1 /* $NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $ */ 2 /*- 3 * SPDX-License-Identifier: BSD-2-Clause 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 47 #include "opt_inet.h" 48 #include "opt_inet6.h" 49 50 #include <sys/param.h> 51 #include <sys/lock.h> 52 #include <sys/priv.h> 53 #include <sys/proc.h> 54 #include <sys/systm.h> 55 #include <sys/jail.h> 56 #include <sys/mbuf.h> 57 #include <sys/module.h> 58 #include <sys/socket.h> 59 #include <sys/eventhandler.h> 60 #include <sys/fcntl.h> 61 #include <sys/filio.h> 62 #include <sys/sockio.h> 63 #include <sys/sx.h> 64 #include <sys/syslog.h> 65 #include <sys/ttycom.h> 66 #include <sys/poll.h> 67 #include <sys/selinfo.h> 68 #include <sys/signalvar.h> 69 #include <sys/filedesc.h> 70 #include <sys/kernel.h> 71 #include <sys/sysctl.h> 72 #include <sys/conf.h> 73 #include <sys/uio.h> 74 #include <sys/malloc.h> 75 #include <sys/random.h> 76 #include <sys/ctype.h> 77 78 #include <net/ethernet.h> 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/if_clone.h> 82 #include <net/if_dl.h> 83 #include <net/if_media.h> 84 #include <net/if_private.h> 85 #include <net/if_types.h> 86 #include <net/if_vlan_var.h> 87 #include <net/netisr.h> 88 #include <net/route.h> 89 #include <net/vnet.h> 90 #include <netinet/in.h> 91 #ifdef INET 92 #include <netinet/ip.h> 93 #endif 94 #ifdef INET6 95 #include <netinet/ip6.h> 96 #include <netinet6/ip6_var.h> 97 #endif 98 #include <netinet/udp.h> 99 #include <netinet/tcp.h> 100 #include <net/bpf.h> 101 #include <net/if_tap.h> 102 #include <net/if_tun.h> 103 104 #include <dev/virtio/network/virtio_net.h> 105 106 #include <sys/queue.h> 107 #include <sys/condvar.h> 108 #include <security/mac/mac_framework.h> 109 110 struct tuntap_driver; 111 112 /* 113 * tun_list is protected by global tunmtx. Other mutable fields are 114 * protected by tun->tun_mtx, or by their owning subsystem. tun_dev is 115 * static for the duration of a tunnel interface. 116 */ 117 struct tuntap_softc { 118 TAILQ_ENTRY(tuntap_softc) tun_list; 119 struct cdev *tun_alias; 120 struct cdev *tun_dev; 121 u_short tun_flags; /* misc flags */ 122 #define TUN_OPEN 0x0001 123 #define TUN_INITED 0x0002 124 #define TUN_UNUSED1 0x0008 125 #define TUN_UNUSED2 0x0010 126 #define TUN_LMODE 0x0020 127 #define TUN_RWAIT 0x0040 128 #define TUN_ASYNC 0x0080 129 #define TUN_IFHEAD 0x0100 130 #define TUN_DYING 0x0200 131 #define TUN_L2 0x0400 132 #define TUN_VMNET 0x0800 133 134 #define TUN_DRIVER_IDENT_MASK (TUN_L2 | TUN_VMNET) 135 #define TUN_READY (TUN_OPEN | TUN_INITED) 136 137 pid_t tun_pid; /* owning pid */ 138 struct ifnet *tun_ifp; /* the interface */ 139 struct sigio *tun_sigio; /* async I/O info */ 140 struct tuntap_driver *tun_drv; /* appropriate driver */ 141 struct selinfo tun_rsel; /* read select */ 142 struct mtx tun_mtx; /* softc field mutex */ 143 struct cv tun_cv; /* for ref'd dev destroy */ 144 struct ether_addr tun_ether; /* remote address */ 145 int tun_busy; /* busy count */ 146 int tun_vhdrlen; /* virtio-net header length */ 147 }; 148 #define TUN2IFP(sc) ((sc)->tun_ifp) 149 150 #define TUNDEBUG if (tundebug) if_printf 151 152 #define TUN_LOCK(tp) mtx_lock(&(tp)->tun_mtx) 153 #define TUN_UNLOCK(tp) mtx_unlock(&(tp)->tun_mtx) 154 #define TUN_LOCK_ASSERT(tp) mtx_assert(&(tp)->tun_mtx, MA_OWNED); 155 156 #define TUN_VMIO_FLAG_MASK 0x0fff 157 158 /* 159 * Interface capabilities of a tap device that supports the virtio-net 160 * header. 161 */ 162 #define TAP_VNET_HDR_CAPS (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6 \ 163 | IFCAP_VLAN_HWCSUM \ 164 | IFCAP_TSO | IFCAP_LRO \ 165 | IFCAP_VLAN_HWTSO) 166 167 #define TAP_ALL_OFFLOAD (CSUM_TSO | CSUM_TCP | CSUM_UDP |\ 168 CSUM_TCP_IPV6 | CSUM_UDP_IPV6) 169 170 /* 171 * All mutable global variables in if_tun are locked using tunmtx, with 172 * the exception of tundebug, which is used unlocked, and the drivers' *clones, 173 * which are static after setup. 174 */ 175 static struct mtx tunmtx; 176 static eventhandler_tag arrival_tag; 177 static eventhandler_tag clone_tag; 178 static const char tunname[] = "tun"; 179 static const char tapname[] = "tap"; 180 static const char vmnetname[] = "vmnet"; 181 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface"); 182 static int tundebug = 0; 183 static int tundclone = 1; 184 static int tap_allow_uopen = 0; /* allow user devfs cloning */ 185 static int tapuponopen = 0; /* IFF_UP on open() */ 186 static int tapdclone = 1; /* enable devfs cloning */ 187 188 static TAILQ_HEAD(,tuntap_softc) tunhead = TAILQ_HEAD_INITIALIZER(tunhead); 189 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, ""); 190 191 static struct sx tun_ioctl_sx; 192 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl"); 193 194 SYSCTL_DECL(_net_link); 195 /* tun */ 196 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 197 "IP tunnel software network interface"); 198 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0, 199 "Enable legacy devfs interface creation"); 200 201 /* tap */ 202 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 203 "Ethernet tunnel software network interface"); 204 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0, 205 "Enable legacy devfs interface creation for all users"); 206 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0, 207 "Bring interface up when /dev/tap is opened"); 208 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0, 209 "Enable legacy devfs interface creation"); 210 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, ""); 211 212 static int tun_create_device(struct tuntap_driver *drv, int unit, 213 struct ucred *cr, struct cdev **dev, const char *name); 214 static int tun_busy_locked(struct tuntap_softc *tp); 215 static void tun_unbusy_locked(struct tuntap_softc *tp); 216 static int tun_busy(struct tuntap_softc *tp); 217 static void tun_unbusy(struct tuntap_softc *tp); 218 219 static int tuntap_name2info(const char *name, int *unit, int *flags); 220 static void tunclone(void *arg, struct ucred *cred, char *name, 221 int namelen, struct cdev **dev); 222 static void tuncreate(struct cdev *dev); 223 static void tundtor(void *data); 224 static void tunrename(void *arg, struct ifnet *ifp); 225 static int tunifioctl(struct ifnet *, u_long, caddr_t); 226 static void tuninit(struct ifnet *); 227 static void tunifinit(void *xtp); 228 static int tuntapmodevent(module_t, int, void *); 229 static int tunoutput(struct ifnet *, struct mbuf *, 230 const struct sockaddr *, struct route *ro); 231 static void tunstart(struct ifnet *); 232 static void tunstart_l2(struct ifnet *); 233 234 static int tun_clone_match(struct if_clone *ifc, const char *name); 235 static int tap_clone_match(struct if_clone *ifc, const char *name); 236 static int