1089         }
1090
1091         tp->tun_pid = td->td_proc->p_pid;
1092         tp->tun_flags |= TUN_OPEN;
1093
1094         if_link_state_change(ifp, LINK_STATE_UP);
1095         TUNDEBUG(ifp, "open\n");
1096         TUN_UNLOCK(tp);
1097
1098         /*
1099          * This can fail with either ENOENT or EBUSY.  This is in the middle of
1100          * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1101          * the only cdevpriv dtor being set will be tundtor and the softc being
1102          * passed is constant for a given cdev.  We ignore the possible error
1103          * because of this as either "unlikely" or "not actually a problem."
1104          */
1105         (void)devfs_set_cdevpriv(tp, tundtor);
1106         CURVNET_RESTORE();
1107         return (0);
1108 }
1109
1110 /*
1111  * tundtor - tear down the device - mark i/f down & delete
1112  * routing info
1113  */
1114 static void
1115 tundtor(void *data)
1116 {
1117         struct proc *p;
1118         struct tuntap_softc *tp;
1119         struct ifnet *ifp;
1120         bool l2tun;
1121
1122         tp = data;
1123         p = curproc;
1124         ifp = TUN2IFP(tp);
1125
1126         TUN_LOCK(tp);
1127
1128         /*
1129          * Realistically, we can't be obstinate here.  This only means that the
1130          * tuntap device was closed out of order, and the last closer wasn't the
1131          * controller.  These are still good to know about, though, as software
1132          * should avoid multiple processes with a tuntap device open and
1133          * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1134          * parent).
1135          */
1136         if (p->p_pid != tp->tun_pid) {
1137                 log(LOG_INFO,
1138                     "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1139                     p->p_pid, p->p_comm, tp->tun_dev->si_name);
1140         }
1141
1142         /*
1143          * junk all pending output
1144          */
1145         CURVNET_SET(ifp->if_vnet);
1146
1147         l2tun = false;
1148         if ((tp->tun_flags & TUN_L2) != 0) {
1149                 l2tun = true;
1150                 IF_DRAIN(&ifp->if_snd);
1151         } else {
1152                 IFQ_PURGE(&ifp->if_snd);
1153         }
1154
1155         /* For vmnet, we won't do most of the address/route bits */
1156         if ((tp->tun_flags & TUN_VMNET) != 0 ||
1157             (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1158                 goto out;
1159
1160         if (ifp->if_flags & IFF_UP) {
1161                 TUN_UNLOCK(tp);
1162                 if_down(ifp);
1163                 TUN_LOCK(tp);
1164         }
1165
1166         /* Delete all addresses and routes which reference this interface. */
1167         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1168                 struct ifaddr *ifa;
1169
1170                 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1171                 TUN_UNLOCK(tp);
1172                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1173                         /* deal w/IPv4 PtP destination; unlocked read */
1174                         if (!l2tun && ifa->ifa_addr->sa_family == AF_INET) {
1175                                 rtinit(ifa, (int)RTM_DELETE,
1176                                     tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
1177                         } else {
1178                                 rtinit(ifa, (int)RTM_DELETE, 0);
1179                         }
1180                 }
1181                 if_purgeaddrs(ifp);
1182                 TUN_LOCK(tp);
1183         }
1184
1185 out:
1186         if_link_state_change(ifp, LINK_STATE_DOWN);
1187         CURVNET_RESTORE();
1188
1189         funsetown(&tp->tun_sigio);
1190         selwakeuppri(&tp->tun_rsel, PZERO + 1);
1191         KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1192         TUNDEBUG (ifp, "closed\n");
