1 /* $NetBSD: if_tun.c,v 1.120 2014/07/25 08:10:40 dholland 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 its 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 17 #include <sys/cdefs.h> 18 __KERNEL_RCSID(0, "$NetBSD: if_tun.c,v 1.120 2014/07/25 08:10:40 dholland Exp $"); 19 20 #include "opt_inet.h" 21 22 #include <sys/param.h> 23 #include <sys/proc.h> 24 #include <sys/systm.h> 25 #include <sys/mbuf.h> 26 #include <sys/buf.h> 27 #include <sys/protosw.h> 28 #include <sys/socket.h> 29 #include <sys/ioctl.h> 30 #include <sys/errno.h> 31 #include <sys/syslog.h> 32 #include <sys/select.h> 33 #include <sys/poll.h> 34 #include <sys/file.h> 35 #include <sys/signalvar.h> 36 #include <sys/conf.h> 37 #include <sys/kauth.h> 38 #include <sys/mutex.h> 39 #include <sys/cpu.h> 40 41 #include <net/if.h> 42 #include <net/if_types.h> 43 #include <net/netisr.h> 44 #include <net/route.h> 45 46 47 #ifdef INET 48 #include <netinet/in.h> 49 #include <netinet/in_systm.h> 50 #include <netinet/in_var.h> 51 #include <netinet/ip.h> 52 #include <netinet/if_inarp.h> 53 #endif 54 55 56 #include <sys/time.h> 57 #include <net/bpf.h> 58 59 #include <net/if_tun.h> 60 61 #define TUNDEBUG if (tundebug) printf 62 int tundebug = 0; 63 64 extern int ifqmaxlen; 65 void tunattach(int); 66 67 static LIST_HEAD(, tun_softc) tun_softc_list; 68 static LIST_HEAD(, tun_softc) tunz_softc_list; 69 static kmutex_t tun_softc_lock; 70 71 static int tun_ioctl(struct ifnet *, u_long, void *); 72 static int tun_output(struct ifnet *, struct mbuf *, 73 const struct sockaddr *, struct rtentry *rt); 74 static int tun_clone_create(struct if_clone *, int); 75 static int tun_clone_destroy(struct ifnet *); 76 77 static struct if_clone tun_cloner = 78 IF_CLONE_INITIALIZER("tun", tun_clone_create, tun_clone_destroy); 79 80 static void tunattach0(struct tun_softc *); 81 static void tuninit(struct tun_softc *); 82 static void tun_i_softintr(void *); 83 static void tun_o_softintr(void *); 84 #ifdef ALTQ 85 static void tunstart(struct ifnet *); 86 #endif 87 static struct tun_softc *tun_find_unit(dev_t); 88 static struct tun_softc *tun_find_zunit(int); 89 90 static dev_type_open(tunopen); 91 static dev_type_close(tunclose); 92 static dev_type_read(tunread); 93 static dev_type_write(tunwrite); 94 static dev_type_ioctl(tunioctl); 95 static dev_type_poll(tunpoll); 96 static dev_type_kqfilter(tunkqfilter); 97 98 const struct cdevsw tun_cdevsw = { 99 .d_open = tunopen, 100 .d_close = tunclose, 101 .d_read = tunread, 102 .d_write = tunwrite, 103 .d_ioctl = tunioctl, 104 .d_stop = nostop, 105 .d_tty = notty, 106 .d_poll = tunpoll, 107 .d_mmap = nommap, 108 .d_kqfilter = tunkqfilter, 109 .d_discard = nodiscard, 110 .d_flag = D_OTHER 111 }; 112 113 void 114 tunattach(int unused) 115 { 116 117 mutex_init(&tun_softc_lock, MUTEX_DEFAULT, IPL_NET); 118 LIST_INIT(&tun_softc_list); 119 LIST_INIT(&tunz_softc_list); 120 if_clone_attach(&tun_cloner); 121 } 122 123 /* 124 * Find driver instance from dev_t. 125 * Returns with tp locked (if found). 