vmnet_clone_match(struct if_clone *ifc, const char *name); 237 static int tun_clone_create(struct if_clone *, char *, size_t, 238 struct ifc_data *, struct ifnet **); 239 static int tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t); 240 static void tun_vnethdr_set(struct ifnet *ifp, int vhdrlen); 241 242 static d_open_t tunopen; 243 static d_read_t tunread; 244 static d_write_t tunwrite; 245 static d_ioctl_t tunioctl; 246 static d_poll_t tunpoll; 247 static d_kqfilter_t tunkqfilter; 248 249 static int tunkqread(struct knote *, long); 250 static int tunkqwrite(struct knote *, long); 251 static void tunkqdetach(struct knote *); 252 253 static struct filterops tun_read_filterops = { 254 .f_isfd = 1, 255 .f_attach = NULL, 256 .f_detach = tunkqdetach, 257 .f_event = tunkqread, 258 }; 259 260 static struct filterops tun_write_filterops = { 261 .f_isfd = 1, 262 .f_attach = NULL, 263 .f_detach = tunkqdetach, 264 .f_event = tunkqwrite, 265 }; 266 267 static struct tuntap_driver { 268 struct cdevsw cdevsw; 269 int ident_flags; 270 struct unrhdr *unrhdr; 271 struct clonedevs *clones; 272 ifc_match_f *clone_match_fn; 273 ifc_create_f *clone_create_fn; 274 ifc_destroy_f *clone_destroy_fn; 275 } tuntap_drivers[] = { 276 { 277 .ident_flags = 0, 278 .cdevsw = { 279 .d_version = D_VERSION, 280 .d_flags = D_NEEDMINOR, 281 .d_open = tunopen, 282 .d_read = tunread, 283 .d_write = tunwrite, 284 .d_ioctl = tunioctl, 285 .d_poll = tunpoll, 286 .d_kqfilter = tunkqfilter, 287 .d_name = tunname, 288 }, 289 .clone_match_fn = tun_clone_match, 290 .clone_create_fn = tun_clone_create, 291 .clone_destroy_fn = tun_clone_destroy, 292 }, 293 { 294 .ident_flags = TUN_L2, 295 .cdevsw = { 296 .d_version = D_VERSION, 297 .d_flags = D_NEEDMINOR, 298 .d_open = tunopen, 299 .d_read = tunread, 300 .d_write = tunwrite, 301 .d_ioctl = tunioctl, 302 .d_poll = tunpoll, 303 .d_kqfilter = tunkqfilter, 304 .d_name = tapname, 305 }, 306 .clone_match_fn = tap_clone_match, 307 .clone_create_fn = tun_clone_create, 308 .clone_destroy_fn = tun_clone_destroy, 309 }, 310 { 311 .ident_flags = TUN_L2 | TUN_VMNET, 312 .cdevsw = { 313 .d_version = D_VERSION, 314 .d_flags = D_NEEDMINOR, 315 .d_open = tunopen, 316 .d_read = tunread, 317 .d_write = tunwrite, 318 .d_ioctl = tunioctl, 319 .d_poll = tunpoll, 320 .d_kqfilter = tunkqfilter, 321 .d_name = vmnetname, 322 }, 323 .clone_match_fn = vmnet_clone_match, 324 .clone_create_fn = tun_clone_create, 325 .clone_destroy_fn = tun_clone_destroy, 326 }, 327 }; 328 329 struct tuntap_driver_cloner { 330 SLIST_ENTRY(tuntap_driver_cloner) link; 331 struct tuntap_driver *drv; 332 struct if_clone *cloner; 333 }; 334 335 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) = 336 SLIST_HEAD_INITIALIZER(tuntap_driver_cloners); 337 338 #define V_tuntap_driver_cloners VNET(tuntap_driver_cloners) 339 340 /* 341 * Mechanism for marking a tunnel device as busy so that we can safely do some 342 * orthogonal operations (such as operations on devices) without racing against 343 * tun_destroy. tun_destroy will wait on the condvar if we're at all busy or 344 * open, to be woken up when the condition is alleviated. 345 */ 346 static int 347 tun_busy_locked(struct tuntap_softc *tp) 348 { 349 350 TUN_LOCK_ASSERT(tp); 351 if ((tp->tun_flags & TUN_DYING) != 0) { 352 /* 353 * Perhaps unintuitive, but the device is busy going away. 354 * Other interpretations of EBUSY from tun_busy make little 355 * sense, since making a busy device even more busy doesn't 356 * sound like a problem. 357 */ 358 return (EBUSY); 359 } 360 361 ++tp->tun_busy; 362 return (0); 363 } 364 365 static void 366 tun_unbusy_locked(struct tuntap_softc *tp) 367 { 368 369 TUN_LOCK_ASSERT(tp); 370 KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel")); 371 372 --tp->tun_busy; 373 /* Wake up anything that may be waiting on our busy tunnel. */ 374 if (tp->tun_busy == 0) 375 cv_broadcast(&tp->tun_cv); 376 } 377 378 static int 379 tun_busy(struct tuntap_softc *tp) 380 { 381 int ret; 382 383 TUN_LOCK(tp); 384 ret = tun_busy_locked(tp); 385 TUN_UNLOCK(tp); 386 return (ret); 387 } 388 389 static void 390 tun_unbusy(struct tuntap_softc *tp) 391 { 392 393 TUN_LOCK(tp); 394 tun_unbusy_locked(tp); 395 TUN_UNLOCK(tp); 396 } 397 398 /* 399 * Sets unit and/or flags given the device name. Must be called with correct 400 * vnet context. 401 */ 402 static int 403 tuntap_name2info(const char *name, int *outunit, int *outflags) 404 { 405 struct tuntap_driver *drv; 406 struct tuntap_driver_cloner *drvc; 407 char *dname; 408 int flags, unit; 409 bool found; 410 411 if (name == NULL) 412 return (EINVAL); 413 414 /* 415 * Needed for dev_stdclone, but dev_stdclone will not modify, it just 416 * wants to be able to pass back a char * through the second param. We 417 * will always set that as NULL here, so we'll fake it. 418 */ 419 dname = __DECONST(char *, name); 420 found = false; 421 422 KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners), 423 ("tuntap_driver_cloners failed to initialize")); 424 SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) { 425 KASSERT(drvc->drv != NULL, 426 ("tuntap_driver_cloners entry not properly initialized")); 427 drv = drvc->drv; 428 429 if (strcmp(name, drv->cdevsw.d_name) == 0) { 430 found = true; 431 unit = -1; 432 flags = drv->ident_flags; 433 break; 434 } 435 436 if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) { 437 found = true; 438 flags = drv->ident_flags; 439 break; 440 } 441 } 442 443 if (!found) 444 return (ENXIO); 445 446 if (outunit != NULL) 447 *outunit = unit; 448 if (outflags != NULL) 449 *outflags = flags; 450 return (0); 451 } 452 453 /* 454 * Get driver information from a set of flags specified. Masks the identifying 455 * part of the flags and compares it against all of the available 456 * tuntap_drivers. Must be called with correct vnet context. 