1193         tp->tun_flags &= ~TUN_OPEN;
1194         tp->tun_pid = 0;
1195         tun_vnethdr_set(ifp, 0);
1196
1197         tun_unbusy_locked(tp);
1198         TUN_UNLOCK(tp);
1199 }
1200
1201 static void
1202 tuninit(struct ifnet *ifp)
1203 {
1204         struct tuntap_softc *tp = ifp->if_softc;
1205 #ifdef INET
1206         struct epoch_tracker et;
1207         struct ifaddr *ifa;
1208 #endif
1209
1210         TUNDEBUG(ifp, "tuninit\n");
1211
1212         TUN_LOCK(tp);
1213         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1214         if ((tp->tun_flags & TUN_L2) == 0) {
1215                 ifp->if_flags |= IFF_UP;
1216                 getmicrotime(&ifp->if_lastchange);
1217 #ifdef INET
1218                 NET_EPOCH_ENTER(et);
1219                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1220                         if (ifa->ifa_addr->sa_family == AF_INET) {
1221                                 struct sockaddr_in *si;
1222
1223                                 si = (struct sockaddr_in *)ifa->ifa_dstaddr;
1224                                 if (si && si->sin_addr.s_addr) {
1225                                         tp->tun_flags |= TUN_DSTADDR;
1226                                         break;
1227                                 }
1228                         }
1229                 }
1230                 NET_EPOCH_EXIT(et);
1231 #endif
1232                 TUN_UNLOCK(tp);
1233         } else {
1234                 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1235                 TUN_UNLOCK(tp);
1236                 /* attempt to start output */
1237                 tunstart_l2(ifp);
1238         }
1239
1240 }
1241
1242 /*
1243  * Used only for l2 tunnel.
1244  */
1245 static void
1246 tunifinit(void *xtp)
1247 {
1248         struct tuntap_softc *tp;
1249
1250         tp = (struct tuntap_softc *)xtp;
1251         tuninit(tp->tun_ifp);
1252 }
1253
1254 /*
1255  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1256  * current value of ifp->if_capenable.
1257  */
1258 static void
1259 tun_caps_changed(struct ifnet *ifp)
1260 {
1261         uint64_t hwassist = 0;
1262
1263         TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1264         if (ifp->if_capenable & IFCAP_TXCSUM)
1265                 hwassist |= CSUM_TCP | CSUM_UDP;
1266         if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1267                 hwassist |= CSUM_TCP_IPV6
1268                     | CSUM_UDP_IPV6;
1269         if (ifp->if_capenable & IFCAP_TSO4)
1270                 hwassist |= CSUM_IP_TSO;
1271         if (ifp->if_capenable & IFCAP_TSO6)
1272                 hwassist |= CSUM_IP6_TSO;
1273         ifp->if_hwassist = hwassist;
1274 }
1275
1276 /*
1277  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1278  * if_capabilities and if_capenable as needed.
1279  */
1280 static void
1281 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1282 {
1283         struct tuntap_softc *tp = ifp->if_softc;
1284
1285         TUN_LOCK_ASSERT(tp);
1286
1287         if (tp->tun_vhdrlen == vhdrlen)
1288                 return;
1289
1290         /*
1291          * Update if_capabilities to reflect the
1292          * functionalities offered by the virtio-net
1293          * header.
1294          */
1295         if (vhdrlen != 0)
1296                 ifp->if_capabilities |=
1297                         TAP_VNET_HDR_CAPS;
1298         else
1299                 ifp->if_capabilities &=
1300                         ~TAP_VNET_HDR_CAPS;
1301         /*
1302          * Disable any capabilities that we don't
1303          * support anymore.
1304          */
1305         ifp->if_capenable &= ifp->if_capabilities;
1306         tun_caps_changed(ifp);
1307         tp->tun_vhdrlen = vhdrlen;
1308
1309         TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1310             vhdrlen, ifp->if_capabilities);
1311 }
1312
1313 /*
1314  * Process an ioctl request.