126 */ 127 static struct tun_softc * 128 tun_find_unit(dev_t dev) 129 { 130 struct tun_softc *tp; 131 int unit = minor(dev); 132 133 mutex_enter(&tun_softc_lock); 134 LIST_FOREACH(tp, &tun_softc_list, tun_list) 135 if (unit == tp->tun_unit) 136 break; 137 if (tp) 138 mutex_enter(&tp->tun_lock); 139 mutex_exit(&tun_softc_lock); 140 141 return (tp); 142 } 143 144 /* 145 * Find zombie driver instance by unit number. 146 * Remove tp from list and return it unlocked (if found). 147 */ 148 static struct tun_softc * 149 tun_find_zunit(int unit) 150 { 151 struct tun_softc *tp; 152 153 mutex_enter(&tun_softc_lock); 154 LIST_FOREACH(tp, &tunz_softc_list, tun_list) 155 if (unit == tp->tun_unit) 156 break; 157 if (tp) 158 LIST_REMOVE(tp, tun_list); 159 mutex_exit(&tun_softc_lock); 160 #ifdef DIAGNOSTIC 161 if (tp != NULL && (tp->tun_flags & (TUN_INITED|TUN_OPEN)) != TUN_OPEN) 162 printf("tun%d: inconsistent flags: %x\n", unit, tp->tun_flags); 163 #endif 164 165 return (tp); 166 } 167 168 static int 169 tun_clone_create(struct if_clone *ifc, int unit) 170 { 171 struct tun_softc *tp; 172 173 if ((tp = tun_find_zunit(unit)) == NULL) { 174 /* Allocate a new instance */ 175 tp = malloc(sizeof(*tp), M_DEVBUF, M_WAITOK|M_ZERO); 176 177 tp->tun_unit = unit; 178 mutex_init(&tp->tun_lock, MUTEX_DEFAULT, IPL_NET); 179 selinit(&tp->tun_rsel); 180 selinit(&tp->tun_wsel); 181 } else { 182 /* Revive tunnel instance; clear ifp part */ 183 (void)memset(&tp->tun_if, 0, sizeof(struct ifnet)); 184 } 185 186 if_initname(&tp->tun_if, ifc->ifc_name, unit); 187 tunattach0(tp); 188 tp->tun_flags |= TUN_INITED; 189 tp->tun_osih = softint_establish(SOFTINT_CLOCK, tun_o_softintr, tp); 190 tp->tun_isih = softint_establish(SOFTINT_CLOCK, tun_i_softintr, tp); 191 192 mutex_enter(&tun_softc_lock); 193 LIST_INSERT_HEAD(&tun_softc_list, tp, tun_list); 194 mutex_exit(&tun_softc_lock); 195 196 return (0); 197 } 198 199 static void 200 tunattach0(struct tun_softc *tp) 201 { 202 struct ifnet *ifp; 203 204 ifp = &tp->tun_if; 205 ifp->if_softc = tp; 206 ifp->if_mtu = TUNMTU; 207 ifp->if_ioctl = tun_ioctl; 208 ifp->if_output = tun_output; 209 #ifdef ALTQ 210 ifp->if_start = tunstart; 211 #endif 212 ifp->if_flags = IFF_POINTOPOINT; 213 ifp->if_type = IFT_TUNNEL; 214 ifp->if_snd.ifq_maxlen = ifqmaxlen; 215 ifp->if_collisions = 0; 216 ifp->if_ierrors = 0; 217 ifp->if_oerrors = 0; 218 ifp->if_ipackets = 0; 219 ifp->if_opackets = 0; 220 ifp->if_ibytes = 0; 221 ifp->if_obytes = 0; 222 ifp->if_dlt = DLT_NULL; 223 IFQ_SET_READY(&ifp->if_snd); 224 if_attach(ifp); 225 if_alloc_sadl(ifp); 226 bpf_attach(ifp, DLT_NULL, sizeof(uint32_t)); 227 } 228 229 static int 230 tun_clone_destroy(struct ifnet *ifp) 231 { 232 struct tun_softc *tp = (void *)ifp; 233 int zombie = 0; 234 235 IF_PURGE(&ifp->if_snd); 236 ifp->if_flags &= ~IFF_RUNNING; 237 238 mutex_enter(&tun_softc_lock); 239 mutex_enter(&tp->tun_lock); 240 LIST_REMOVE(tp, tun_list); 241 if (tp->tun_flags & TUN_OPEN) { 242 /* Hang on to storage until last close */ 243 zombie = 1; 244 tp->tun_flags &= ~TUN_INITED; 245 LIST_INSERT_HEAD(&tunz_softc_list, tp, tun_list); 246 } 247 mutex_exit(&tun_softc_lock); 248 249 if (tp->tun_flags & TUN_RWAIT) { 250 tp->tun_flags &= ~TUN_RWAIT; 251 wakeup((void *)tp); 252 } 253 selnotify(&tp->tun_rsel, 0, 0); 254 255 