1 /* $NetBSD: if_tun.c,v 1.106 2008/04/24 15:35:30 ad 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 1.106 2008/04/24 15:35:30 ad 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/simplelock.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 "bpfilter.h" 57 #if NBPFILTER > 0 58 #include <sys/time.h> 59 #include <net/bpf.h> 60 #endif 61 62 #include <net/if_tun.h> 63 64 #define TUNDEBUG if (tundebug) printf 65 int tundebug = 0; 66 67 extern int ifqmaxlen; 68 void tunattach(int); 69 70 static LIST_HEAD(, tun_softc) tun_softc_list; 71 static LIST_HEAD(, tun_softc) tunz_softc_list; 72 static struct simplelock tun_softc_lock; 73 74 static int tun_ioctl(struct ifnet *, u_long, void *); 75 static int tun_output(struct ifnet *, struct mbuf *, 76 const struct sockaddr *, struct rtentry *rt); 77 static int tun_clone_create(struct if_clone *, int); 78 static int tun_clone_destroy(struct ifnet *); 79 80 static struct if_clone tun_cloner = 81 IF_CLONE_INITIALIZER("tun", tun_clone_create, tun_clone_destroy); 82 83 static void tunattach0(struct tun_softc *); 84 static void tuninit(struct tun_softc *); 85 static void tun_i_softintr(void *); 86 static void tun_o_softintr(void *); 87 #ifdef ALTQ 88 static void tunstart(struct ifnet *); 89 #endif 90 static struct tun_softc *tun_find_unit(dev_t); 91 static struct tun_softc *tun_find_zunit(int); 92 93 static dev_type_open(tunopen); 94 static dev_type_close(tunclose); 95 static dev_type_read(tunread); 96 static dev_type_write(tunwrite); 97 static dev_type_ioctl(tunioctl); 98 static dev_type_poll(tunpoll); 99 static dev_type_kqfilter(tunkqfilter); 100 101 const struct cdevsw tun_cdevsw = { 102 tunopen, tunclose, tunread, tunwrite, tunioctl, 103 nostop, notty, tunpoll, nommap, tunkqfilter, D_OTHER, 104 }; 105 106 void 107 tunattach(int unused) 108 { 109 110 simple_lock_init(&tun_softc_lock); 111 LIST_INIT(&tun_softc_list); 112 LIST_INIT(&tunz_softc_list); 113 if_clone_attach(&tun_cloner); 114 } 115 116 /* 117 * Find driver instance from dev_t. 118 * Call at splnet(). 119 * Returns with tp locked (if found). 120 */ 121 static struct tun_softc * 122 tun_find_unit(dev_t dev) 123 { 124 struct tun_softc *tp; 125 int unit = minor(dev); 126 127 simple_lock(&tun_softc_lock); 128 LIST_FOREACH(tp, &tun_softc_list, tun_list) 129 if (unit == tp->tun_unit) 130 break; 131 if (tp) 132 simple_lock(&tp->tun_lock); 133 simple_unlock(&tun_softc_lock); 134 135 return (tp); 136 } 137 138 /* 139 * Find zombie driver instance by unit number. 140 * Call at splnet(). 141 * Remove tp from list and return it unlocked (if found). 