457 */ 458 static struct tuntap_driver * 459 tuntap_driver_from_flags(int tun_flags) 460 { 461 struct tuntap_driver *drv; 462 struct tuntap_driver_cloner *drvc; 463 464 KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners), 465 ("tuntap_driver_cloners failed to initialize")); 466 SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) { 467 KASSERT(drvc->drv != NULL, 468 ("tuntap_driver_cloners entry not properly initialized")); 469 drv = drvc->drv; 470 if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->ident_flags) 471 return (drv); 472 } 473 474 return (NULL); 475 } 476 477 static int 478 tun_clone_match(struct if_clone *ifc, const char *name) 479 { 480 int tunflags; 481 482 if (tuntap_name2info(name, NULL, &tunflags) == 0) { 483 if ((tunflags & TUN_L2) == 0) 484 return (1); 485 } 486 487 return (0); 488 } 489 490 static int 491 tap_clone_match(struct if_clone *ifc, const char *name) 492 { 493 int tunflags; 494 495 if (tuntap_name2info(name, NULL, &tunflags) == 0) { 496 if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2) 497 return (1); 498 } 499 500 return (0); 501 } 502 503 static int 504 vmnet_clone_match(struct if_clone *ifc, const char *name) 505 { 506 int tunflags; 507 508 if (tuntap_name2info(name, NULL, &tunflags) == 0) { 509 if ((tunflags & TUN_VMNET) != 0) 510 return (1); 511 } 512 513 return (0); 514 } 515 516 static int 517 tun_clone_create(struct if_clone *ifc, char *name, size_t len, 518 struct ifc_data *ifd, struct ifnet **ifpp) 519 { 520 struct tuntap_driver *drv; 521 struct cdev *dev; 522 int err, i, tunflags, unit; 523 524 tunflags = 0; 525 /* The name here tells us exactly what we're creating */ 526 err = tuntap_name2info(name, &unit, &tunflags); 527 if (err != 0) 528 return (err); 529 530 drv = tuntap_driver_from_flags(tunflags); 531 if (drv == NULL) 532 return (ENXIO); 533 534 if (unit != -1) { 535 /* If this unit number is still available that's okay. */ 536 if (alloc_unr_specific(drv->unrhdr, unit) == -1) 537 return (EEXIST); 538 } else { 539 unit = alloc_unr(drv->unrhdr); 540 } 541 542 snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit); 543 544 /* find any existing device, or allocate new unit number */ 545 dev = NULL; 546 i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0); 547 /* No preexisting struct cdev *, create one */ 548 if (i != 0) 549 i = tun_create_device(drv, unit, NULL, &dev, name); 550 if (i == 0) { 551 dev_ref(dev); 552 tuncreate(dev); 553 struct tuntap_softc *tp = dev->si_drv1; 554 *ifpp = tp->tun_ifp; 555 } 556 557 return (i); 558 } 559 560 static void 561 tunclone(void *arg, struct ucred *cred, char *name, int namelen, 562 struct cdev **dev) 563 { 564 char devname[SPECNAMELEN + 1]; 565 struct tuntap_driver *drv; 566 int append_unit, i, u, tunflags; 567 bool mayclone; 568 569 if (*dev != NULL) 570 return; 571 572 tunflags = 0; 573 CURVNET_SET(CRED_TO_VNET(cred)); 574 if (tuntap_name2info(name, &u, &tunflags) != 0) 575 goto out; /* Not recognized */ 576 577 if (u != -1 && u > IF_MAXUNIT) 578 goto out; /* Unit number too high */ 579 580 mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0; 581 if ((tunflags & TUN_L2) != 0) { 582 /* tap/vmnet allow user open with a sysctl */ 583 mayclone = (mayclone || tap_allow_uopen) && tapdclone; 584 } else { 585 mayclone = mayclone && tundclone; 586 } 587 588 /* 589 * If tun cloning is enabled, only the superuser can create an 590 * interface. 591 */ 592 if (!mayclone) 593 goto out; 594 595 if (u == -1) 596 append_unit = 1; 597 else 598 append_unit = 0; 599 600 drv = tuntap_driver_from_flags(tunflags); 601 if (drv == NULL) 602 goto out; 603 604 /* find any existing device, or allocate new unit number */ 605 i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0); 606 if (i) { 607 if (append_unit) { 608 namelen = snprintf(devname, sizeof(devname), "%s%d", 609 name, u); 610 name = devname; 611 } 612 613 i = tun_create_device(drv, u, cred, dev, name); 614 } 615 if (i == 0) { 616 dev_ref(*dev); 617 if_clone_create(name, namelen, NULL); 618 } 619 out: 620 CURVNET_RESTORE(); 621 } 622 623 static void 624 tun_destroy(struct tuntap_softc *tp) 625 { 626 627 TUN_LOCK(tp); 628 tp->tun_flags |= TUN_DYING; 629 if (tp->tun_busy != 0) 630 cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx); 631 else 632 TUN_UNLOCK(tp); 633 634 CURVNET_SET(TUN2IFP(tp)->if_vnet); 635 636 /* destroy_dev will take care of any alias. */ 637 destroy_dev(tp->tun_dev); 638 seldrain(&tp->tun_rsel); 639 knlist_clear(&tp->tun_rsel.si_note, 0); 640 knlist_destroy(&tp->tun_rsel.si_note); 641 if ((tp->tun_flags & TUN_L2) != 0) { 642 ether_ifdetach(TUN2IFP(tp)); 643 } else { 644 bpfdetach(TUN2IFP(tp)); 645 if_detach(TUN2IFP(tp)); 646 } 647 sx_xlock(&tun_ioctl_sx); 648 TUN2IFP(tp)->if_softc = NULL; 649 sx_xunlock(&tun_ioctl_sx); 650 free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit); 651 if_free(TUN2IFP(tp)); 652 mtx_destroy(&tp->tun_mtx); 653 cv_destroy(&tp->tun_cv); 654 free(tp, M_TUN); 655 CURVNET_RESTORE(); 656 } 657 658 static int 659 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags) 660 { 661 struct tuntap_softc *tp = ifp->if_softc; 662 663 mtx_lock(&tunmtx); 664 TAILQ_REMOVE(&tunhead, tp, tun_list); 665 mtx_unlock(&tunmtx); 666 tun_destroy(tp); 667 668 return (0); 669 } 670 671 static void 672 vnet_tun_init(const void *unused __unused) 673 { 674 struct tuntap_driver *drv; 675 struct tuntap_driver_cloner *drvc; 676 int i; 677 678 for (i = 0; i < nitems(tuntap_drivers); ++i) { 679 drv = &tuntap_drivers[i]; 680 drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO); 681 682 drvc->drv = drv; 683 struct if_clone_addreq req = { 684 .match_f = drv->clone_match_fn, 685 .create_f = drv->clone_create_fn, 686 .destroy_f = drv->clone_destroy_fn, 687 }; 688 drvc->cloner = ifc_attach_cloner(drv->cdevsw.d_name, &req); 689 SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link); 690 }; 691 } 692 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY, 693 vnet_tun_init, NULL); 694 695 static void 696 vnet_tun_uninit(const void *unused __unused) 697 { 698 struct tuntap_driver_cloner *drvc; 699 700 while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) { 701 drvc = SLIST_FIRST(&V_tuntap_driver_cloners); 702 SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link); 703 704 if_clone_detach(drvc->cloner); 705 free(drvc, M_TUN); 706 } 707 } 708 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, 709 vnet_tun_uninit, NULL); 710 711 static void 712 tun_uninit(const void *unused __unused) 713 { 714 struct tuntap_driver *drv; 715 struct tuntap_softc *tp; 716 int i; 717 718 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag); 719 EVENTHANDLER_DEREGISTER(dev_clone, clone_tag); 720 721 mtx_lock(&tunmtx); 722 while ((tp = TAILQ_FIRST(&tunhead)) != NULL) { 723 TAILQ_REMOVE(&tunhead, tp, tun_list); 724 mtx_unlock(&tunmtx); 725 tun_destroy(tp); 726 mtx_lock(&tunmtx); 727 } 728 mtx_unlock(&tunmtx); 729 for (i = 0; i < nitems(tuntap_drivers); ++i) { 730 drv = &tuntap_drivers[i]; 731 delete_unrhdr(drv->unrhdr); 732 clone_cleanup(&drv->clones); 733 } 734 mtx_destroy(&tunmtx); 735 } 736 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL); 737 738 static struct tuntap_driver * 739 tuntap_driver_from_ifnet(const struct ifnet *ifp) 740 { 741 struct tuntap_driver *drv; 742 int i; 743 744 if (ifp == NULL) 745 return (NULL); 746 747 for (i = 0; i < nitems(tuntap_drivers); ++i) { 748 drv = &tuntap_drivers[i]; 749 if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0) 750 return (drv); 751 } 752 753 return (NULL); 754 } 755 756 static int 757 tuntapmodevent(module_t mod, int type, void *data) 758 { 759 struct tuntap_driver *drv; 760 