mutex_exit(&tp->tun_lock); 256 257 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 258 fownsignal(tp->tun_pgid, SIGIO, POLL_HUP, 0, NULL); 259 260 bpf_detach(ifp); 261 if_detach(ifp); 262 263 if (!zombie) { 264 seldestroy(&tp->tun_rsel); 265 seldestroy(&tp->tun_wsel); 266 softint_disestablish(tp->tun_osih); 267 softint_disestablish(tp->tun_isih); 268 mutex_destroy(&tp->tun_lock); 269 free(tp, M_DEVBUF); 270 } 271 272 return (0); 273 } 274 275 /* 276 * tunnel open - must be superuser & the device must be 277 * configured in 278 */ 279 static int 280 tunopen(dev_t dev, int flag, int mode, struct lwp *l) 281 { 282 struct ifnet *ifp; 283 struct tun_softc *tp; 284 int error; 285 286 error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE_TUN, 287 KAUTH_REQ_NETWORK_INTERFACE_TUN_ADD, NULL, NULL, NULL); 288 if (error) 289 return (error); 290 291 tp = tun_find_unit(dev); 292 293 if (tp == NULL) { 294 (void)tun_clone_create(&tun_cloner, minor(dev)); 295 tp = tun_find_unit(dev); 296 if (tp == NULL) { 297 error = ENXIO; 298 goto out_nolock; 299 } 300 } 301 302 if (tp->tun_flags & TUN_OPEN) { 303 error = EBUSY; 304 goto out; 305 } 306 307 ifp = &tp->tun_if; 308 tp->tun_flags |= TUN_OPEN; 309 TUNDEBUG("%s: open\n", ifp->if_xname); 310 out: 311 mutex_exit(&tp->tun_lock); 312 out_nolock: 313 return (error); 314 } 315 316 /* 317 * tunclose - close the device - mark i/f down & delete 318 * routing info 319 */ 320 int 321 tunclose(dev_t dev, int flag, int mode, 322 struct lwp *l) 323 { 324 struct tun_softc *tp; 325 struct ifnet *ifp; 326 327 if ((tp = tun_find_zunit(minor(dev))) != NULL) { 328 /* interface was "destroyed" before the close */ 329 seldestroy(&tp->tun_rsel); 330 seldestroy(&tp->tun_wsel); 331 softint_disestablish(tp->tun_osih); 332 softint_disestablish(tp->tun_isih); 333 mutex_destroy(&tp->tun_lock); 334 free(tp, M_DEVBUF); 335 goto out_nolock; 336 } 337 338 if ((tp = tun_find_unit(dev)) == NULL) 339 goto out_nolock; 340 341 ifp = &tp->tun_if; 342 343 tp->tun_flags &= ~TUN_OPEN; 344 345 tp->tun_pgid = 0; 346 selnotify(&tp->tun_rsel, 0, 0); 347 348 TUNDEBUG ("%s: closed\n", ifp->if_xname); 349 mutex_exit(&tp->tun_lock); 350 351 /* 352 * junk all pending output 353 */ 354 IFQ_PURGE(&ifp->if_snd); 355 356 if (ifp->if_flags & IFF_UP) { 357 if_down(ifp); 358 if (ifp->if_flags & IFF_RUNNING) { 359 /* find internet addresses and delete routes */ 360 struct ifaddr *ifa; 361 IFADDR_FOREACH(ifa, ifp) { 362 #if defined(INET) || defined(INET6) 363 if (ifa->ifa_addr->sa_family == AF_INET || 364 ifa->ifa_addr->sa_family == AF_INET6) { 365 rtinit(ifa, (int)RTM_DELETE, 366 tp->tun_flags & TUN_DSTADDR 367 ? RTF_HOST 368 : 0); 369 } 370 #endif 371 } 372 } 373 } 374 out_nolock: 375 return (0); 376 } 377 378 /* 379 * Call at splnet(). 