1315  */
1316 static int
1317 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1318 {
1319         struct ifreq *ifr = (struct ifreq *)data;
1320         struct tuntap_softc *tp;
1321         struct ifstat *ifs;
1322         struct ifmediareq       *ifmr;
1323         int             dummy, error = 0;
1324         bool            l2tun;
1325
1326         ifmr = NULL;
1327         sx_xlock(&tun_ioctl_sx);
1328         tp = ifp->if_softc;
1329         if (tp == NULL) {
1330                 error = ENXIO;
1331                 goto bad;
1332         }
1333         l2tun = (tp->tun_flags & TUN_L2) != 0;
1334         switch(cmd) {
1335         case SIOCGIFSTATUS:
1336                 ifs = (struct ifstat *)data;
1337                 TUN_LOCK(tp);
1338                 if (tp->tun_pid)
1339                         snprintf(ifs->ascii, sizeof(ifs->ascii),
1340                             "\tOpened by PID %d\n", tp->tun_pid);
1341                 else
1342                         ifs->ascii[0] = '\0';
1343                 TUN_UNLOCK(tp);
1344                 break;
1345         case SIOCSIFADDR:
1346                 if (l2tun)
1347                         error = ether_ioctl(ifp, cmd, data);
1348                 else
1349                         tuninit(ifp);
1350                 if (error == 0)
1351                     TUNDEBUG(ifp, "address set\n");
1352                 break;
1353         case SIOCSIFMTU:
1354                 ifp->if_mtu = ifr->ifr_mtu;
1355                 TUNDEBUG(ifp, "mtu set\n");
1356                 break;
1357         case SIOCSIFFLAGS:
1358         case SIOCADDMULTI:
1359         case SIOCDELMULTI:
1360                 break;
1361         case SIOCGIFMEDIA:
1362                 if (!l2tun) {
1363                         error = EINVAL;
1364                         break;
1365                 }
1366
1367                 ifmr = (struct ifmediareq *)data;
1368                 dummy = ifmr->ifm_count;
1369                 ifmr->ifm_count = 1;
1370                 ifmr->ifm_status = IFM_AVALID;
1371                 ifmr->ifm_active = IFM_ETHER;
1372                 if (tp->tun_flags & TUN_OPEN)
1373                         ifmr->ifm_status |= IFM_ACTIVE;
1374                 ifmr->ifm_current = ifmr->ifm_active;
1375                 if (dummy >= 1) {
1376                         int media = IFM_ETHER;
1377                         error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1378                 }
1379                 break;
1380         case SIOCSIFCAP:
1381                 TUN_LOCK(tp);
1382                 ifp->if_capenable = ifr->ifr_reqcap;
1383                 tun_caps_changed(ifp);
1384                 TUN_UNLOCK(tp);
1385                 VLAN_CAPABILITIES(ifp);
1386                 break;
1387         default:
1388                 if (l2tun) {
1389                         error = ether_ioctl(ifp, cmd, data);
1390                 } else {
1391                         error = EINVAL;
1392                 }
1393         }
1394 bad:
1395         sx_xunlock(&tun_ioctl_sx);
1396         return (error);
1397 }
1398
1399 /*
1400  * tunoutput - queue packets from higher level ready to put out.
1401  */
1402 static int
1403 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1404     struct route *ro)
1405 {
1406         struct tuntap_softc *tp = ifp->if_softc;
1407         u_short cached_tun_flags;
1408         int error;
1409         u_int32_t af;
1410
1411         TUNDEBUG (ifp, "tunoutput\n");
1412
1413 #ifdef MAC
1414         error = mac_ifnet_check_transmit(ifp, m0);
1415         if (error) {
1416                 m_freem(m0);
1417                 return (error);
1418         }
1419 #endif
1420
1421         /* Could be unlocked read? */
1422         TUN_LOCK(tp);
1423         cached_tun_flags = tp->tun_flags;
1424         TUN_UNLOCK(tp);
1425         if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1426                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1427                 m_freem (m0);
1428                 return (EHOSTDOWN);
1429         }
1430
1431         if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1432                 m_freem (m0);
1433                 return (EHOSTDOWN);
1434         }
1435
1436         /* BPF writes need to be handled specially. */
1437         if (dst->sa_family == AF_UNSPEC)
1438                 bcopy(dst->sa_data, &af, sizeof(af));
1439         else
1440                 af = dst->sa_family;
1441
1442         if (bpf_peers_present(ifp->if_bpf))
1443                 bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1444
1445         /* prepend sockaddr? this may abort if the mbuf allocation fails */