int i; 761 762 switch (type) { 763 case MOD_LOAD: 764 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF); 765 for (i = 0; i < nitems(tuntap_drivers); ++i) { 766 drv = &tuntap_drivers[i]; 767 clone_setup(&drv->clones); 768 drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx); 769 } 770 arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event, 771 tunrename, 0, 1000); 772 if (arrival_tag == NULL) 773 return (ENOMEM); 774 clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000); 775 if (clone_tag == NULL) 776 return (ENOMEM); 777 break; 778 case MOD_UNLOAD: 779 /* See tun_uninit, so it's done after the vnet_sysuninit() */ 780 break; 781 default: 782 return EOPNOTSUPP; 783 } 784 return 0; 785 } 786 787 static moduledata_t tuntap_mod = { 788 "if_tuntap", 789 tuntapmodevent, 790 0 791 }; 792 793 /* We'll only ever have these two, so no need for a macro. */ 794 static moduledata_t tun_mod = { "if_tun", NULL, 0 }; 795 static moduledata_t tap_mod = { "if_tap", NULL, 0 }; 796 797 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 798 MODULE_VERSION(if_tuntap, 1); 799 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 800 MODULE_VERSION(if_tun, 1); 801 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 802 MODULE_VERSION(if_tap, 1); 803 804 static int 805 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr, 806 struct cdev **dev, const char *name) 807 { 808 struct make_dev_args args; 809 struct tuntap_softc *tp; 810 int error; 811 812 tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO); 813 mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF); 814 cv_init(&tp->tun_cv, "tun_condvar"); 815 tp->tun_flags = drv->ident_flags; 816 tp->tun_drv = drv; 817 818 make_dev_args_init(&args); 819 if (cr != NULL) 820 args.mda_flags = MAKEDEV_REF; 821 args.mda_devsw = &drv->cdevsw; 822 args.mda_cr = cr; 823 args.mda_uid = UID_UUCP; 824 args.mda_gid = GID_DIALER; 825 args.mda_mode = 0600; 826 args.mda_unit = unit; 827 args.mda_si_drv1 = tp; 828 error = make_dev_s(&args, dev, "%s", name); 829 if (error != 0) { 830 free(tp, M_TUN); 831 return (error); 832 } 833 834 KASSERT((*dev)->si_drv1 != NULL, 835 ("Failed to set si_drv1 at %s creation", name)); 836 tp->tun_dev = *dev; 837 knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx); 838 mtx_lock(&tunmtx); 839 TAILQ_INSERT_TAIL(&tunhead, tp, tun_list); 840 mtx_unlock(&tunmtx); 841 return (0); 842 } 843 844 static void 845 tunstart(struct ifnet *ifp) 846 { 847 struct tuntap_softc *tp = ifp->if_softc; 848 struct mbuf *m; 849 850 TUNDEBUG(ifp, "starting\n"); 851 if (ALTQ_IS_ENABLED(&ifp->if_snd)) { 852 IFQ_LOCK(&ifp->if_snd); 853 IFQ_POLL_NOLOCK(&ifp->if_snd, m); 854 if (m == NULL) { 855 IFQ_UNLOCK(&ifp->if_snd); 856 return; 857 } 858 IFQ_UNLOCK(&ifp->if_snd); 859 } 860 861 TUN_LOCK(tp); 862 if (tp->tun_flags & TUN_RWAIT) { 863 tp->tun_flags &= ~TUN_RWAIT; 864 wakeup(tp); 865 } 866 selwakeuppri(&tp->tun_rsel, PZERO + 1); 867 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0); 868 if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) { 869 TUN_UNLOCK(tp); 870 pgsigio(&tp->tun_sigio, SIGIO, 0); 871 } else 872 TUN_UNLOCK(tp); 873 } 874 875 /* 876 * tunstart_l2 877 * 878 * queue packets from higher level ready to put out 879 */ 880 static void 881 tunstart_l2(struct ifnet *ifp) 882 { 883 struct tuntap_softc *tp = ifp->if_softc; 884 885 TUNDEBUG(ifp, "starting\n"); 886 887 /* 888 * do not junk pending output if we are in VMnet mode. 889 * XXX: can this do any harm because of queue overflow? 890 */ 891 892 TUN_LOCK(tp); 893 if (((tp->tun_flags & TUN_VMNET) == 0) && 894 ((tp->tun_flags & TUN_READY) != TUN_READY)) { 895 struct mbuf *m; 896 897 /* Unlocked read. */ 898 TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags); 899 900 for (;;) { 901 IF_DEQUEUE(&ifp->if_snd, m); 902 if (m != NULL) { 903 m_freem(m); 904 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 905 } else 906 break; 907 } 908 TUN_UNLOCK(tp); 909 910 return; 911 } 912 913 ifp->if_drv_flags |= IFF_DRV_OACTIVE; 914 915 if (!IFQ_IS_EMPTY(&ifp->if_snd)) { 916 if (tp->tun_flags & TUN_RWAIT) { 917 tp->tun_flags &= ~TUN_RWAIT; 918 wakeup(tp); 919 } 920 921 if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) { 922 TUN_UNLOCK(tp); 923 pgsigio(&tp->tun_sigio, SIGIO, 0); 924 TUN_LOCK(tp); 925 } 926 927 selwakeuppri(&tp->tun_rsel, PZERO+1); 928 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0); 929 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */ 930 } 931 932 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 933 TUN_UNLOCK(tp); 934 } /* tunstart_l2 */ 935 936 static int 937 tap_transmit(struct ifnet *ifp, struct mbuf *m) 938 { 939 int error; 940 941 BPF_MTAP(ifp, m); 942 IFQ_HANDOFF(ifp, m, error); 943 return (error); 944 } 945 946 /* XXX: should return an error code so it can fail. */ 947 static void 948 tuncreate(struct cdev *dev) 949 { 950 struct tuntap_driver *drv; 951 struct tuntap_softc *tp; 952 struct ifnet *ifp; 953 struct ether_addr eaddr; 954 int iflags; 955 u_char type; 956 957 tp = dev->si_drv1; 958 KASSERT(tp != NULL, 959 ("si_drv1 should have been initialized at creation")); 960 961 drv = tp->tun_drv; 962 iflags = IFF_MULTICAST; 963 if ((tp->tun_flags & TUN_L2) != 0) { 964 type = IFT_ETHER; 965 iflags |= IFF_BROADCAST | IFF_SIMPLEX; 966 } else { 967 type = IFT_PPP; 968 iflags |= IFF_POINTOPOINT; 969 } 970 ifp = tp->tun_ifp = if_alloc(type); 971 if (ifp == NULL) 972 panic("%s%d: failed to if_alloc() interface.\n", 973 drv->cdevsw.d_name, dev2unit(dev)); 974 ifp->if_softc = tp; 975 if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev)); 976 ifp->if_ioctl = tunifioctl; 977 ifp->if_flags = iflags; 978 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 979 ifp->if_capabilities |= IFCAP_LINKSTATE; 980 if ((tp->tun_flags & TUN_L2) != 0) 981 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6; 982 ifp->if_capenable |= IFCAP_LINKSTATE; 983 984 if ((tp->tun_flags & TUN_L2) != 0) { 985 ifp->if_init = tunifinit; 986 ifp->if_start = tunstart_l2; 987 ifp->if_transmit = tap_transmit; 988 ifp->if_qflush = if_qflush; 989 990 ether_gen_addr(ifp, &eaddr); 991 ether_ifattach(ifp, eaddr.octet); 992 } else { 993 ifp->if_mtu = TUNMTU; 994 ifp->if_start = tunstart; 995 ifp->if_output = tunoutput; 996 997 ifp->if_snd.ifq_drv_maxlen = 0; 998 IFQ_SET_READY(&ifp->if_snd); 999 1000 if_attach(ifp); 1001 bpfattach(ifp, DLT_NULL, sizeof(u_int32_t)); 1002 } 1003 1004 TUN_LOCK(tp); 1005 tp->tun_flags |= TUN_INITED; 1006 TUN_UNLOCK(tp); 1007 1008 TUNDEBUG(ifp, "interface %s is created, minor = %#x\n", 1009 ifp->if_xname, dev2unit(dev)); 1010 } 1011 1012 static void 1013 tunrename(void *arg __unused, struct ifnet *ifp) 1014 { 1015 struct tuntap_softc *tp; 1016 int error; 1017 1018 if ((ifp->if_flags & IFF_RENAMING) == 0) 1019 return; 1020 1021 if (tuntap_driver_from_ifnet(ifp) == NULL) 1022 return; 1023 1024 /* 1025 * We need to grab the ioctl sx long enough to make sure the softc is 1026 * still there. If it is, we can safely try to busy the tun device. 