if_tun.c revision 130585
1/* $NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $ */ 2 3/* 4 * Copyright (c) 1988, Julian Onions <jpo@cs.nott.ac.uk> 5 * Nottingham University 1987. 6 * 7 * This source may be freely distributed, however I would be interested 8 * in any changes that are made. 9 * 10 * This driver takes packets off the IP i/f and hands them up to a 11 * user process to have its wicked way with. This driver has it's 12 * roots in a similar driver written by Phil Cockcroft (formerly) at 13 * UCL. This driver is based much more on read/write/poll mode of 14 * operation though. 15 * 16 * $FreeBSD: head/sys/net/if_tun.c 130585 2004-06-16 09:47:26Z phk $ 17 */ 18 19#include "opt_atalk.h" 20#include "opt_inet.h" 21#include "opt_inet6.h" 22#include "opt_ipx.h" 23#include "opt_mac.h" 24 25#include <sys/param.h> 26#include <sys/proc.h> 27#include <sys/systm.h> 28#include <sys/mac.h> 29#include <sys/mbuf.h> 30#include <sys/module.h> 31#include <sys/socket.h> 32#include <sys/filio.h> 33#include <sys/sockio.h> 34#include <sys/ttycom.h> 35#include <sys/poll.h> 36#include <sys/signalvar.h> 37#include <sys/filedesc.h> 38#include <sys/kernel.h> 39#include <sys/sysctl.h> 40#include <sys/conf.h> 41#include <sys/uio.h> 42#include <sys/vnode.h> 43#include <sys/malloc.h> 44#include <sys/random.h> 45 46#include <net/if.h> 47#include <net/if_types.h> 48#include <net/netisr.h> 49#include <net/route.h> 50#ifdef INET 51#include <netinet/in.h> 52#endif 53#include <net/bpf.h> 54#include <net/if_tun.h> 55 56#include <sys/queue.h> 57 58/* 59 * tun_list is protected by global tunmtx. Other mutable fields are 60 * protected by tun->tun_mtx, or by their owning subsystem. tun_dev is 61 * static for the duration of a tunnel interface. 62 */ 63struct tun_softc { 64 TAILQ_ENTRY(tun_softc) tun_list; 65 struct cdev *tun_dev; 66 u_short tun_flags; /* misc flags */ 67#define TUN_OPEN 0x0001 68#define TUN_INITED 0x0002 69#define TUN_RCOLL 0x0004 70#define TUN_IASET 0x0008 71#define TUN_DSTADDR 0x0010 72#define TUN_LMODE 0x0020 73#define TUN_RWAIT 0x0040 74#define TUN_ASYNC 0x0080 75#define TUN_IFHEAD 0x0100 76 77#define TUN_READY (TUN_OPEN | TUN_INITED) 78 79 /* 80 * XXXRW: tun_pid is used to exclusively lock /dev/tun. Is this 81 * actually needed? Can we just return EBUSY if already open? 82 * Problem is that this involved inherent races when a tun device 83 * is handed off from one process to another, as opposed to just 84 * being slightly stale informationally. 85 */ 86 pid_t tun_pid; /* owning pid */ 87 struct ifnet tun_if; /* the interface */ 88 struct sigio *tun_sigio; /* information for async I/O */ 89 struct selinfo tun_rsel; /* read select */ 90 struct mtx tun_mtx; /* protect mutable softc fields */ 91}; 92 93#define TUNDEBUG if (tundebug) if_printf 94#define TUNNAME "tun" 95 96/* 97 * All mutable global variables in if_tun are locked using tunmtx, with 98 * the exception of tundebug, which is used unlocked, and tunclones, 99 * which is static after setup. 