1446         if (cached_tun_flags & TUN_LMODE) {
1447                 /* allocate space for sockaddr */
1448                 M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1449
1450                 /* if allocation failed drop packet */
1451                 if (m0 == NULL) {
1452                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1453                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1454                         return (ENOBUFS);
1455                 } else {
1456                         bcopy(dst, m0->m_data, dst->sa_len);
1457                 }
1458         }
1459
1460         if (cached_tun_flags & TUN_IFHEAD) {
1461                 /* Prepend the address family */
1462                 M_PREPEND(m0, 4, M_NOWAIT);
1463
1464                 /* if allocation failed drop packet */
1465                 if (m0 == NULL) {
1466                         if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1467                         if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1468                         return (ENOBUFS);
1469                 } else
1470                         *(u_int32_t *)m0->m_data = htonl(af);
1471         } else {
1472 #ifdef INET
1473                 if (af != AF_INET)
1474 #endif
1475                 {
1476                         m_freem(m0);
1477                         return (EAFNOSUPPORT);
1478                 }
1479         }
1480
1481         error = (ifp->if_transmit)(ifp, m0);
1482         if (error)
1483                 return (ENOBUFS);
1484         if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1485         return (0);
1486 }
1487
1488 /*
1489  * the cdevsw interface is now pretty minimal.
1490  */
1491 static  int
1492 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1493     struct thread *td)
1494 {
1495         struct ifreq ifr, *ifrp;
1496         struct tuntap_softc *tp = dev->si_drv1;
1497         struct ifnet *ifp = TUN2IFP(tp);
1498         struct tuninfo *tunp;
1499         int error, iflags, ival;
1500         bool    l2tun;
1501
1502         l2tun = (tp->tun_flags & TUN_L2) != 0;
1503         if (l2tun) {
1504                 /* tap specific ioctls */
1505                 switch(cmd) {
1506                 /* VMware/VMnet port ioctl's */
1507 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1508     defined(COMPAT_FREEBSD4)
1509                 case _IO('V', 0):
1510                         ival = IOCPARM_IVAL(data);
1511                         data = (caddr_t)&ival;
1512                         /* FALLTHROUGH */
1513 #endif
1514                 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1515                         iflags = *(int *)data;
1516                         iflags &= TUN_VMIO_FLAG_MASK;
1517                         iflags &= ~IFF_CANTCHANGE;
1518                         iflags |= IFF_UP;
1519
1520                         TUN_LOCK(tp);
1521                         ifp->if_flags = iflags |
1522                             (ifp->if_flags & IFF_CANTCHANGE);
1523                         TUN_UNLOCK(tp);
1524
1525                         return (0);
1526                 case SIOCGIFADDR:       /* get MAC address of the remote side */
1527                         TUN_LOCK(tp);
1528                         bcopy(&tp->tun_ether.octet, data,
1529                             sizeof(tp->tun_ether.octet));
1530                         TUN_UNLOCK(tp);
1531
1532                         return (0);
1533                 case SIOCSIFADDR:       /* set MAC address of the remote side */
1534                         TUN_LOCK(tp);
1535                         bcopy(data, &tp->tun_ether.octet,
1536                             sizeof(tp->tun_ether.octet));
1537                         TUN_UNLOCK(tp);
1538
1539                         return (0);
1540                 case TAPSVNETHDR:
1541                         ival = *(int *)data;
1542                         if (ival != 0 &&
1543                             ival != sizeof(struct virtio_net_hdr) &&
1544                             ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1545                                 return (EINVAL);
1546                         }
1547                         TUN_LOCK(tp);
1548                         tun_vnethdr_set(ifp, ival);
1549                         TUN_UNLOCK(tp);
1550
1551                         return (0);
1552                 case TAPGVNETHDR:
1553                         TUN_LOCK(tp);
1554                         *(int *)data = tp->tun_vhdrlen;
1555                         TUN_UNLOCK(tp);
1556
1557                         return (0);
1558                 }
1559