380 */ 381 static void 382 tuninit(struct tun_softc *tp) 383 { 384 struct ifnet *ifp = &tp->tun_if; 385 struct ifaddr *ifa; 386 387 TUNDEBUG("%s: tuninit\n", ifp->if_xname); 388 389 mutex_enter(&tp->tun_lock); 390 ifp->if_flags |= IFF_UP | IFF_RUNNING; 391 392 tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR); 393 IFADDR_FOREACH(ifa, ifp) { 394 #ifdef INET 395 if (ifa->ifa_addr->sa_family == AF_INET) { 396 struct sockaddr_in *sin; 397 398 sin = satosin(ifa->ifa_addr); 399 if (sin && sin->sin_addr.s_addr) 400 tp->tun_flags |= TUN_IASET; 401 402 if (ifp->if_flags & IFF_POINTOPOINT) { 403 sin = satosin(ifa->ifa_dstaddr); 404 if (sin && sin->sin_addr.s_addr) 405 tp->tun_flags |= TUN_DSTADDR; 406 } 407 } 408 #endif 409 #ifdef INET6 410 if (ifa->ifa_addr->sa_family == AF_INET6) { 411 struct sockaddr_in6 *sin; 412 413 sin = (struct sockaddr_in6 *)ifa->ifa_addr; 414 if (!IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr)) 415 tp->tun_flags |= TUN_IASET; 416 417 if (ifp->if_flags & IFF_POINTOPOINT) { 418 sin = (struct sockaddr_in6 *)ifa->ifa_dstaddr; 419 if (sin && 420 !IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr)) 421 tp->tun_flags |= TUN_DSTADDR; 422 } else 423 tp->tun_flags &= ~TUN_DSTADDR; 424 } 425 #endif /* INET6 */ 426 } 427 mutex_exit(&tp->tun_lock); 428 } 429 430 /* 431 * Process an ioctl request. 432 */ 433 static int 434 tun_ioctl(struct ifnet *ifp, u_long cmd, void *data) 435 { 436 int error = 0, s; 437 struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc); 438 struct ifreq *ifr = data; 439 440 s = splnet(); 441 442 switch (cmd) { 443 case SIOCINITIFADDR: 444 tuninit(tp); 445 TUNDEBUG("%s: address set\n", ifp->if_xname); 446 break; 447 case SIOCSIFBRDADDR: 448 TUNDEBUG("%s: broadcast address set\n", ifp->if_xname); 449 break; 450 case SIOCSIFMTU: 451 if (ifr->ifr_mtu > TUNMTU || ifr->ifr_mtu < 576) { 452 error = EINVAL; 453 break; 454 } 455 TUNDEBUG("%s: interface mtu set\n", ifp->if_xname); 456 if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET) 457 error = 0; 458 break; 459 case SIOCADDMULTI: 460 case SIOCDELMULTI: 461 if (ifr == NULL) { 462 error = EAFNOSUPPORT; /* XXX */ 463 break; 464 } 465 switch (ifreq_getaddr(cmd, ifr)->sa_family) { 466 #ifdef INET 467 case AF_INET: 468 break; 469 #endif 470 #ifdef INET6 471 case AF_INET6: 472 break; 473 #endif 474 default: 475 error = EAFNOSUPPORT; 476 break; 477 } 478 break; 479 default: 480 error = ifioctl_common(ifp, cmd, data); 481 } 482 483 splx(s); 484 return (error); 485 } 486 487 /* 488 * tun_output - queue packets from higher level ready to put out. 489 */ 490 static int 491 tun_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst, 492 struct rtentry *rt) 493 { 494 struct tun_softc *tp = ifp->if_softc; 495 int s; 496 int error; 497 #if defined(INET) || defined(INET6) 498 int mlen; 499 uint32_t *af; 500 #endif 501 ALTQ_DECL(struct altq_pktattr pktattr;) 502 503 s = splnet(); 504 mutex_enter(&tp->tun_lock); 505 TUNDEBUG ("%s: tun_output\n", ifp->if_xname); 506 507 if ((tp->tun_flags & TUN_READY) != TUN_READY) { 508 TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, 509 tp->tun_flags); 510 error = EHOSTDOWN; 511 goto out; 512 } 513 514 /* 515 * if the queueing discipline needs packet classification, 516 * do it before prepending link headers. 