142 */ 143 static struct tun_softc * 144 tun_find_zunit(int unit) 145 { 146 struct tun_softc *tp; 147 148 simple_lock(&tun_softc_lock); 149 LIST_FOREACH(tp, &tunz_softc_list, tun_list) 150 if (unit == tp->tun_unit) 151 break; 152 if (tp) 153 LIST_REMOVE(tp, tun_list); 154 simple_unlock(&tun_softc_lock); 155 #ifdef DIAGNOSTIC 156 if (tp != NULL && (tp->tun_flags & (TUN_INITED|TUN_OPEN)) != TUN_OPEN) 157 printf("tun%d: inconsistent flags: %x\n", unit, tp->tun_flags); 158 #endif 159 160 return (tp); 161 } 162 163 static int 164 tun_clone_create(struct if_clone *ifc, int unit) 165 { 166 struct tun_softc *tp; 167 168 if ((tp = tun_find_zunit(unit)) == NULL) { 169 /* Allocate a new instance */ 170 tp = malloc(sizeof(struct tun_softc), M_DEVBUF, M_WAITOK); 171 (void)memset(tp, 0, sizeof(struct tun_softc)); 172 173 tp->tun_unit = unit; 174 simple_lock_init(&tp->tun_lock); 175 selinit(&tp->tun_rsel); 176 selinit(&tp->tun_wsel); 177 } else { 178 /* Revive tunnel instance; clear ifp part */ 179 (void)memset(&tp->tun_if, 0, sizeof(struct ifnet)); 180 } 181 182 (void)snprintf(tp->tun_if.if_xname, sizeof(tp->tun_if.if_xname), 183 "%s%d", ifc->ifc_name, unit); 184 tunattach0(tp); 185 tp->tun_flags |= TUN_INITED; 186 tp->tun_osih = softint_establish(SOFTINT_CLOCK, tun_o_softintr, tp); 187 tp->tun_isih = softint_establish(SOFTINT_CLOCK, tun_i_softintr, tp); 188 189 simple_lock(&tun_softc_lock); 190 LIST_INSERT_HEAD(&tun_softc_list, tp, tun_list); 191 simple_unlock(&tun_softc_lock); 192 193 return (0); 194 } 195 196 static void 197 tunattach0(struct tun_softc *tp) 198 { 199 struct ifnet *ifp; 200 201 ifp = &tp->tun_if; 202 ifp->if_softc = tp; 203 ifp->if_mtu = TUNMTU; 204 ifp->if_ioctl = tun_ioctl; 205 ifp->if_output = tun_output; 206 #ifdef ALTQ 207 ifp->if_start = tunstart; 208 #endif 209 ifp->if_flags = IFF_POINTOPOINT; 210 ifp->if_type = IFT_TUNNEL; 211 ifp->if_snd.ifq_maxlen = ifqmaxlen; 212 ifp->if_collisions = 0; 213 ifp->if_ierrors = 0; 214 ifp->if_oerrors = 0; 215 ifp->if_ipackets = 0; 216 ifp->if_opackets = 0; 217 ifp->if_ibytes = 0; 218 ifp->if_obytes = 0; 219 ifp->if_dlt = DLT_NULL; 220 IFQ_SET_READY(&ifp->if_snd); 221 if_attach(ifp); 222 if_alloc_sadl(ifp); 223 #if NBPFILTER > 0 224 bpfattach(ifp, DLT_NULL, sizeof(uint32_t)); 225 #endif 226 } 227 228 static int 229 tun_clone_destroy(struct ifnet *ifp) 230 { 231 struct tun_softc *tp = (void *)ifp; 232 int s, zombie = 0; 233 234 s = splnet(); 235 simple_lock(&tun_softc_lock); 236 simple_lock(&tp->tun_lock); 237 LIST_REMOVE(tp, tun_list); 238 if (tp->tun_flags & TUN_OPEN) { 239 /* Hang on to storage until last close */ 240 zombie = 1; 241 tp->tun_flags &= ~TUN_INITED; 242 LIST_INSERT_HEAD(&tunz_softc_list, tp, tun_list); 243 } 244 simple_unlock(&tun_softc_lock); 245 246 IF_PURGE(&ifp->if_snd); 247 ifp->if_flags &= ~IFF_RUNNING; 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 simple_unlock(&tp->tun_lock); 256 splx(s); 257 258 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 259 fownsignal(tp->tun_pgid, SIGIO, POLL_HUP, 0, NULL); 260 261 #if NBPFILTER > 0 262 bpfdetach(ifp); 263 #endif 264 if_detach(ifp); 265 266 if (!zombie) { 267 seldestroy(&tp->tun_rsel); 268 seldestroy(&tp->tun_wsel); 