1560                 /* Fall through to the common ioctls if unhandled */
1561         } else {
1562                 switch (cmd) {
1563                 case TUNSLMODE:
1564                         TUN_LOCK(tp);
1565                         if (*(int *)data) {
1566                                 tp->tun_flags |= TUN_LMODE;
1567                                 tp->tun_flags &= ~TUN_IFHEAD;
1568                         } else
1569                                 tp->tun_flags &= ~TUN_LMODE;
1570                         TUN_UNLOCK(tp);
1571
1572                         return (0);
1573                 case TUNSIFHEAD:
1574                         TUN_LOCK(tp);
1575                         if (*(int *)data) {
1576                                 tp->tun_flags |= TUN_IFHEAD;
1577                                 tp->tun_flags &= ~TUN_LMODE;
1578                         } else
1579                                 tp->tun_flags &= ~TUN_IFHEAD;
1580                         TUN_UNLOCK(tp);
1581
1582                         return (0);
1583                 case TUNGIFHEAD:
1584                         TUN_LOCK(tp);
1585                         *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1586                         TUN_UNLOCK(tp);
1587
1588                         return (0);
1589                 case TUNSIFMODE:
1590                         /* deny this if UP */
1591                         if (TUN2IFP(tp)->if_flags & IFF_UP)
1592                                 return (EBUSY);
1593
1594                         switch (*(int *)data & ~IFF_MULTICAST) {
1595                         case IFF_POINTOPOINT:
1596                         case IFF_BROADCAST:
1597                                 TUN_LOCK(tp);
1598                                 TUN2IFP(tp)->if_flags &=
1599                                     ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1600                                 TUN2IFP(tp)->if_flags |= *(int *)data;
1601                                 TUN_UNLOCK(tp);
1602
1603                                 break;
1604                         default:
1605                                 return (EINVAL);
1606                         }
1607
1608                         return (0);
1609                 case TUNSIFPID:
1610                         TUN_LOCK(tp);
1611                         tp->tun_pid = curthread->td_proc->p_pid;
1612                         TUN_UNLOCK(tp);
1613
1614                         return (0);
1615                 }
1616                 /* Fall through to the common ioctls if unhandled */
1617         }
1618
1619         switch (cmd) {
1620         case TUNGIFNAME:
1621                 ifrp = (struct ifreq *)data;
1622                 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1623
1624                 return (0);
1625         case TUNSIFINFO:
1626                 tunp = (struct tuninfo *)data;
1627                 if (TUN2IFP(tp)->if_type != tunp->type)
1628                         return (EPROTOTYPE);
1629                 TUN_LOCK(tp);
1630                 if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1631                         strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1632                         ifr.ifr_mtu = tunp->mtu;
1633                         CURVNET_SET(TUN2IFP(tp)->if_vnet);
1634                         error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1635                             (caddr_t)&ifr, td);
1636                         CURVNET_RESTORE();
1637                         if (error) {
1638                                 TUN_UNLOCK(tp);
1639                                 return (error);
1640                         }
1641                 }
1642                 TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1643                 TUN_UNLOCK(tp);
1644                 break;
1645         case TUNGIFINFO:
1646                 tunp = (struct tuninfo *)data;
1647                 TUN_LOCK(tp);
1648                 tunp->mtu = TUN2IFP(tp)->if_mtu;
1649                 tunp->type = TUN2IFP(tp)->if_type;
1650                 tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1651                 TUN_UNLOCK(tp);
1652                 break;
1653         case TUNSDEBUG:
1654                 tundebug = *(int *)data;
1655                 break;
1656         case TUNGDEBUG:
1657                 *(int *)data = tundebug;
1658                 break;
1659         case FIONBIO:
1660                 break;
1661         case FIOASYNC:
1662                 TUN_LOCK(tp);
1663                 if (*(int *)data)
1664                         tp->tun_flags |= TUN_ASYNC;
1665                 else
1666                         tp->tun_flags &= ~TUN_ASYNC;