517 */ 518 IFQ_CLASSIFY(&ifp->if_snd, m0, dst->sa_family, &pktattr); 519 520 bpf_mtap_af(ifp, dst->sa_family, m0); 521 522 switch(dst->sa_family) { 523 #ifdef INET6 524 case AF_INET6: 525 #endif 526 #ifdef INET 527 case AF_INET: 528 #endif 529 #if defined(INET) || defined(INET6) 530 if (tp->tun_flags & TUN_PREPADDR) { 531 /* Simple link-layer header */ 532 M_PREPEND(m0, dst->sa_len, M_DONTWAIT); 533 if (m0 == NULL) { 534 IF_DROP(&ifp->if_snd); 535 error = ENOBUFS; 536 goto out; 537 } 538 bcopy(dst, mtod(m0, char *), dst->sa_len); 539 } 540 541 if (tp->tun_flags & TUN_IFHEAD) { 542 /* Prepend the address family */ 543 M_PREPEND(m0, sizeof(*af), M_DONTWAIT); 544 if (m0 == NULL) { 545 IF_DROP(&ifp->if_snd); 546 error = ENOBUFS; 547 goto out; 548 } 549 af = mtod(m0,uint32_t *); 550 *af = htonl(dst->sa_family); 551 } else { 552 #ifdef INET 553 if (dst->sa_family != AF_INET) 554 #endif 555 { 556 error = EAFNOSUPPORT; 557 goto out; 558 } 559 } 560 /* FALLTHROUGH */ 561 case AF_UNSPEC: 562 IFQ_ENQUEUE(&ifp->if_snd, m0, &pktattr, error); 563 if (error) { 564 ifp->if_collisions++; 565 error = EAFNOSUPPORT; 566 m0 = NULL; 567 goto out; 568 } 569 mlen = m0->m_pkthdr.len; 570 ifp->if_opackets++; 571 ifp->if_obytes += mlen; 572 break; 573 #endif 574 default: 575 error = EAFNOSUPPORT; 576 goto out; 577 } 578 579 if (tp->tun_flags & TUN_RWAIT) { 580 tp->tun_flags &= ~TUN_RWAIT; 581 wakeup((void *)tp); 582 } 583 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 584 softint_schedule(tp->tun_isih); 585 586 selnotify(&tp->tun_rsel, 0, 0); 587 out: 588 mutex_exit(&tp->tun_lock); 589 splx(s); 590 591 if (error && m0) { 592 m_freem(m0); 593 } 594 return 0; 595 } 596 597 static void 598 tun_i_softintr(void *cookie) 599 { 600 struct tun_softc *tp = cookie; 601 602 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 603 fownsignal(tp->tun_pgid, SIGIO, POLL_IN, POLLIN|POLLRDNORM, 604 NULL); 605 } 606 607 static void 608 tun_o_softintr(void *cookie) 609 { 610 struct tun_softc *tp = cookie; 611 612 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 613 fownsignal(tp->tun_pgid, SIGIO, POLL_OUT, POLLOUT|POLLWRNORM, 614 NULL); 615 } 616 617 /* 618 * the cdevsw interface is now pretty minimal. 619 */ 620 int 621 tunioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 622 { 623 struct tun_softc *tp; 624 int s, error = 0; 625 626 s = splnet(); 627 tp = tun_find_unit(dev); 628 629 /* interface was "destroyed" already */ 630 if (tp == NULL) { 631 error = ENXIO; 632 goto out_nolock; 633 } 634 635 switch (cmd) { 636 case TUNSDEBUG: 637 tundebug = *(int *)data; 638 break; 639 640 case TUNGDEBUG: 641 *(int *)data = tundebug; 642 break; 643 644 case TUNSIFMODE: 645 switch (*(int *)data & (IFF_POINTOPOINT|IFF_BROADCAST)) { 646 case IFF_POINTOPOINT: 647 case IFF_BROADCAST: 648 if (tp->tun_if.if_flags & IFF_UP) { 649 error = EBUSY; 650 goto out; 651 } 652 tp->tun_if.if_flags &= 653 ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST); 654 tp->tun_if.if_flags |= *(int *)data; 655 break; 656 default: 657 error = EINVAL; 658 goto out; 659 } 660 break; 661 662 case TUNSLMODE: 663 if (*(int *)data) { 664 tp->tun_flags |= TUN_PREPADDR; 665 tp->tun_flags &= ~TUN_IFHEAD; 666 } else 667 tp->tun_flags &= ~TUN_PREPADDR; 668 break; 669 670 case TUNSIFHEAD: 671 if (*(int *)data) { 672 tp->tun_flags |= TUN_IFHEAD; 673 tp->tun_flags &= ~TUN_PREPADDR; 674 } else 675 tp->tun_flags &= ~TUN_IFHEAD; 676 break; 677 678 