269 softint_disestablish(tp->tun_osih); 270 softint_disestablish(tp->tun_isih); 271 free(tp, M_DEVBUF); 272 } 273 274 return (0); 275 } 276 277 /* 278 * tunnel open - must be superuser & the device must be 279 * configured in 280 */ 281 static int 282 tunopen(dev_t dev, int flag, int mode, struct lwp *l) 283 { 284 struct ifnet *ifp; 285 struct tun_softc *tp; 286 int s, error; 287 288 if ((error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 289 NULL)) != 0) 290 return (error); 291 292 s = splnet(); 293 tp = tun_find_unit(dev); 294 295 if (tp == NULL) { 296 (void)tun_clone_create(&tun_cloner, minor(dev)); 297 tp = tun_find_unit(dev); 298 if (tp == NULL) { 299 error = ENXIO; 300 goto out_nolock; 301 } 302 } 303 304 if (tp->tun_flags & TUN_OPEN) { 305 error = EBUSY; 306 goto out; 307 } 308 309 ifp = &tp->tun_if; 310 tp->tun_flags |= TUN_OPEN; 311 TUNDEBUG("%s: open\n", ifp->if_xname); 312 out: 313 simple_unlock(&tp->tun_lock); 314 out_nolock: 315 splx(s); 316 return (error); 317 } 318 319 /* 320 * tunclose - close the device - mark i/f down & delete 321 * routing info 322 */ 323 int 324 tunclose(dev_t dev, int flag, int mode, 325 struct lwp *l) 326 { 327 int s; 328 struct tun_softc *tp; 329 struct ifnet *ifp; 330 331 s = splnet(); 332 if ((tp = tun_find_zunit(minor(dev))) != NULL) { 333 /* interface was "destroyed" before the close */ 334 seldestroy(&tp->tun_rsel); 335 seldestroy(&tp->tun_wsel); 336 softint_disestablish(tp->tun_osih); 337 softint_disestablish(tp->tun_isih); 338 free(tp, M_DEVBUF); 339 goto out_nolock; 340 } 341 342 if ((tp = tun_find_unit(dev)) == NULL) 343 goto out_nolock; 344 345 ifp = &tp->tun_if; 346 347 tp->tun_flags &= ~TUN_OPEN; 348 349 /* 350 * junk all pending output 351 */ 352 IFQ_PURGE(&ifp->if_snd); 353 354 if (ifp->if_flags & IFF_UP) { 355 if_down(ifp); 356 if (ifp->if_flags & IFF_RUNNING) { 357 /* find internet addresses and delete routes */ 358 struct ifaddr *ifa; 359 IFADDR_FOREACH(ifa, ifp) { 360 #if defined(INET) || defined(INET6) 361 if (ifa->ifa_addr->sa_family == AF_INET || 362 ifa->ifa_addr->sa_family == AF_INET6) { 363 rtinit(ifa, (int)RTM_DELETE, 364 tp->tun_flags & TUN_DSTADDR 365 ? RTF_HOST 366 : 0); 367 } 368 #endif 369 } 370 } 371 } 372 tp->tun_pgid = 0; 373 selnotify(&tp->tun_rsel, 0, 0); 374 375 TUNDEBUG ("%s: closed\n", ifp->if_xname); 376 simple_unlock(&tp->tun_lock); 377 out_nolock: 378 splx(s); 379 return (0); 380 } 381 382 /* 383 * Call at splnet() with tp locked. 