100 */ 101static struct mtx tunmtx; 102static MALLOC_DEFINE(M_TUN, TUNNAME, "Tunnel Interface"); 103static int tundebug = 0; 104static struct clonedevs *tunclones; 105static TAILQ_HEAD(,tun_softc) tunhead = TAILQ_HEAD_INITIALIZER(tunhead); 106SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, ""); 107 108static void tunclone(void *arg, char *name, int namelen, struct cdev **dev); 109static void tuncreate(struct cdev *dev); 110static int tunifioctl(struct ifnet *, u_long, caddr_t); 111static int tuninit(struct ifnet *); 112static int tunmodevent(module_t, int, void *); 113static int tunoutput(struct ifnet *, struct mbuf *, struct sockaddr *, 114 struct rtentry *rt); 115static void tunstart(struct ifnet *); 116 117static d_open_t tunopen; 118static d_close_t tunclose; 119static d_read_t tunread; 120static d_write_t tunwrite; 121static d_ioctl_t tunioctl; 122static d_poll_t tunpoll; 123 124static struct cdevsw tun_cdevsw = { 125 .d_version = D_VERSION, 126 .d_flags = D_PSEUDO | D_NEEDGIANT, 127 .d_open = tunopen, 128 .d_close = tunclose, 129 .d_read = tunread, 130 .d_write = tunwrite, 131 .d_ioctl = tunioctl, 132 .d_poll = tunpoll, 133 .d_name = TUNNAME, 134}; 135 136static void 137tunclone(void *arg, char *name, int namelen, struct cdev **dev) 138{ 139 int u, i; 140 141 if (*dev != NODEV) 142 return; 143 144 if (strcmp(name, TUNNAME) == 0) { 145 u = -1; 146 } else if (dev_stdclone(name, NULL, TUNNAME, &u) != 1) 147 return; /* Don't recognise the name */ 148 if (u != -1 && u > IF_MAXUNIT) 149 return; /* Unit number too high */ 150 151 /* find any existing device, or allocate new unit number */ 152 i = clone_create(&tunclones, &tun_cdevsw, &u, dev, 0); 153 if (i) { 154 /* No preexisting struct cdev *, create one */ 155 *dev = make_dev(&tun_cdevsw, unit2minor(u), 156 UID_UUCP, GID_DIALER, 0600, "tun%d", u); 157 if (*dev != NULL) 158 (*dev)->si_flags |= SI_CHEAPCLONE; 159 } 160} 161 162static void 163tun_destroy(struct tun_softc *tp) 164{ 165 struct cdev *dev; 166 167 /* Unlocked read. */ 168 KASSERT((tp->tun_flags & TUN_OPEN) == 0, 169 ("tununits is out of sync - unit %d", tp->tun_if.if_dunit)); 170 171 dev = tp->tun_dev; 172 bpfdetach(&tp->tun_if); 173 if_detach(&tp->tun_if); 174 destroy_dev(dev); 175 mtx_destroy(&tp->tun_mtx); 176 free(tp, M_TUN); 177} 178 179static int 180tunmodevent(module_t mod, int type, void *data) 181{ 182 static eventhandler_tag tag; 183 struct tun_softc *tp; 184 185 switch (type) { 186 case MOD_LOAD: 187 mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF); 188 clone_setup(&tunclones); 189 tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000); 190 if (tag == NULL) 191 return (ENOMEM); 192 break; 193 case MOD_UNLOAD: 194 EVENTHANDLER_DEREGISTER(dev_clone, tag); 195 196 mtx_lock(&tunmtx); 197 while ((tp = TAILQ_FIRST(&tunhead)) != NULL) { 198 TAILQ_REMOVE(&tunhead, tp, tun_list); 199 mtx_unlock(&tunmtx); 200 tun_destroy(tp); 201 mtx_lock(&tunmtx); 202 } 203 mtx_unlock(&tunmtx); 204 clone_cleanup(&tunclones); 205 mtx_destroy(&tunmtx); 206 break; 207 } 208 return 0; 209} 210 211static moduledata_t tun_mod = { 212 "if_tun", 213 tunmodevent, 214 