case TUNGIFHEAD: 679 *(int *)data = (tp->tun_flags & TUN_IFHEAD); 680 break; 681 682 case FIONBIO: 683 if (*(int *)data) 684 tp->tun_flags |= TUN_NBIO; 685 else 686 tp->tun_flags &= ~TUN_NBIO; 687 break; 688 689 case FIOASYNC: 690 if (*(int *)data) 691 tp->tun_flags |= TUN_ASYNC; 692 else 693 tp->tun_flags &= ~TUN_ASYNC; 694 break; 695 696 case FIONREAD: 697 if (tp->tun_if.if_snd.ifq_head) 698 *(int *)data = tp->tun_if.if_snd.ifq_head->m_pkthdr.len; 699 else 700 *(int *)data = 0; 701 break; 702 703 case TIOCSPGRP: 704 case FIOSETOWN: 705 error = fsetown(&tp->tun_pgid, cmd, data); 706 break; 707 708 case TIOCGPGRP: 709 case FIOGETOWN: 710 error = fgetown(tp->tun_pgid, cmd, data); 711 break; 712 713 default: 714 error = ENOTTY; 715 } 716 717 out: 718 mutex_exit(&tp->tun_lock); 719 out_nolock: 720 splx(s); 721 return (error); 722 } 723 724 /* 725 * The cdevsw read interface - reads a packet at a time, or at 726 * least as much of a packet as can be read. 727 */ 728 int 729 tunread(dev_t dev, struct uio *uio, int ioflag) 730 { 731 struct tun_softc *tp; 732 struct ifnet *ifp; 733 struct mbuf *m, *m0; 734 int error = 0, len, s, index; 735 736 s = splnet(); 737 tp = tun_find_unit(dev); 738 739 /* interface was "destroyed" already */ 740 if (tp == NULL) { 741 error = ENXIO; 742 goto out_nolock; 743 } 744 745 index = tp->tun_if.if_index; 746 ifp = &tp->tun_if; 747 748 TUNDEBUG ("%s: read\n", ifp->if_xname); 749 if ((tp->tun_flags & TUN_READY) != TUN_READY) { 750 TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags); 751 error = EHOSTDOWN; 752 goto out; 753 } 754 755 tp->tun_flags &= ~TUN_RWAIT; 756 757 do { 758 IFQ_DEQUEUE(&ifp->if_snd, m0); 759 if (m0 == 0) { 760 if (tp->tun_flags & TUN_NBIO) { 761 error = EWOULDBLOCK; 762 goto out; 763 } 764 tp->tun_flags |= TUN_RWAIT; 765 if (mtsleep((void *)tp, PZERO|PCATCH|PNORELOCK, 766 "tunread", 0, &tp->tun_lock) != 0) { 767 error = EINTR; 768 goto out_nolock; 769 } else { 770 /* 771 * Maybe the interface was destroyed while 772 * we were sleeping, so let's ensure that 773 * we're looking at the same (valid) tun 774 * interface before looping. 775 */ 776 tp = tun_find_unit(dev); 777 if (tp == NULL) { 778 error = ENXIO; 779 goto out_nolock; 780 } 781 if (tp->tun_if.if_index != index) { 782 error = ENXIO; 783 goto out; 784 } 785 } 786 } 787 } while (m0 == 0); 788 789 mutex_exit(&tp->tun_lock); 790 splx(s); 791 792 /* Copy the mbuf chain */ 793 while (m0 && uio->uio_resid > 0 && error == 0) { 794 len = min(uio->uio_resid, m0->m_len); 795 if (len != 0) 796 error = uiomove(mtod(m0, void *), len, uio); 797 MFREE(m0, m); 798 m0 = m; 799 } 800 801 if (m0) { 802 TUNDEBUG("Dropping mbuf\n"); 803 m_freem(m0); 804 } 805 if (error) 806 ifp->if_ierrors++; 807 808 return (error); 809 810 out: 811 mutex_exit(&tp->tun_lock); 812 out_nolock: 813 splx(s); 814 return (error); 815 } 816 817 /* 818 * the cdevsw write interface - an atomic write is a packet - or else! 819 */ 820 int 821 tunwrite(dev_t dev, struct uio *uio, int ioflag) 822 { 823 struct tun_softc *tp; 824 struct ifnet *ifp; 825 struct mbuf *top, **mp, *m; 826 pktqueue_t *pktq; 827 struct sockaddr dst; 828 int error = 0, s, tlen, mlen; 