1667                 TUN_UNLOCK(tp);
1668                 break;
1669         case FIONREAD:
1670                 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1671                         struct mbuf *mb;
1672                         IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1673                         IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1674                         for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1675                                 *(int *)data += mb->m_len;
1676                         IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1677                 } else
1678                         *(int *)data = 0;
1679                 break;
1680         case FIOSETOWN:
1681                 return (fsetown(*(int *)data, &tp->tun_sigio));
1682
1683         case FIOGETOWN:
1684                 *(int *)data = fgetown(&tp->tun_sigio);
1685                 return (0);
1686
1687         /* This is deprecated, FIOSETOWN should be used instead. */
1688         case TIOCSPGRP:
1689                 return (fsetown(-(*(int *)data), &tp->tun_sigio));
1690
1691         /* This is deprecated, FIOGETOWN should be used instead. */
1692         case TIOCGPGRP:
1693                 *(int *)data = -fgetown(&tp->tun_sigio);
1694                 return (0);
1695
1696         default:
1697                 return (ENOTTY);
1698         }
1699         return (0);
1700 }
1701
1702 /*
1703  * The cdevsw read interface - reads a packet at a time, or at
1704  * least as much of a packet as can be read.
1705  */
1706 static  int
1707 tunread(struct cdev *dev, struct uio *uio, int flag)
1708 {
1709         struct tuntap_softc *tp = dev->si_drv1;
1710         struct ifnet    *ifp = TUN2IFP(tp);
1711         struct mbuf     *m;
1712         size_t          len;
1713         int             error = 0;
1714
1715         TUNDEBUG (ifp, "read\n");
1716         TUN_LOCK(tp);
1717         if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1718                 TUN_UNLOCK(tp);
1719                 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1720                 return (EHOSTDOWN);
1721         }
1722
1723         tp->tun_flags &= ~TUN_RWAIT;
1724
1725         for (;;) {
1726                 IFQ_DEQUEUE(&ifp->if_snd, m);
1727                 if (m != NULL)
1728                         break;
1729                 if (flag & O_NONBLOCK) {
1730                         TUN_UNLOCK(tp);
1731                         return (EWOULDBLOCK);
1732                 }
1733                 tp->tun_flags |= TUN_RWAIT;
1734                 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1735                     "tunread", 0);
1736                 if (error != 0) {
1737                         TUN_UNLOCK(tp);
1738                         return (error);
1739                 }
1740         }
1741         TUN_UNLOCK(tp);
1742
1743         if ((tp->tun_flags & TUN_L2) != 0)
1744                 BPF_MTAP(ifp, m);
1745
1746         len = min(tp->tun_vhdrlen, uio->uio_resid);
1747         if (len > 0) {
1748                 struct virtio_net_hdr_mrg_rxbuf vhdr;
1749
1750                 bzero(&vhdr, sizeof(vhdr));
1751                 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1752                         m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1753                 }
1754
1755                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1756                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1757                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1758                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1759                     vhdr.hdr.csum_offset);
1760                 error = uiomove(&vhdr, len, uio);
1761         }
1762
1763         while (m && uio->uio_resid > 0 && error == 0) {
1764                 len = min(uio->uio_resid, m->m_len);
1765                 if (len != 0)
1766                         error = uiomove(mtod(m, void *), len, uio);
1767                 m = m_free(m);
1768         }
1769
1770         if (m) {
1771                 TUNDEBUG(ifp, "Dropping mbuf\n");
1772                 m_freem(m);
1773         }
1774         return (error);
1775 }
1776
1777 static int
1778 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1779             struct virtio_net_hdr_mrg_rxbuf *vhdr)
1780 {
1781         struct epoch_tracker et;
1782         struct ether_header *eh;
1783         struct ifnet *ifp;
1784
1785         ifp = TUN2IFP(tp);
1786
1787         /*
1788          * Only pass a unicast frame to ether_input(), if it would
1789          * actually have been received by non-virtual hardware.