384 */ 385 static void 386 tuninit(struct tun_softc *tp) 387 { 388 struct ifnet *ifp = &tp->tun_if; 389 struct ifaddr *ifa; 390 391 TUNDEBUG("%s: tuninit\n", ifp->if_xname); 392 393 ifp->if_flags |= IFF_UP | IFF_RUNNING; 394 395 tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR); 396 IFADDR_FOREACH(ifa, ifp) { 397 #ifdef INET 398 if (ifa->ifa_addr->sa_family == AF_INET) { 399 struct sockaddr_in *sin; 400 401 sin = satosin(ifa->ifa_addr); 402 if (sin && sin->sin_addr.s_addr) 403 tp->tun_flags |= TUN_IASET; 404 405 if (ifp->if_flags & IFF_POINTOPOINT) { 406 sin = satosin(ifa->ifa_dstaddr); 407 if (sin && sin->sin_addr.s_addr) 408 tp->tun_flags |= TUN_DSTADDR; 409 } 410 } 411 #endif 412 #ifdef INET6 413 if (ifa->ifa_addr->sa_family == AF_INET6) { 414 struct sockaddr_in6 *sin; 415 416 sin = (struct sockaddr_in6 *)ifa->ifa_addr; 417 if (!IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr)) 418 tp->tun_flags |= TUN_IASET; 419 420 if (ifp->if_flags & IFF_POINTOPOINT) { 421 sin = (struct sockaddr_in6 *)ifa->ifa_dstaddr; 422 if (sin && 423 !IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr)) 424 tp->tun_flags |= TUN_DSTADDR; 425 } else 426 tp->tun_flags &= ~TUN_DSTADDR; 427 } 428 #endif /* INET6 */ 429 } 430 431 return; 432 } 433 434 /* 435 * Process an ioctl request. 436 */ 437 static int 438 tun_ioctl(struct ifnet *ifp, u_long cmd, void *data) 439 { 440 int error = 0, s; 441 struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc); 442 struct ifreq *ifr = data; 443 444 s = splnet(); 445 simple_lock(&tp->tun_lock); 446 447 switch (cmd) { 448 case SIOCSIFADDR: 449 tuninit(tp); 450 TUNDEBUG("%s: address set\n", ifp->if_xname); 451 break; 452 case SIOCSIFDSTADDR: 453 tuninit(tp); 454 TUNDEBUG("%s: destination address set\n", ifp->if_xname); 455 break; 456 case SIOCSIFBRDADDR: 457 TUNDEBUG("%s: broadcast address set\n", ifp->if_xname); 458 break; 459 case SIOCSIFMTU: 460 if (ifr->ifr_mtu > TUNMTU || ifr->ifr_mtu < 576) { 461 error = EINVAL; 462 break; 463 } 464 TUNDEBUG("%s: interface mtu set\n", ifp->if_xname); 465 if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET) 466 error = 0; 467 break; 468 case SIOCADDMULTI: 469 case SIOCDELMULTI: 470 if (ifr == NULL) { 471 error = EAFNOSUPPORT; /* XXX */ 472 break; 473 } 474 switch (ifreq_getaddr(cmd, ifr)->sa_family) { 475 #ifdef INET 476 case AF_INET: 477 break; 478 #endif 479 #ifdef INET6 480 case AF_INET6: 481 break; 482 #endif 483 default: 484 error = EAFNOSUPPORT; 485 break; 486 } 487 break; 488 case SIOCSIFFLAGS: 489 break; 490 default: 491 error = EINVAL; 492 } 493 494 simple_unlock(&tp->tun_lock); 495 splx(s); 496 return (error); 497 } 498 499 /* 500 * tun_output - queue packets from higher level ready to put out. 501 */ 502 static int 503 tun_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst, 504 struct rtentry *rt) 505 { 506 struct tun_softc *tp = ifp->if_softc; 507 int s; 508 int error; 509 #if defined(INET) || defined(INET6) 510 int mlen; 511 uint32_t *af; 512 #endif 513 ALTQ_DECL(struct altq_pktattr pktattr;) 514 515 s = splnet(); 516 simple_lock(&tp->tun_lock); 517 TUNDEBUG ("%s: tun_output\n", ifp->if_xname); 518 519 if ((tp->tun_flags & TUN_READY) != TUN_READY) { 520 TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, 521 tp->tun_flags); 522 m_freem (m0); 523 error = EHOSTDOWN; 524 goto out; 525 } 526 527 /* 528 * if the queueing discipline needs packet classification, 529 * do it before prepending link headers. 