829 uint32_t family; 830 831 s = splnet(); 832 tp = tun_find_unit(dev); 833 834 /* interface was "destroyed" already */ 835 if (tp == NULL) { 836 error = ENXIO; 837 goto out_nolock; 838 } 839 840 /* Unlock until we've got the data */ 841 mutex_exit(&tp->tun_lock); 842 splx(s); 843 844 ifp = &tp->tun_if; 845 846 TUNDEBUG("%s: tunwrite\n", ifp->if_xname); 847 848 if (tp->tun_flags & TUN_PREPADDR) { 849 if (uio->uio_resid < sizeof(dst)) { 850 error = EIO; 851 goto out0; 852 } 853 error = uiomove((void *)&dst, sizeof(dst), uio); 854 if (dst.sa_len > sizeof(dst)) { 855 /* Duh.. */ 856 char discard; 857 int n = dst.sa_len - sizeof(dst); 858 while (n--) 859 if ((error = uiomove(&discard, 1, uio)) != 0) { 860 goto out0; 861 } 862 } 863 } else if (tp->tun_flags & TUN_IFHEAD) { 864 if (uio->uio_resid < sizeof(family)){ 865 error = EIO; 866 goto out0; 867 } 868 error = uiomove((void *)&family, sizeof(family), uio); 869 dst.sa_family = ntohl(family); 870 } else { 871 #ifdef INET 872 dst.sa_family = AF_INET; 873 #endif 874 } 875 876 if (uio->uio_resid > TUNMTU) { 877 TUNDEBUG("%s: len=%lu!\n", ifp->if_xname, 878 (unsigned long)uio->uio_resid); 879 error = EIO; 880 goto out0; 881 } 882 883 switch (dst.sa_family) { 884 #ifdef INET 885 case AF_INET: 886 pktq = ip_pktq; 887 break; 888 #endif 889 #ifdef INET6 890 case AF_INET6: 891 pktq = ip6_pktq; 892 break; 893 #endif 894 default: 895 error = EAFNOSUPPORT; 896 goto out0; 897 } 898 899 tlen = uio->uio_resid; 900 901 /* get a header mbuf */ 902 MGETHDR(m, M_DONTWAIT, MT_DATA); 903 if (m == NULL) { 904 error = ENOBUFS; 905 goto out0; 906 } 907 mlen = MHLEN; 908 909 top = NULL; 910 mp = ⊤ 911 while (error == 0 && uio->uio_resid > 0) { 912 m->m_len = min(mlen, uio->uio_resid); 913 error = uiomove(mtod(m, void *), m->m_len, uio); 914 *mp = m; 915 mp = &m->m_next; 916 if (error == 0 && uio->uio_resid > 0) { 917 MGET(m, M_DONTWAIT, MT_DATA); 918 if (m == NULL) { 919 error = ENOBUFS; 920 break; 921 } 922 mlen = MLEN; 923 } 924 } 925 if (error) { 926 if (top != NULL) 927 m_freem (top); 928 ifp->if_ierrors++; 929 goto out0; 930 } 931 932 top->m_pkthdr.len = tlen; 933 top->m_pkthdr.rcvif = ifp; 934 935 bpf_mtap_af(ifp, dst.sa_family, top); 936 937 s = splnet(); 938 mutex_enter(&tp->tun_lock); 939 if ((tp->tun_flags & TUN_INITED) == 0) { 940 /* Interface was destroyed */ 941 error = ENXIO; 942 goto out; 943 } 944 if (__predict_false(!pktq_enqueue(pktq, top, 0))) { 945 ifp->if_collisions++; 946 mutex_exit(&tp->tun_lock); 947 error = ENOBUFS; 948 m_freem(top); 949 goto out_nolock; 950 } 951 ifp->if_ipackets++; 952 ifp->if_ibytes += tlen; 953 out: 954 mutex_exit(&tp->tun_lock); 955 out_nolock: 956 splx(s); 957 out0: 958 return (error); 959 } 960 961 #ifdef ALTQ 962 /* 963 * Start packet transmission on the interface. 964 * when the interface queue is rate-limited by ALTQ or TBR, 965 * if_start is needed to drain packets from the queue in order 966 * to notify readers when outgoing packets become ready. 967 * 968 * Should be called at splnet. 