1790          */
1791         if (m->m_len < sizeof(struct ether_header)) {
1792                 m_freem(m);
1793                 return (0);
1794         }
1795
1796         eh = mtod(m, struct ether_header *);
1797
1798         if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1799             !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1800             bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1801                 m_freem(m);
1802                 return (0);
1803         }
1804
1805         if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1806                 m_freem(m);
1807                 return (0);
1808         }
1809
1810         /* Pass packet up to parent. */
1811         CURVNET_SET(ifp->if_vnet);
1812         NET_EPOCH_ENTER(et);
1813         (*ifp->if_input)(ifp, m);
1814         NET_EPOCH_EXIT(et);
1815         CURVNET_RESTORE();
1816         /* ibytes are counted in parent */
1817         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1818         return (0);
1819 }
1820
1821 static int
1822 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1823 {
1824         struct epoch_tracker et;
1825         struct ifnet *ifp;
1826         int family, isr;
1827
1828         ifp = TUN2IFP(tp);
1829         /* Could be unlocked read? */
1830         TUN_LOCK(tp);
1831         if (tp->tun_flags & TUN_IFHEAD) {
1832                 TUN_UNLOCK(tp);
1833                 if (m->m_len < sizeof(family) &&
1834                 (m = m_pullup(m, sizeof(family))) == NULL)
1835                         return (ENOBUFS);
1836                 family = ntohl(*mtod(m, u_int32_t *));
1837                 m_adj(m, sizeof(family));
1838         } else {
1839                 TUN_UNLOCK(tp);
1840                 family = AF_INET;
1841         }
1842
1843         BPF_MTAP2(ifp, &family, sizeof(family), m);
1844
1845         switch (family) {
1846 #ifdef INET
1847         case AF_INET:
1848                 isr = NETISR_IP;
1849                 break;
1850 #endif
1851 #ifdef INET6
1852         case AF_INET6:
1853                 isr = NETISR_IPV6;
1854                 break;
1855 #endif
1856         default:
1857                 m_freem(m);
1858                 return (EAFNOSUPPORT);
1859         }
1860         random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1861         if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1862         if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1863         CURVNET_SET(ifp->if_vnet);
1864         M_SETFIB(m, ifp->if_fib);
1865         NET_EPOCH_ENTER(et);
1866         netisr_dispatch(isr, m);
1867         NET_EPOCH_EXIT(et);
1868         CURVNET_RESTORE();
1869         return (0);
1870 }
1871
1872 /*
1873  * the cdevsw write interface - an atomic write is a packet - or else!
1874  */
1875 static  int
1876 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1877 {
1878         struct virtio_net_hdr_mrg_rxbuf vhdr;
1879         struct tuntap_softc *tp;
1880         struct ifnet    *ifp;
1881         struct mbuf     *m;
1882         uint32_t        mru;
1883         int             align, vhdrlen, error;
1884         bool            l2tun;
1885
1886         tp = dev->si_drv1;
1887         ifp = TUN2IFP(tp);
1888         TUNDEBUG(ifp, "tunwrite\n");
1889         if ((ifp->if_flags & IFF_UP) != IFF_UP)
1890                 /* ignore silently */
1891                 return (0);
1892
1893         if (uio->uio_resid == 0)
1894                 return (0);
1895
1896         l2tun = (tp->tun_flags & TUN_L2) != 0;
1897         mru = l2tun ? TAPMRU : TUNMRU;
1898         vhdrlen = tp->tun_vhdrlen;
1899         align = 0;
1900         if (l2tun) {
1901                 align = ETHER_ALIGN;
1902                 mru += vhdrlen;
1903         } else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1904                 mru += sizeof(uint32_t);        /* family */
1905         if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1906                 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1907                 return (EIO);
1908         }
1909
1910         if (vhdrlen > 0) {
1911                 error = uiomove(&vhdr, vhdrlen, uio);
1912                 if (error != 0)
1913                         return (error);
1914                 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1915                     "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1916                     vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1917                     vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1918                     vhdr.hdr.csum_offset);
1919         }
1920
1921         if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1922                 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1923                 return (ENOBUFS);
1924         }
1925
1926         m->m_pkthdr.rcvif = ifp;
1927 #ifdef MAC
1928         mac_ifnet_create_mbuf(ifp, m);
1929 #endif
1930
1931         if (l2tun)
1932                 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1933
1934         return (tunwrite_l3(tp, m));
1935 }
1936
1937 /*
1938  * tunpoll - the poll interface, this is only useful on reads
1939  * really. The write detect always returns true, write never blocks
1940  * anyway, it either accepts the packet or drops it.