1027 * The busy may fail if the device is currently dying, in which case 1028 * we do nothing. If it doesn't fail, the busy count stops the device 1029 * from dying until we've created the alias (that will then be 1030 * subsequently destroyed). 1031 */ 1032 sx_xlock(&tun_ioctl_sx); 1033 tp = ifp->if_softc; 1034 if (tp == NULL) { 1035 sx_xunlock(&tun_ioctl_sx); 1036 return; 1037 } 1038 error = tun_busy(tp); 1039 sx_xunlock(&tun_ioctl_sx); 1040 if (error != 0) 1041 return; 1042 if (tp->tun_alias != NULL) { 1043 destroy_dev(tp->tun_alias); 1044 tp->tun_alias = NULL; 1045 } 1046 1047 if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0) 1048 goto out; 1049 1050 /* 1051 * Failure's ok, aliases are created on a best effort basis. If a 1052 * tun user/consumer decides to rename the interface to conflict with 1053 * another device (non-ifnet) on the system, we will assume they know 1054 * what they are doing. make_dev_alias_p won't touch tun_alias on 1055 * failure, so we use it but ignore the return value. 1056 */ 1057 make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s", 1058 ifp->if_xname); 1059 out: 1060 tun_unbusy(tp); 1061 } 1062 1063 static int 1064 tunopen(struct cdev *dev, int flag, int mode, struct thread *td) 1065 { 1066 struct ifnet *ifp; 1067 struct tuntap_softc *tp; 1068 int error __diagused, tunflags; 1069 1070 tunflags = 0; 1071 CURVNET_SET(TD_TO_VNET(td)); 1072 error = tuntap_name2info(dev->si_name, NULL, &tunflags); 1073 if (error != 0) { 1074 CURVNET_RESTORE(); 1075 return (error); /* Shouldn't happen */ 1076 } 1077 1078 tp = dev->si_drv1; 1079 KASSERT(tp != NULL, 1080 ("si_drv1 should have been initialized at creation")); 1081 1082 TUN_LOCK(tp); 1083 if ((tp->tun_flags & TUN_INITED) == 0) { 1084 TUN_UNLOCK(tp); 1085 CURVNET_RESTORE(); 1086 return (ENXIO); 1087 } 1088 if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) { 1089 TUN_UNLOCK(tp); 1090 CURVNET_RESTORE(); 1091 return (EBUSY); 1092 } 1093 1094 error = tun_busy_locked(tp); 1095 KASSERT(error == 0, ("Must be able to busy an unopen tunnel")); 1096 ifp = TUN2IFP(tp); 1097 1098 if ((tp->tun_flags & TUN_L2) != 0) { 1099 bcopy(IF_LLADDR(ifp), tp->tun_ether.octet, 1100 sizeof(tp->tun_ether.octet)); 1101 1102 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1103 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1104 1105 if (tapuponopen) 1106 ifp->if_flags |= IFF_UP; 1107 } 1108 1109 tp->tun_pid = td->td_proc->p_pid; 1110 tp->tun_flags |= TUN_OPEN; 1111 1112 if_link_state_change(ifp, LINK_STATE_UP); 1113 TUNDEBUG(ifp, "open\n"); 1114 TUN_UNLOCK(tp); 1115 1116 /* 1117 * This can fail with either ENOENT or EBUSY. This is in the middle of 1118 * d_open, so ENOENT should not be possible. EBUSY is possible, but 1119 * the only cdevpriv dtor being set will be tundtor and the softc being 1120 * passed is constant for a given cdev. We ignore the possible error 1121 * because of this as either "unlikely" or "not actually a problem." 1122 */ 1123 (void)devfs_set_cdevpriv(tp, tundtor); 1124 CURVNET_RESTORE(); 1125 return (0); 1126 } 1127 1128 /* 1129 * tundtor - tear down the device - mark i/f down & delete 1130 * routing info 1131 */ 1132 static void 1133 tundtor(void *data) 1134 { 1135 struct proc *p; 1136 struct tuntap_softc *tp; 1137 struct ifnet *ifp; 1138 bool l2tun; 1139 1140 tp = data; 1141 p = curproc; 1142 ifp = TUN2IFP(tp); 1143 1144 TUN_LOCK(tp); 1145 1146 /* 1147 * Realistically, we can't be obstinate here. This only means that the 1148 * tuntap device was closed out of order, and the last closer wasn't the 1149 * controller. These are still good to know about, though, as software 1150 * should avoid multiple processes with a tuntap device open and 1151 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in 1152 * parent). 1153 */ 1154 if (p->p_pid != tp->tun_pid) { 1155 log(LOG_INFO, 1156 "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n", 1157 p->p_pid, p->p_comm, tp->tun_dev->si_name); 1158 } 1159 1160 /* 1161 * junk all pending output 1162 */ 1163 CURVNET_SET(ifp->if_vnet); 1164 1165 l2tun = false; 1166 if ((tp->tun_flags & TUN_L2) != 0) { 1167 l2tun = true; 1168 IF_DRAIN(&ifp->if_snd); 1169 } else { 1170 IFQ_PURGE(&ifp->if_snd); 1171 } 1172 1173 /* For vmnet, we won't do most of the address/route bits */ 1174 if ((tp->tun_flags & TUN_VMNET) != 0 || 1175 (l2tun && (ifp->if_flags & IFF_LINK0) != 0)) 1176 goto out; 1177 1178 if (ifp->if_flags & IFF_UP) { 1179 TUN_UNLOCK(tp); 1180 if_down(ifp); 1181 TUN_LOCK(tp); 1182 } 1183 1184 /* Delete all addresses and routes which reference this interface. */ 1185 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1186 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1187 TUN_UNLOCK(tp); 1188 if_purgeaddrs(ifp); 1189 TUN_LOCK(tp); 1190 } 1191 1192 out: 1193 if_link_state_change(ifp, LINK_STATE_DOWN); 1194 CURVNET_RESTORE(); 1195 1196 funsetown(&tp->tun_sigio); 1197 selwakeuppri(&tp->tun_rsel, PZERO + 1); 1198 KNOTE_LOCKED(&tp->tun_rsel.si_note, 0); 1199 TUNDEBUG (ifp, "closed\n"); 1200 tp->tun_flags &= ~TUN_OPEN; 1201 tp->tun_pid = 0; 1202 tun_vnethdr_set(ifp, 0); 1203 1204 tun_unbusy_locked(tp); 1205 TUN_UNLOCK(tp); 1206 } 1207 1208 static void 1209 tuninit(struct ifnet *ifp) 1210 { 1211 struct tuntap_softc *tp = ifp->if_softc; 1212 1213 TUNDEBUG(ifp, "tuninit\n"); 1214 1215 TUN_LOCK(tp); 1216 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1217 if ((tp->tun_flags & TUN_L2) == 0) { 1218 ifp->if_flags |= IFF_UP; 1219 getmicrotime(&ifp->if_lastchange); 1220 TUN_UNLOCK(tp); 1221 } else { 1222 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1223 TUN_UNLOCK(tp); 1224 /* attempt to start output */ 1225 tunstart_l2(ifp); 1226 } 1227 1228 } 1229 1230 /* 1231 * Used only for l2 tunnel. 1232 */ 1233 static void 1234 tunifinit(void *xtp) 1235 { 1236 struct tuntap_softc *tp; 1237 1238 tp = (struct tuntap_softc *)xtp; 1239 tuninit(tp->tun_ifp); 1240 } 1241 1242 /* 1243 * To be called under TUN_LOCK. Update ifp->if_hwassist according to the 1244 * current value of ifp->if_capenable. 1245 */ 1246 static void 1247 tun_caps_changed(struct ifnet *ifp) 1248 { 1249 uint64_t hwassist = 0; 1250 1251 TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc); 1252 if (ifp->if_capenable & IFCAP_TXCSUM) 1253 hwassist |= CSUM_TCP | CSUM_UDP; 1254 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) 1255 hwassist |= CSUM_TCP_IPV6 1256 | CSUM_UDP_IPV6; 1257 if (ifp->if_capenable & IFCAP_TSO4) 1258 hwassist |= CSUM_IP_TSO; 1259 if (ifp->if_capenable & IFCAP_TSO6) 1260 hwassist |= CSUM_IP6_TSO; 1261 ifp->if_hwassist = hwassist; 1262 } 1263 1264 /* 1265 * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust 1266 * if_capabilities and if_capenable as needed. 