969 */ 970 static void 971 tunstart(struct ifnet *ifp) 972 { 973 struct tun_softc *tp = ifp->if_softc; 974 975 if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd)) 976 return; 977 978 mutex_enter(&tp->tun_lock); 979 if (!IF_IS_EMPTY(&ifp->if_snd)) { 980 if (tp->tun_flags & TUN_RWAIT) { 981 tp->tun_flags &= ~TUN_RWAIT; 982 wakeup((void *)tp); 983 } 984 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 985 softint_schedule(tp->tun_osih); 986 987 selnotify(&tp->tun_rsel, 0, 0); 988 } 989 mutex_exit(&tp->tun_lock); 990 } 991 #endif /* ALTQ */ 992 /* 993 * tunpoll - the poll interface, this is only useful on reads 994 * really. The write detect always returns true, write never blocks 995 * anyway, it either accepts the packet or drops it. 996 */ 997 int 998 tunpoll(dev_t dev, int events, struct lwp *l) 999 { 1000 struct tun_softc *tp; 1001 struct ifnet *ifp; 1002 int s, revents = 0; 1003 1004 s = splnet(); 1005 tp = tun_find_unit(dev); 1006 1007 /* interface was "destroyed" already */ 1008 if (tp == NULL) 1009 goto out_nolock; 1010 1011 ifp = &tp->tun_if; 1012 1013 TUNDEBUG("%s: tunpoll\n", ifp->if_xname); 1014 1015 if (events & (POLLIN | POLLRDNORM)) { 1016 if (!IFQ_IS_EMPTY(&ifp->if_snd)) { 1017 TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname, 1018 ifp->if_snd.ifq_len); 1019 revents |= events & (POLLIN | POLLRDNORM); 1020 } else { 1021 TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname); 1022 selrecord(l, &tp->tun_rsel); 1023 } 1024 } 1025 1026 if (events & (POLLOUT | POLLWRNORM)) 1027 revents |= events & (POLLOUT | POLLWRNORM); 1028 1029 mutex_exit(&tp->tun_lock); 1030 out_nolock: 1031 splx(s); 1032 return (revents); 1033 } 1034 1035 static void 1036 filt_tunrdetach(struct knote *kn) 1037 { 1038 struct tun_softc *tp = kn->kn_hook; 1039 int s; 1040 1041 s = splnet(); 1042 SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext); 1043 splx(s); 1044 } 1045 1046 static int 1047 filt_tunread(struct knote *kn, long hint) 1048 { 1049 struct tun_softc *tp = kn->kn_hook; 1050 struct ifnet *ifp = &tp->tun_if; 1051 struct mbuf *m; 1052 int s; 1053 1054 s = splnet(); 1055 IF_POLL(&ifp->if_snd, m); 1056 if (m == NULL) { 1057 splx(s); 1058 return (0); 1059 } 1060 1061 for (kn->kn_data = 0; m != NULL; m = m->m_next) 1062 kn->kn_data += m->m_len; 1063 1064 splx(s); 1065 return (1); 1066 } 1067 1068 static const struct filterops tunread_filtops = 1069 { 1, NULL, filt_tunrdetach, filt_tunread }; 1070 1071 static const struct filterops tun_seltrue_filtops = 1072 { 1, NULL, filt_tunrdetach, filt_seltrue }; 1073 1074 int 1075 tunkqfilter(dev_t dev, struct knote *kn) 1076 { 1077 struct tun_softc *tp; 1078 struct klist *klist; 1079 int rv = 0, s; 1080 1081 s = splnet(); 1082 tp = tun_find_unit(dev); 1083 if (tp == NULL) 1084 goto out_nolock; 1085 1086 switch (kn->kn_filter) { 1087 case EVFILT_READ: 1088 klist = &tp->tun_rsel.sel_klist; 1089 kn->kn_fop = &tunread_filtops; 1090 break; 1091 1092 case EVFILT_WRITE: 1093 klist = &tp->tun_rsel.sel_klist; 1094 kn->kn_fop = &tun_seltrue_filtops; 1095 break; 1096 1097 default: 1098 rv = EINVAL; 1099 goto out; 1100 } 1101 1102 kn->kn_hook = tp; 1103 1104 SLIST_INSERT_HEAD(klist, kn, kn_selnext); 1105 1106 out: 1107 mutex_exit(&tp->tun_lock); 1108 out_nolock: 1109 splx(s); 1110 return (rv); 1111 } 1112