0 215}; 216 217DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 218 219static void 220tunstart(struct ifnet *ifp) 221{ 222 struct tun_softc *tp = ifp->if_softc; 223 224 mtx_lock(&tp->tun_mtx); 225 if (tp->tun_flags & TUN_RWAIT) { 226 tp->tun_flags &= ~TUN_RWAIT; 227 wakeup(tp); 228 } 229 if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) { 230 mtx_unlock(&tp->tun_mtx); 231 pgsigio(&tp->tun_sigio, SIGIO, 0); 232 } else 233 mtx_unlock(&tp->tun_mtx); 234 selwakeuppri(&tp->tun_rsel, PZERO + 1); 235} 236 237static void 238tuncreate(struct cdev *dev) 239{ 240 struct tun_softc *sc; 241 struct ifnet *ifp; 242 243 dev->si_flags &= ~SI_CHEAPCLONE; 244 245 MALLOC(sc, struct tun_softc *, sizeof(*sc), M_TUN, M_WAITOK | M_ZERO); 246 mtx_init(&sc->tun_mtx, "tun_mtx", NULL, MTX_DEF); 247 sc->tun_flags = TUN_INITED; 248 sc->tun_dev = dev; 249 mtx_lock(&tunmtx); 250 TAILQ_INSERT_TAIL(&tunhead, sc, tun_list); 251 mtx_unlock(&tunmtx); 252 253 ifp = &sc->tun_if; 254 if_initname(ifp, TUNNAME, dev2unit(dev)); 255 ifp->if_mtu = TUNMTU; 256 ifp->if_ioctl = tunifioctl; 257 ifp->if_output = tunoutput; 258 ifp->if_start = tunstart; 259 ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST; 260 ifp->if_type = IFT_PPP; 261 ifp->if_snd.ifq_maxlen = ifqmaxlen; 262 ifp->if_softc = sc; 263 if_attach(ifp); 264 bpfattach(ifp, DLT_NULL, sizeof(u_int)); 265 dev->si_drv1 = sc; 266} 267 268static int 269tunopen(struct cdev *dev, int flag, int mode, struct thread *td) 270{ 271 struct ifnet *ifp; 272 struct tun_softc *tp; 273 274 /* 275 * XXXRW: Non-atomic test and set of dev->si_drv1 requires 276 * synchronization. 277 */ 278 tp = dev->si_drv1; 279 if (!tp) { 280 tuncreate(dev); 281 tp = dev->si_drv1; 282 } 283 284 /* 285 * XXXRW: This use of tun_pid is subject to error due to the 286 * fact that a reference to the tunnel can live beyond the 287 * death of the process that created it. Can we replace this 288 * with a simple busy flag? 289 */ 290 mtx_lock(&tp->tun_mtx); 291 if (tp->tun_pid != 0 && tp->tun_pid != td->td_proc->p_pid) { 292 mtx_unlock(&tp->tun_mtx); 293 return (EBUSY); 294 } 295 tp->tun_pid = td->td_proc->p_pid; 296 297 tp->tun_flags |= TUN_OPEN; 298 mtx_unlock(&tp->tun_mtx); 299 ifp = &tp->tun_if; 300 TUNDEBUG(ifp, "open\n"); 301 302 return (0); 303} 304 305/* 306 * tunclose - close the device - mark i/f down & delete 307 * routing info 308 */ 309static int 310tunclose(struct cdev *dev, int foo, int bar, struct thread *td) 311{ 312 struct tun_softc *tp; 313 struct ifnet *ifp; 314 int s; 315 316 tp = dev->si_drv1; 317 ifp = &tp->tun_if; 318 319 mtx_lock(&tp->tun_mtx); 320 tp->tun_flags &= ~TUN_OPEN; 321 tp->tun_pid = 0; 322 mtx_unlock(&tp->tun_mtx); 323 324 /* 325 * junk all pending output 326 */ 327 IF_DRAIN(&ifp->if_snd); 328 329 if (ifp->if_flags & IFF_UP) { 330 s = splimp(); 331 if_down(ifp); 332 splx(s); 333 } 334 335 if (ifp->if_flags & IFF_RUNNING) { 336 struct ifaddr *ifa; 337 338 s = splimp(); 339 /* find internet addresses and delete routes */ 340 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) 341 if (ifa->ifa_addr->sa_family == AF_INET) 342 /* Unlocked read. */ 343 rtinit(ifa, (int)RTM_DELETE, 344 tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0); 345 ifp->if_flags &= ~IFF_RUNNING; 346 splx(s); 347 } 348 349 funsetown(&tp->tun_sigio); 350 selwakeuppri(&tp->tun_rsel, PZERO + 1); 351 TUNDEBUG (ifp, "closed\n"); 352 return (0); 353} 354 355static int 356tuninit(struct ifnet *ifp) 357{ 358 struct tun_softc *tp = ifp->if_softc; 359 struct ifaddr *ifa; 360 int error = 0; 361 362 TUNDEBUG(ifp, "tuninit\n"); 363 364 ifp->if_flags |= IFF_UP | IFF_RUNNING; 365 getmicrotime(&ifp->if_lastchange); 366 367 for (ifa = TAILQ_FIRST(&ifp->if_addrhead); ifa; 368 ifa = TAILQ_NEXT(ifa, ifa_link)) { 369 if (ifa->ifa_addr == NULL) 370 error = EFAULT; 371 /* XXX: Should maybe return straight off? */ 372 else { 373#ifdef INET 374 if (ifa->ifa_addr->sa_family == AF_INET) { 375 struct sockaddr_in *si; 376 377 si = (struct sockaddr_in *)ifa->ifa_addr; 378 mtx_lock(&tp->tun_mtx); 379 if (si->sin_addr.s_addr) 380 tp->tun_flags |= TUN_IASET; 381 382 si = (struct sockaddr_in *)ifa->ifa_dstaddr; 383 if (si && si->sin_addr.s_addr) 384 tp->tun_flags |= TUN_DSTADDR; 385 mtx_unlock(&tp->tun_mtx); 386 } 387#endif 388 } 389 } 390 return (error); 391} 392 393/* 394 * Process an ioctl request. 395 */ 396static int 397tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 398{ 399 struct ifreq *ifr = (struct ifreq *)data; 400 struct tun_softc *tp = ifp->if_softc; 401 struct ifstat *ifs; 402 int error = 0, s; 403 404 s = splimp(); 405 switch(cmd) { 406 case SIOCGIFSTATUS: 407 ifs = (struct ifstat *)data; 408 mtx_lock(&tp->tun_mtx); 409 if (tp->tun_pid) 410 sprintf(ifs->ascii + strlen(ifs->ascii), 411 "\tOpened by PID %d\n", tp->tun_pid); 412 mtx_unlock(&tp->tun_mtx); 413 break; 414 case SIOCSIFADDR: 415 error = tuninit(ifp); 416 TUNDEBUG(ifp, "address set, error=%d\n", error); 417 break; 418 case SIOCSIFDSTADDR: 419 error = tuninit(ifp); 420 TUNDEBUG(ifp, "destination address set, error=%d\n", error); 421 break; 422 case SIOCSIFMTU: 423 ifp->if_mtu = ifr->ifr_mtu; 424 TUNDEBUG(ifp, "mtu set\n"); 425 break; 426 case SIOCSIFFLAGS: 427 case SIOCADDMULTI: 428 case SIOCDELMULTI: 429 break; 430 default: 431 error = EINVAL; 432 } 433 splx(s); 434 return (error); 435} 436 437/* 438 * tunoutput - queue packets from higher level ready to put out. 439 */ 440static int 441tunoutput( 442 struct ifnet *ifp, 443 struct mbuf *m0, 444 struct sockaddr *dst, 445 struct rtentry *rt) 446{ 447 struct tun_softc *tp = ifp->if_softc; 448 u_short cached_tun_flags; 449 int error; 450 451 TUNDEBUG (ifp, "tunoutput\n"); 452 453#ifdef MAC 454 error = mac_check_ifnet_transmit(ifp, m0); 455 if (error) { 456 m_freem(m0); 457 return (error); 458 } 459#endif 460 461 /* Could be unlocked read? */ 462 mtx_lock(&tp->tun_mtx); 463 cached_tun_flags = tp->tun_flags; 464 mtx_unlock(&tp->tun_mtx); 465 if ((cached_tun_flags & TUN_READY) != TUN_READY) { 466 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags); 467 m_freem (m0); 468 return (EHOSTDOWN); 469 } 470 471 if ((ifp->if_flags & IFF_UP) != IFF_UP) { 472 m_freem (m0); 473 return (EHOSTDOWN); 474 } 475 476 /* BPF write needs to be handled specially */ 477 if (dst->sa_family == AF_UNSPEC) { 478 dst->sa_family = *(mtod(m0, int *)); 479 m0->m_len -= sizeof(int); 480 m0->m_pkthdr.len -= sizeof(int); 481 m0->m_data += sizeof(int); 482 } 483 484 if (ifp->if_bpf) { 485 uint32_t af = dst->sa_family; 486 bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0); 487 } 488 489 /* prepend sockaddr? this may abort if the mbuf allocation fails */ 490 if (cached_tun_flags & TUN_LMODE) { 491 /* allocate space for sockaddr */ 492 M_PREPEND(m0, dst->sa_len, M_DONTWAIT); 493 494 /* if allocation failed drop packet */ 495 if (m0 == NULL) { 496 ifp->if_iqdrops++; 497 ifp->if_oerrors++; 498 return (ENOBUFS); 499 } else { 500 bcopy(dst, m0->m_data, dst->sa_len); 501 } 502 } 503 504 if (cached_tun_flags & TUN_IFHEAD) { 505 /* Prepend the address family */ 506 M_PREPEND(m0, 4, M_DONTWAIT); 507 508 /* if allocation failed drop packet */ 509 if (m0 == NULL) { 510 ifp->if_iqdrops++; 511 ifp->if_oerrors++; 512 return (ENOBUFS); 513 } else 514 *(u_int32_t *)m0->m_data = htonl(dst->sa_family); 515 } else { 516#ifdef INET 517 if (dst->sa_family != AF_INET) 518#endif 519 { 520 m_freem(m0); 521 return (EAFNOSUPPORT); 522 } 523 } 524 525 IFQ_HANDOFF(ifp, m0, error); 526 if (error) { 527 ifp->if_collisions++; 528 return (ENOBUFS); 529 } 530 ifp->if_opackets++; 531 return (0); 532} 533 534/* 535 * the cdevsw interface is now pretty minimal. 536 */ 537static int 538tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td) 539{ 540 int s; 541 int error; 542 struct tun_softc *tp = dev->si_drv1; 543 struct tuninfo *tunp; 544 545 switch (cmd) { 546 case TUNSIFINFO: 547 tunp = (struct tuninfo *)data; 548 if (tunp->mtu < IF_MINMTU) 549 return (EINVAL); 550 if (tp->tun_if.if_mtu != tunp->mtu 551 && (error = suser(td)) != 0) 552 return (error); 553 tp->tun_if.if_mtu = tunp->mtu; 554 tp->tun_if.if_type = tunp->type; 555 tp->tun_if.if_baudrate = tunp->baudrate; 556 break; 557 case TUNGIFINFO: 558 tunp = (struct tuninfo *)data; 559 tunp->mtu = tp->tun_if.if_mtu; 560 tunp->type = tp->tun_if.if_type; 561 tunp->baudrate = tp->tun_if.if_baudrate; 562 break; 563 case TUNSDEBUG: 564 tundebug = *(int *)data; 565 break; 566 case TUNGDEBUG: 567 *(int *)data = tundebug; 568 break; 569 case TUNSLMODE: 570 mtx_lock(&tp->tun_mtx); 571 if (*(int *)data) { 572 tp->tun_flags |= TUN_LMODE; 573 tp->tun_flags &= ~TUN_IFHEAD; 574 } else 575 tp->tun_flags &= ~TUN_LMODE; 576 mtx_unlock(&tp->tun_mtx); 577 break; 578 case TUNSIFHEAD: 579 mtx_lock(&tp->tun_mtx); 580 if (*(int *)data) { 581 tp->tun_flags |= TUN_IFHEAD; 582 tp->tun_flags &= ~TUN_LMODE; 583 } else 584 tp->tun_flags &= ~TUN_IFHEAD; 585 mtx_unlock(&tp->tun_mtx); 586 break; 587 case TUNGIFHEAD: 588 /* Could be unlocked read? */ 589 mtx_lock(&tp->tun_mtx); 590 *(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0; 591 mtx_unlock(&tp->tun_mtx); 592 break; 593 case TUNSIFMODE: 594 /* deny this if UP */ 595 if (tp->tun_if.if_flags & IFF_UP) 596 