1267 */ 1268 static void 1269 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen) 1270 { 1271 struct tuntap_softc *tp = ifp->if_softc; 1272 1273 TUN_LOCK_ASSERT(tp); 1274 1275 if (tp->tun_vhdrlen == vhdrlen) 1276 return; 1277 1278 /* 1279 * Update if_capabilities to reflect the 1280 * functionalities offered by the virtio-net 1281 * header. 1282 */ 1283 if (vhdrlen != 0) 1284 ifp->if_capabilities |= 1285 TAP_VNET_HDR_CAPS; 1286 else 1287 ifp->if_capabilities &= 1288 ~TAP_VNET_HDR_CAPS; 1289 /* 1290 * Disable any capabilities that we don't 1291 * support anymore. 1292 */ 1293 ifp->if_capenable &= ifp->if_capabilities; 1294 tun_caps_changed(ifp); 1295 tp->tun_vhdrlen = vhdrlen; 1296 1297 TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n", 1298 vhdrlen, ifp->if_capabilities); 1299 } 1300 1301 /* 1302 * Process an ioctl request. 1303 */ 1304 static int 1305 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1306 { 1307 struct ifreq *ifr = (struct ifreq *)data; 1308 struct tuntap_softc *tp; 1309 struct ifstat *ifs; 1310 struct ifmediareq *ifmr; 1311 int dummy, error = 0; 1312 bool l2tun; 1313 1314 ifmr = NULL; 1315 sx_xlock(&tun_ioctl_sx); 1316 tp = ifp->if_softc; 1317 if (tp == NULL) { 1318 error = ENXIO; 1319 goto bad; 1320 } 1321 l2tun = (tp->tun_flags & TUN_L2) != 0; 1322 switch(cmd) { 1323 case SIOCGIFSTATUS: 1324 ifs = (struct ifstat *)data; 1325 TUN_LOCK(tp); 1326 if (tp->tun_pid) 1327 snprintf(ifs->ascii, sizeof(ifs->ascii), 1328 "\tOpened by PID %d\n", tp->tun_pid); 1329 else 1330 ifs->ascii[0] = '\0'; 1331 TUN_UNLOCK(tp); 1332 break; 1333 case SIOCSIFADDR: 1334 if (l2tun) 1335 error = ether_ioctl(ifp, cmd, data); 1336 else 1337 tuninit(ifp); 1338 if (error == 0) 1339 TUNDEBUG(ifp, "address set\n"); 1340 break; 1341 case SIOCSIFMTU: 1342 ifp->if_mtu = ifr->ifr_mtu; 1343 TUNDEBUG(ifp, "mtu set\n"); 1344 break; 1345 case SIOCSIFFLAGS: 1346 case SIOCADDMULTI: 1347 case SIOCDELMULTI: 1348 break; 1349 case SIOCGIFMEDIA: 1350 if (!l2tun) { 1351 error = EINVAL; 1352 break; 1353 } 1354 1355 ifmr = (struct ifmediareq *)data; 1356 dummy = ifmr->ifm_count; 1357 ifmr->ifm_count = 1; 1358 ifmr->ifm_status = IFM_AVALID; 1359 ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T; 1360 if (tp->tun_flags & TUN_OPEN) 1361 ifmr->ifm_status |= IFM_ACTIVE; 1362 ifmr->ifm_current = ifmr->ifm_active; 1363 if (dummy >= 1) { 1364 int media = IFM_ETHER; 1365 error = copyout(&media, ifmr->ifm_ulist, sizeof(int)); 1366 } 1367 break; 1368 case SIOCSIFCAP: 1369 TUN_LOCK(tp); 1370 ifp->if_capenable = ifr->ifr_reqcap; 1371 tun_caps_changed(ifp); 1372 TUN_UNLOCK(tp); 1373 VLAN_CAPABILITIES(ifp); 1374 break; 1375 default: 1376 if (l2tun) { 1377 error = ether_ioctl(ifp, cmd, data); 1378 } else { 1379 error = EINVAL; 1380 } 1381 } 1382 bad: 1383 sx_xunlock(&tun_ioctl_sx); 1384 return (error); 1385 } 1386 1387 /* 1388 * tunoutput - queue packets from higher level ready to put out. 1389 */ 1390 static int 1391 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst, 1392 struct route *ro) 1393 { 1394 struct tuntap_softc *tp = ifp->if_softc; 1395 u_short cached_tun_flags; 1396 int error; 1397 u_int32_t af; 1398 1399 TUNDEBUG (ifp, "tunoutput\n"); 1400 1401 #ifdef MAC 1402 error = mac_ifnet_check_transmit(ifp, m0); 1403 if (error) { 1404 m_freem(m0); 1405 return (error); 1406 } 1407 #endif 1408 1409 /* Could be unlocked read? */ 1410 TUN_LOCK(tp); 1411 cached_tun_flags = tp->tun_flags; 1412 TUN_UNLOCK(tp); 1413 if ((cached_tun_flags & TUN_READY) != TUN_READY) { 1414 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags); 1415 m_freem (m0); 1416 return (EHOSTDOWN); 1417 } 1418 1419 if ((ifp->if_flags & IFF_UP) != IFF_UP) { 1420 m_freem (m0); 1421 return (EHOSTDOWN); 1422 } 1423 1424 /* BPF writes need to be handled specially. */ 1425 if (dst->sa_family == AF_UNSPEC) 1426 bcopy(dst->sa_data, &af, sizeof(af)); 1427 else 1428 af = RO_GET_FAMILY(ro, dst); 1429 1430 BPF_MTAP2(ifp, &af, sizeof(af), m0); 1431 1432 /* prepend sockaddr? this may abort if the mbuf allocation fails */ 1433 if (cached_tun_flags & TUN_LMODE) { 1434 /* allocate space for sockaddr */ 1435 M_PREPEND(m0, dst->sa_len, 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 bcopy(dst, m0->m_data, dst->sa_len); 1444 } 1445 } 1446 1447 if (cached_tun_flags & TUN_IFHEAD) { 1448 /* Prepend the address family */ 1449 M_PREPEND(m0, 4, M_NOWAIT); 1450 1451 /* if allocation failed drop packet */ 1452 if (m0 == NULL) { 1453 if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1); 1454 if_inc_counter(ifp, IFCOUNTER_OERRORS, 1); 1455 return (ENOBUFS); 1456 } else 1457 *(u_int32_t *)m0->m_data = htonl(af); 1458 } else { 1459 #ifdef INET 1460 if (af != AF_INET) 1461 #endif 1462 { 1463 m_freem(m0); 1464 return (EAFNOSUPPORT); 1465 } 1466 } 1467 1468 error = (ifp->if_transmit)(ifp, m0); 1469 if (error) 1470 return (ENOBUFS); 1471 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 1472 return (0); 1473 } 1474 1475 /* 1476 * the cdevsw interface is now pretty minimal. 1477 */ 1478 static int 1479 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, 1480 struct thread *td) 1481 { 1482 struct ifreq ifr, *ifrp; 1483 struct tuntap_softc *tp = dev->si_drv1; 1484 struct ifnet *ifp = TUN2IFP(tp); 1485 struct tuninfo *tunp; 1486 int error, iflags, ival; 1487 bool l2tun; 1488 1489 l2tun = (tp->tun_flags & TUN_L2) != 0; 1490 if (l2tun) { 1491 /* tap specific ioctls */ 1492 switch(cmd) { 1493 /* VMware/VMnet port ioctl's */ 1494 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \ 1495 defined(COMPAT_FREEBSD4) 1496 case _IO('V', 0): 1497 ival = IOCPARM_IVAL(data); 1498 data = (caddr_t)&ival; 1499 /* FALLTHROUGH */ 1500 #endif 1501 case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */ 1502 iflags = *(int *)data; 1503 iflags &= TUN_VMIO_FLAG_MASK; 1504 iflags &= ~IFF_CANTCHANGE; 1505 iflags |= IFF_UP; 1506 1507 TUN_LOCK(tp); 1508 ifp->if_flags = iflags | 1509 (ifp->if_flags & IFF_CANTCHANGE); 1510 TUN_UNLOCK(tp); 1511 1512 return (0); 1513 case SIOCGIFADDR: /* get MAC address of the remote side */ 1514 TUN_LOCK(tp); 1515 bcopy(&tp->tun_ether.octet, data, 1516 sizeof(tp->tun_ether.octet)); 1517 TUN_UNLOCK(tp); 1518 1519 return (0); 1520 case SIOCSIFADDR: /* set MAC address of the remote side */ 1521 TUN_LOCK(tp); 1522 bcopy(data, &tp->tun_ether.octet, 1523 sizeof(tp->tun_ether.octet)); 1524 TUN_UNLOCK(tp); 1525 1526 return (0); 1527 case TAPSVNETHDR: 1528 ival = *(int *)data; 1529 if (ival != 0 && 1530 ival != sizeof(struct virtio_net_hdr) && 1531 ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) { 1532 return (EINVAL); 1533 } 1534 TUN_LOCK(tp); 1535 tun_vnethdr_set(ifp, ival); 1536 TUN_UNLOCK(tp); 1537 1538 return (0); 1539 case TAPGVNETHDR: 1540 TUN_LOCK(tp); 1541 *(int *)data = tp->tun_vhdrlen; 1542 TUN_UNLOCK(tp); 1543 1544 return (0); 1545 } 1546 1547 /* Fall through to the common ioctls if unhandled */ 1548 } else { 1549 switch (cmd) { 1550 case TUNSLMODE: 1551 TUN_LOCK(tp); 1552 if (*(int *)data) { 1553 tp->tun_flags |= TUN_LMODE; 1554 tp->tun_flags &= ~TUN_IFHEAD; 1555 } else 1556 tp->tun_flags &= ~TUN_LMODE; 1557 TUN_UNLOCK(tp); 1558 1559 return (0); 1560 case TUNSIFHEAD: 1561 TUN_LOCK(tp); 1562 if (*(int *)data) { 1563 tp->tun_flags |= TUN_IFHEAD; 1564 tp->tun_flags &= ~TUN_LMODE; 1565 } else 1566 tp->tun_flags &= ~TUN_IFHEAD; 1567 TUN_UNLOCK(tp); 1568 1569 return (0); 1570 case TUNGIFHEAD: 1571 TUN_LOCK(tp); 1572 *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0; 1573 TUN_UNLOCK(tp); 1574 1575 return (0); 1576 case TUNSIFMODE: 1577 /* deny this if UP */ 1578 if (TUN2IFP(tp)->if_flags & IFF_UP) 1579 return (EBUSY); 1580 1581 switch (*(int *)data & ~IFF_MULTICAST) { 1582 case IFF_POINTOPOINT: 1583 case IFF_BROADCAST: 1584 TUN_LOCK(tp); 1585 TUN2IFP(tp)->if_flags &= 1586 ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST); 1587 TUN2IFP(tp)->if_flags |= *(int *)data; 1588 TUN_UNLOCK(tp); 1589 1590 break; 1591 default: 1592 return (EINVAL); 1593 } 1594 1595 return (0); 1596 case TUNSIFPID: 1597 TUN_LOCK(tp); 1598 tp->tun_pid = curthread->td_proc->p_pid; 1599 TUN_UNLOCK(tp); 1600 1601 return (0); 1602 } 1603 /* Fall through to the common ioctls if unhandled */ 1604 } 1605 1606 switch (cmd) { 1607 case TUNGIFNAME: 1608 ifrp = (struct ifreq *)data; 1609 strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ); 1610 1611 return (0); 1612 case TUNSIFINFO: 1613 tunp = (struct tuninfo *)data; 1614 if (TUN2IFP(tp)->if_type != tunp->type) 1615 return (EPROTOTYPE); 1616 TUN_LOCK(tp); 1617 if (TUN2IFP(tp)->if_mtu != tunp->mtu) { 1618 strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ); 1619 ifr.ifr_mtu = tunp->mtu; 1620 CURVNET_SET(TUN2IFP(tp)->if_vnet); 1621 error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp), 1622 (caddr_t)&ifr, td); 1623 CURVNET_RESTORE(); 1624 if (error) { 1625 TUN_UNLOCK(tp); 1626 return (error); 1627 } 1628 } 1629 TUN2IFP(tp)->if_baudrate = tunp->baudrate; 1630 TUN_UNLOCK(tp); 1631 break; 1632 case TUNGIFINFO: 1633 tunp = (struct tuninfo *)data; 1634 TUN_LOCK(tp); 1635 tunp->mtu = TUN2IFP(tp)->if_mtu; 1636 tunp->type = TUN2IFP(tp)->if_type; 1637 tunp->baudrate = TUN2IFP(tp)->if_baudrate; 1638 TUN_UNLOCK(tp); 1639 break; 1640 case TUNSDEBUG: 1641 tundebug = *(int *)data; 1642 break; 1643 case TUNGDEBUG: 1644 *(int *)data = tundebug; 1645 break; 1646 case FIONBIO: 1647 break; 1648 case FIOASYNC: 1649 TUN_LOCK(tp); 1650 if (*(int *)data) 1651 tp->tun_flags |= TUN_ASYNC; 1652 else 1653 tp->tun_flags &= ~TUN_ASYNC; 1654 TUN_UNLOCK(tp); 1655 break; 1656 case FIONREAD: 1657 if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) { 1658 struct mbuf *mb; 1659 IFQ_LOCK(&TUN2IFP(tp)->if_snd); 1660 IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb); 1661 for (*(int *)data = 0; mb != NULL; mb = mb->m_next) 1662 *(int *)data += mb->m_len; 1663 IFQ_UNLOCK(&TUN2IFP(tp)->if_snd); 1664 } else 1665 *(int *)data = 0; 1666 break; 1667 case FIOSETOWN: 1668 return (fsetown(*(int *)data, &tp->tun_sigio)); 1669 1670 case FIOGETOWN: 1671 *(int *)data = fgetown(&tp->tun_sigio); 1672 return (0); 1673 1674 /* This is deprecated, FIOSETOWN should be used instead. */ 1675 case TIOCSPGRP: 1676 return (fsetown(-(*(int *)data), &tp->tun_sigio)); 1677 1678 /* This is deprecated, FIOGETOWN should be used instead. */ 1679 case TIOCGPGRP: 1680 *(int *)data = -fgetown(&tp->tun_sigio); 1681 return (0); 1682 1683 default: 1684 return (ENOTTY); 1685 } 1686 return (0); 1687 } 1688 1689 /* 1690 * The cdevsw read interface - reads a packet at a time, or at 1691 * least as much of a packet as can be read. 1692 */ 1693 static int 1694 tunread(struct cdev *dev, struct uio *uio, int flag) 1695 { 1696 struct tuntap_softc *tp = dev->si_drv1; 1697 struct ifnet *ifp = TUN2IFP(tp); 1698 struct mbuf *m; 1699 size_t len; 1700 int error = 0; 1701 1702 TUNDEBUG (ifp, "read\n"); 1703 TUN_LOCK(tp); 1704 if ((tp->tun_flags & TUN_READY) != TUN_READY) { 1705 TUN_UNLOCK(tp); 1706 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags); 1707 return (EHOSTDOWN); 1708 } 1709 1710 tp->tun_flags &= ~TUN_RWAIT; 1711 1712 for (;;) { 1713 IFQ_DEQUEUE(&ifp->if_snd, m); 1714 if (m != NULL) 1715 break; 1716 if (flag & O_NONBLOCK) { 1717 TUN_UNLOCK(tp); 1718 return (EWOULDBLOCK); 1719 } 1720 tp->tun_flags |= TUN_RWAIT; 1721 error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1), 1722 "tunread", 0); 1723 if (error != 0) { 1724 TUN_UNLOCK(tp); 1725 return (error); 1726 } 1727 } 1728 TUN_UNLOCK(tp); 1729 1730 len = min(tp->tun_vhdrlen, uio->uio_resid); 1731 if (len > 0) { 1732 struct virtio_net_hdr_mrg_rxbuf vhdr; 1733 1734 bzero(&vhdr, sizeof(vhdr)); 1735 if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) { 1736 m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr); 1737 } 1738 1739 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, " 1740 "gs %u, cs %u, co %u\n", vhdr.hdr.flags, 1741 vhdr.hdr.gso_type, vhdr.hdr.hdr_len, 1742 vhdr.hdr.gso_size, vhdr.hdr.csum_start, 1743 vhdr.hdr.csum_offset); 1744 error = uiomove(&vhdr, len, uio); 1745 } 1746 1747 while (m && uio->uio_resid > 0 && error == 0) { 1748 len = min(uio->uio_resid, m->m_len); 1749 if (len != 0) 1750 error = uiomove(mtod(m, void *), len, uio); 1751 m = m_free(m); 1752 } 1753 1754 if (m) { 1755 TUNDEBUG(ifp, "Dropping mbuf\n"); 1756 m_freem(m); 1757 } 1758 return (error); 1759 } 1760 1761 static int 1762 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m, 1763 struct virtio_net_hdr_mrg_rxbuf *vhdr) 1764 { 1765 struct epoch_tracker et; 1766 struct ether_header *eh; 1767 struct ifnet *ifp; 1768 1769 ifp = TUN2IFP(tp); 1770 1771 /* 1772 * Only pass a unicast frame to ether_input(), if it would 1773 * actually have been received by non-virtual hardware. 