1941  */
1942 static  int
1943 tunpoll(struct cdev *dev, int events, struct thread *td)
1944 {
1945         struct tuntap_softc *tp = dev->si_drv1;
1946         struct ifnet    *ifp = TUN2IFP(tp);
1947         int             revents = 0;
1948
1949         TUNDEBUG(ifp, "tunpoll\n");
1950
1951         if (events & (POLLIN | POLLRDNORM)) {
1952                 IFQ_LOCK(&ifp->if_snd);
1953                 if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1954                         TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1955                         revents |= events & (POLLIN | POLLRDNORM);
1956                 } else {
1957                         TUNDEBUG(ifp, "tunpoll waiting\n");
1958                         selrecord(td, &tp->tun_rsel);
1959                 }
1960                 IFQ_UNLOCK(&ifp->if_snd);
1961         }
1962         revents |= events & (POLLOUT | POLLWRNORM);
1963
1964         return (revents);
1965 }
1966
1967 /*
1968  * tunkqfilter - support for the kevent() system call.
1969  */
1970 static int
1971 tunkqfilter(struct cdev *dev, struct knote *kn)
1972 {
1973         struct tuntap_softc     *tp = dev->si_drv1;
1974         struct ifnet    *ifp = TUN2IFP(tp);
1975
1976         switch(kn->kn_filter) {
1977         case EVFILT_READ:
1978                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1979                     ifp->if_xname, dev2unit(dev));
1980                 kn->kn_fop = &tun_read_filterops;
1981                 break;
1982
1983         case EVFILT_WRITE:
1984                 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1985                     ifp->if_xname, dev2unit(dev));
1986                 kn->kn_fop = &tun_write_filterops;
1987                 break;
1988
1989         default:
1990                 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1991                     ifp->if_xname, dev2unit(dev));
1992                 return(EINVAL);
1993         }
1994
1995         kn->kn_hook = tp;
1996         knlist_add(&tp->tun_rsel.si_note, kn, 0);
1997
1998         return (0);
1999 }
2000
2001 /*
2002  * Return true of there is data in the interface queue.
2003  */
2004 static int
2005 tunkqread(struct knote *kn, long hint)
2006 {
2007         int                     ret;
2008         struct tuntap_softc     *tp = kn->kn_hook;
2009         struct cdev             *dev = tp->tun_dev;
2010         struct ifnet    *ifp = TUN2IFP(tp);
2011
2012         if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2013                 TUNDEBUG(ifp,
2014                     "%s have data in the queue.  Len = %d, minor = %#x\n",
2015                     ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2016                 ret = 1;
2017         } else {
2018                 TUNDEBUG(ifp,
2019                     "%s waiting for data, minor = %#x\n", ifp->if_xname,
2020                     dev2unit(dev));
2021                 ret = 0;
2022         }
2023
2024         return (ret);
2025 }
2026
2027 /*
2028  * Always can write, always return MTU in kn->data.
2029  */
2030 static int
2031 tunkqwrite(struct knote *kn, long hint)
2032 {
2033         struct tuntap_softc     *tp = kn->kn_hook;
2034         struct ifnet    *ifp = TUN2IFP(tp);
2035
2036         kn->kn_data = ifp->if_mtu;
2037
2038         return (1);
2039 }
2040
2041 static void
2042 tunkqdetach(struct knote *kn)
2043 {
2044         struct tuntap_softc     *tp = kn->kn_hook;
2045
2046         knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2047 }