return(EBUSY); 597 598 switch (*(int *)data & ~IFF_MULTICAST) { 599 case IFF_POINTOPOINT: 600 case IFF_BROADCAST: 601 tp->tun_if.if_flags &= 602 ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST); 603 tp->tun_if.if_flags |= *(int *)data; 604 break; 605 default: 606 return(EINVAL); 607 } 608 break; 609 case TUNSIFPID: 610 mtx_lock(&tp->tun_mtx); 611 tp->tun_pid = curthread->td_proc->p_pid; 612 mtx_unlock(&tp->tun_mtx); 613 break; 614 case FIONBIO: 615 break; 616 case FIOASYNC: 617 mtx_lock(&tp->tun_mtx); 618 if (*(int *)data) 619 tp->tun_flags |= TUN_ASYNC; 620 else 621 tp->tun_flags &= ~TUN_ASYNC; 622 mtx_unlock(&tp->tun_mtx); 623 break; 624 case FIONREAD: 625 s = splimp(); 626 if (tp->tun_if.if_snd.ifq_head) { 627 struct mbuf *mb = tp->tun_if.if_snd.ifq_head; 628 for( *(int *)data = 0; mb != 0; mb = mb->m_next) 629 *(int *)data += mb->m_len; 630 } else 631 *(int *)data = 0; 632 splx(s); 633 break; 634 case FIOSETOWN: 635 return (fsetown(*(int *)data, &tp->tun_sigio)); 636 637 case FIOGETOWN: 638 *(int *)data = fgetown(&tp->tun_sigio); 639 return (0); 640 641 /* This is deprecated, FIOSETOWN should be used instead. */ 642 case TIOCSPGRP: 643 return (fsetown(-(*(int *)data), &tp->tun_sigio)); 644 645 /* This is deprecated, FIOGETOWN should be used instead. */ 646 case TIOCGPGRP: 647 *(int *)data = -fgetown(&tp->tun_sigio); 648 return (0); 649 650 default: 651 return (ENOTTY); 652 } 653 return (0); 654} 655 656/* 657 * The cdevsw read interface - reads a packet at a time, or at 658 * least as much of a packet as can be read. 659 */ 660static int 661tunread(struct cdev *dev, struct uio *uio, int flag) 662{ 663 struct tun_softc *tp = dev->si_drv1; 664 struct ifnet *ifp = &tp->tun_if; 665 struct mbuf *m; 666 int error=0, len, s; 667 668 TUNDEBUG (ifp, "read\n"); 669 mtx_lock(&tp->tun_mtx); 670 if ((tp->tun_flags & TUN_READY) != TUN_READY) { 671 mtx_unlock(&tp->tun_mtx); 672 TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags); 673 return (EHOSTDOWN); 674 } 675 676 tp->tun_flags &= ~TUN_RWAIT; 677 mtx_unlock(&tp->tun_mtx); 678 679 s = splimp(); 680 do { 681 IF_DEQUEUE(&ifp->if_snd, m); 682 if (m == NULL) { 683 if (flag & IO_NDELAY) { 684 splx(s); 685 return (EWOULDBLOCK); 686 } 687 mtx_lock(&tp->tun_mtx); 688 tp->tun_flags |= TUN_RWAIT; 689 mtx_unlock(&tp->tun_mtx); 690 if((error = tsleep(tp, PCATCH | (PZERO + 1), 691 "tunread", 0)) != 0) { 692 splx(s); 693 return (error); 694 } 695 } 696 } while (m == NULL); 697 splx(s); 698 699 while (m && uio->uio_resid > 0 && error == 0) { 700 len = min(uio->uio_resid, m->m_len); 701 if (len != 0) 702 error = uiomove(mtod(m, void *), len, uio); 703 m = m_free(m); 704 } 705 706 if (m) { 707 TUNDEBUG(ifp, "Dropping mbuf\n"); 708 m_freem(m); 709 } 710 return (error); 711} 712 713/* 714 * the cdevsw write interface - an atomic write is a packet - or else! 715 */ 716static int 717tunwrite(struct cdev *dev, struct uio *uio, int flag) 718{ 719 struct tun_softc *tp = dev->si_drv1; 720 struct ifnet *ifp = &tp->tun_if; 721 struct mbuf *top, **mp, *m; 722 int error=0, tlen, mlen; 723 uint32_t family; 724 int isr; 725 726 TUNDEBUG(ifp, "tunwrite\n"); 727 728 if ((ifp->if_flags & IFF_UP) != IFF_UP) 729 /* ignore silently */ 730 return (0); 731 732 if (uio->uio_resid == 0) 733 return (0); 734 735 if (uio->uio_resid < 0 || uio->uio_resid > TUNMRU) { 736 TUNDEBUG(ifp, "len=%d!