1774 */ 1775 if (m->m_len < sizeof(struct ether_header)) { 1776 m_freem(m); 1777 return (0); 1778 } 1779 1780 eh = mtod(m, struct ether_header *); 1781 1782 if ((ifp->if_flags & IFF_PROMISC) == 0 && 1783 !ETHER_IS_MULTICAST(eh->ether_dhost) && 1784 bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) { 1785 m_freem(m); 1786 return (0); 1787 } 1788 1789 if (vhdr != NULL) { 1790 if (virtio_net_rx_csum(m, &vhdr->hdr)) { 1791 m_freem(m); 1792 return (0); 1793 } 1794 } else { 1795 switch (ntohs(eh->ether_type)) { 1796 #ifdef INET 1797 case ETHERTYPE_IP: 1798 if (ifp->if_capenable & IFCAP_RXCSUM) { 1799 m->m_pkthdr.csum_flags |= 1800 CSUM_IP_CHECKED | CSUM_IP_VALID | 1801 CSUM_DATA_VALID | CSUM_SCTP_VALID | 1802 CSUM_PSEUDO_HDR; 1803 m->m_pkthdr.csum_data = 0xffff; 1804 } 1805 break; 1806 #endif 1807 #ifdef INET6 1808 case ETHERTYPE_IPV6: 1809 if (ifp->if_capenable & IFCAP_RXCSUM_IPV6) { 1810 m->m_pkthdr.csum_flags |= 1811 CSUM_DATA_VALID_IPV6 | CSUM_SCTP_VALID | 1812 CSUM_PSEUDO_HDR; 1813 m->m_pkthdr.csum_data = 0xffff; 1814 } 1815 break; 1816 #endif 1817 } 1818 } 1819 1820 /* Pass packet up to parent. */ 1821 CURVNET_SET(ifp->if_vnet); 1822 NET_EPOCH_ENTER(et); 1823 (*ifp->if_input)(ifp, m); 1824 NET_EPOCH_EXIT(et); 1825 CURVNET_RESTORE(); 1826 /* ibytes are counted in parent */ 1827 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); 1828 return (0); 1829 } 1830 1831 static int 1832 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m) 1833 { 1834 struct epoch_tracker et; 1835 struct ifnet *ifp; 1836 int family, isr; 1837 1838 ifp = TUN2IFP(tp); 1839 /* Could be unlocked read? */ 1840 TUN_LOCK(tp); 1841 if (tp->tun_flags & TUN_IFHEAD) { 1842 TUN_UNLOCK(tp); 1843 if (m->m_len < sizeof(family) && 1844 (m = m_pullup(m, sizeof(family))) == NULL) 1845 return (ENOBUFS); 1846 family = ntohl(*mtod(m, u_int32_t *)); 1847 m_adj(m, sizeof(family)); 1848 } else { 1849 TUN_UNLOCK(tp); 1850 family = AF_INET; 1851 } 1852 1853 BPF_MTAP2(ifp, &family, sizeof(family), m); 1854 1855 switch (family) { 1856 #ifdef INET 1857 case AF_INET: 1858 isr = NETISR_IP; 1859 break; 1860 #endif 1861 #ifdef INET6 1862 case AF_INET6: 1863 isr = NETISR_IPV6; 1864 break; 1865 #endif 1866 default: 1867 m_freem(m); 1868 return (EAFNOSUPPORT); 1869 } 1870 random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN); 1871 if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len); 1872 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); 1873 CURVNET_SET(ifp->if_vnet); 1874 M_SETFIB(m, ifp->if_fib); 1875 NET_EPOCH_ENTER(et); 1876 netisr_dispatch(isr, m); 1877 NET_EPOCH_EXIT(et); 1878 CURVNET_RESTORE(); 1879 return (0); 1880 } 1881 1882 /* 1883 * the cdevsw write interface - an atomic write is a packet - or else! 1884 */ 1885 static int 1886 tunwrite(struct cdev *dev, struct uio *uio, int flag) 1887 { 1888 struct virtio_net_hdr_mrg_rxbuf vhdr; 1889 struct tuntap_softc *tp; 1890 struct ifnet *ifp; 1891 struct mbuf *m; 1892 uint32_t mru; 1893 int align, vhdrlen, error; 1894 bool l2tun; 1895 1896 tp = dev->si_drv1; 1897 ifp = TUN2IFP(tp); 1898 TUNDEBUG(ifp, "tunwrite\n"); 1899 if ((ifp->if_flags & IFF_UP) != IFF_UP) 1900 /* ignore silently */ 1901 return (0); 1902 1903 if (uio->uio_resid == 0) 1904 return (0); 1905 1906 l2tun = (tp->tun_flags & TUN_L2) != 0; 1907 mru = l2tun ? TAPMRU : TUNMRU; 1908 vhdrlen = tp->tun_vhdrlen; 1909 align = 0; 1910 if (l2tun) { 1911 align = ETHER_ALIGN; 1912 mru += vhdrlen; 1913 } else if ((tp->tun_flags & TUN_IFHEAD) != 0) 1914 mru += sizeof(uint32_t); /* family */ 1915 if (uio->uio_resid < 0 || uio->uio_resid > mru) { 1916 TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid); 1917 return (EIO); 1918 } 1919 1920 if (vhdrlen > 0) { 1921 error = uiomove(&vhdr, vhdrlen, uio); 1922 if (error != 0) 1923 return (error); 1924 TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, " 1925 "gs %u, cs %u, co %u\n", vhdr.hdr.flags, 1926 vhdr.hdr.gso_type, vhdr.hdr.hdr_len, 1927 vhdr.hdr.gso_size, vhdr.hdr.csum_start, 1928 vhdr.hdr.csum_offset); 1929 } 1930 1931 if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) { 1932 if_inc_counter(ifp, IFCOUNTER_IERRORS, 1); 1933 return (ENOBUFS); 1934 } 1935 1936 m->m_pkthdr.rcvif = ifp; 1937 #ifdef MAC 1938 mac_ifnet_create_mbuf(ifp, m); 1939 #endif 1940 1941 if (l2tun) 1942 return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL)); 1943 1944 return (tunwrite_l3(tp, m)); 1945 } 1946 1947 /* 1948 * tunpoll - the poll interface, this is only useful on reads 1949 * really. The write detect always returns true, write never blocks 1950 * anyway, it either accepts the packet or drops it. 1951 */ 1952 static int 1953 tunpoll(struct cdev *dev, int events, struct thread *td) 1954 { 1955 struct tuntap_softc *tp = dev->si_drv1; 1956 struct ifnet *ifp = TUN2IFP(tp); 1957 int revents = 0; 1958 1959 TUNDEBUG(ifp, "tunpoll\n"); 1960 1961 if (events & (POLLIN | POLLRDNORM)) { 1962 IFQ_LOCK(&ifp->if_snd); 1963 if (!IFQ_IS_EMPTY(&ifp->if_snd)) { 1964 TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len); 1965 revents |= events & (POLLIN | POLLRDNORM); 1966 } else { 1967 TUNDEBUG(ifp, "tunpoll waiting\n"); 1968 selrecord(td, &tp->tun_rsel); 1969 } 1970 IFQ_UNLOCK(&ifp->if_snd); 1971 } 1972 revents |= events & (POLLOUT | POLLWRNORM); 1973 1974 return (revents); 1975 } 1976 1977 /* 1978 * tunkqfilter - support for the kevent() system call. 1979 */ 1980 static int 1981 tunkqfilter(struct cdev *dev, struct knote *kn) 1982 { 1983 struct tuntap_softc *tp = dev->si_drv1; 1984 struct ifnet *ifp = TUN2IFP(tp); 1985 1986 switch(kn->kn_filter) { 1987 case EVFILT_READ: 1988 TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n", 1989 ifp->if_xname, dev2unit(dev)); 1990 kn->kn_fop = &tun_read_filterops; 1991 break; 1992 1993 case EVFILT_WRITE: 1994 TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n", 1995 ifp->if_xname, dev2unit(dev)); 1996 kn->kn_fop = &tun_write_filterops; 1997 break; 1998 1999 default: 2000 TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n", 2001 ifp->if_xname, dev2unit(dev)); 2002 return(EINVAL); 2003 } 2004 2005 kn->kn_hook = tp; 2006 knlist_add(&tp->tun_rsel.si_note, kn, 0); 2007 2008 return (0); 2009 } 2010 2011 /* 2012 * Return true of there is data in the interface queue. 2013 */ 2014 static int 2015 tunkqread(struct knote *kn, long hint) 2016 { 2017 int ret; 2018 struct tuntap_softc *tp = kn->kn_hook; 2019 struct cdev *dev = tp->tun_dev; 2020 struct ifnet *ifp = TUN2IFP(tp); 2021 2022 if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) { 2023 TUNDEBUG(ifp, 2024 "%s have data in the queue. Len = %d, minor = %#x\n", 2025 ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev)); 2026 ret = 1; 2027 } else { 2028 TUNDEBUG(ifp, 2029 "%s waiting for data, minor = %#x\n", ifp->if_xname, 2030 dev2unit(dev)); 2031 ret = 0; 2032 } 2033 2034 return (ret); 2035 } 2036 2037 /* 2038 * Always can write, always return MTU in kn->data. 2039 */ 2040 static int 2041 tunkqwrite(struct knote *kn, long hint) 2042 { 2043 struct tuntap_softc *tp = kn->kn_hook; 2044 struct ifnet *ifp = TUN2IFP(tp); 2045 2046 kn->kn_data = ifp->if_mtu; 2047 2048 return (1); 2049 } 2050 2051 static void 2052 tunkqdetach(struct knote *kn) 2053 { 2054 struct tuntap_softc *tp = kn->kn_hook; 2055 2056 knlist_remove(&tp->tun_rsel.si_note, kn, 0); 2057 } 2058