\n", uio->uio_resid); 737 return (EIO); 738 } 739 tlen = uio->uio_resid; 740 741 /* get a header mbuf */ 742 MGETHDR(m, M_DONTWAIT, MT_DATA); 743 if (m == NULL) 744 return (ENOBUFS); 745 mlen = MHLEN; 746 747 top = 0; 748 mp = ⊤ 749 while (error == 0 && uio->uio_resid > 0) { 750 m->m_len = min(mlen, uio->uio_resid); 751 error = uiomove(mtod(m, void *), m->m_len, uio); 752 *mp = m; 753 mp = &m->m_next; 754 if (uio->uio_resid > 0) { 755 MGET (m, M_DONTWAIT, MT_DATA); 756 if (m == 0) { 757 error = ENOBUFS; 758 break; 759 } 760 mlen = MLEN; 761 } 762 } 763 if (error) { 764 if (top) 765 m_freem (top); 766 ifp->if_ierrors++; 767 return (error); 768 } 769 770 top->m_pkthdr.len = tlen; 771 top->m_pkthdr.rcvif = ifp; 772#ifdef MAC 773 mac_create_mbuf_from_ifnet(ifp, top); 774#endif 775 776 /* Could be unlocked read? */ 777 mtx_lock(&tp->tun_mtx); 778 if (tp->tun_flags & TUN_IFHEAD) { 779 mtx_unlock(&tp->tun_mtx); 780 if (top->m_len < sizeof(family) && 781 (top = m_pullup(top, sizeof(family))) == NULL) 782 return (ENOBUFS); 783 family = ntohl(*mtod(top, u_int32_t *)); 784 m_adj(top, sizeof(family)); 785 } else { 786 mtx_unlock(&tp->tun_mtx); 787 family = AF_INET; 788 } 789 790 BPF_MTAP2(ifp, &family, sizeof(family), top); 791 792 switch (family) { 793#ifdef INET 794 case AF_INET: 795 isr = NETISR_IP; 796 break; 797#endif 798#ifdef INET6 799 case AF_INET6: 800 isr = NETISR_IPV6; 801 break; 802#endif 803#ifdef IPX 804 case AF_IPX: 805 isr = NETISR_IPX; 806 break; 807#endif 808#ifdef NETATALK 809 case AF_APPLETALK: 810 isr = NETISR_ATALK2; 811 break; 812#endif 813 default: 814 m_freem(m); 815 return (EAFNOSUPPORT); 816 } 817 /* First chunk of an mbuf contains good junk */ 818 if (harvest.point_to_point) 819 random_harvest(m, 16, 3, 0, RANDOM_NET); 820 ifp->if_ibytes += top->m_pkthdr.len; 821 ifp->if_ipackets++; 822 netisr_dispatch(isr, top); 823 return (0); 824} 825 826/* 827 * tunpoll - the poll interface, this is only useful on reads 828 * really. The write detect always returns true, write never blocks 829 * anyway, it either accepts the packet or drops it. 830 */ 831static int 832tunpoll(struct cdev *dev, int events, struct thread *td) 833{ 834 int s; 835 struct tun_softc *tp = dev->si_drv1; 836 struct ifnet *ifp = &tp->tun_if; 837 int revents = 0; 838 839 s = splimp(); 840 TUNDEBUG(ifp, "tunpoll\n"); 841 842 if (events & (POLLIN | POLLRDNORM)) { 843 if (ifp->if_snd.ifq_len > 0) { 844 TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len); 845 revents |= events & (POLLIN | POLLRDNORM); 846 } else { 847 TUNDEBUG(ifp, "tunpoll waiting\n"); 848 selrecord(td, &tp->tun_rsel); 849 } 850 } 851 if (events & (POLLOUT | POLLWRNORM)) 852 revents |= events & (POLLOUT | POLLWRNORM); 853 854 splx(s); 855 return (revents); 856} 857