530 */ 531 IFQ_CLASSIFY(&ifp->if_snd, m0, dst->sa_family, &pktattr); 532 533 #if NBPFILTER > 0 534 if (ifp->if_bpf) 535 bpf_mtap_af(ifp->if_bpf, dst->sa_family, m0); 536 #endif 537 538 switch(dst->sa_family) { 539 #ifdef INET6 540 case AF_INET6: 541 #endif 542 #ifdef INET 543 case AF_INET: 544 #endif 545 #if defined(INET) || defined(INET6) 546 if (tp->tun_flags & TUN_PREPADDR) { 547 /* Simple link-layer header */ 548 M_PREPEND(m0, dst->sa_len, M_DONTWAIT); 549 if (m0 == NULL) { 550 IF_DROP(&ifp->if_snd); 551 error = ENOBUFS; 552 goto out; 553 } 554 bcopy(dst, mtod(m0, char *), dst->sa_len); 555 } 556 557 if (tp->tun_flags & TUN_IFHEAD) { 558 /* Prepend the address family */ 559 M_PREPEND(m0, sizeof(*af), M_DONTWAIT); 560 if (m0 == NULL) { 561 IF_DROP(&ifp->if_snd); 562 error = ENOBUFS; 563 goto out; 564 } 565 af = mtod(m0,uint32_t *); 566 *af = htonl(dst->sa_family); 567 } else { 568 #ifdef INET 569 if (dst->sa_family != AF_INET) 570 #endif 571 { 572 m_freem(m0); 573 error = EAFNOSUPPORT; 574 goto out; 575 } 576 } 577 /* FALLTHROUGH */ 578 case AF_UNSPEC: 579 IFQ_ENQUEUE(&ifp->if_snd, m0, &pktattr, error); 580 if (error) { 581 ifp->if_collisions++; 582 error = EAFNOSUPPORT; 583 goto out; 584 } 585 mlen = m0->m_pkthdr.len; 586 ifp->if_opackets++; 587 ifp->if_obytes += mlen; 588 break; 589 #endif 590 default: 591 m_freem(m0); 592 error = EAFNOSUPPORT; 593 goto out; 594 } 595 596 if (tp->tun_flags & TUN_RWAIT) { 597 tp->tun_flags &= ~TUN_RWAIT; 598 wakeup((void *)tp); 599 } 600 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 601 softint_schedule(tp->tun_isih); 602 603 selnotify(&tp->tun_rsel, 0, 0); 604 out: 605 simple_unlock(&tp->tun_lock); 606 splx(s); 607 return (0); 608 } 609 610 static void 611 tun_i_softintr(void *cookie) 612 { 613 struct tun_softc *tp = cookie; 614 615 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 616 fownsignal(tp->tun_pgid, SIGIO, POLL_IN, POLLIN|POLLRDNORM, 617 NULL); 618 } 619 620 static void 621 tun_o_softintr(void *cookie) 622 { 623 struct tun_softc *tp = cookie; 624 625 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 626 fownsignal(tp->tun_pgid, SIGIO, POLL_OUT, POLLOUT|POLLWRNORM, 627 NULL); 628 } 629 630 /* 631 * the cdevsw interface is now pretty minimal. 632 */ 633 int 634 tunioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 635 { 636 struct tun_softc *tp; 637 int s, error = 0; 638 639 s = splnet(); 640 tp = tun_find_unit(dev); 641 642 /* interface was "destroyed" already */ 643 if (tp == NULL) { 644 error = ENXIO; 645 goto out_nolock; 646 } 647 648 switch (cmd) { 649 case TUNSDEBUG: 650 tundebug = *(int *)data; 651 break; 652 653 case TUNGDEBUG: 654 *(int *)data = tundebug; 655 break; 656 657 case TUNSIFMODE: 658 switch (*(int *)data & (IFF_POINTOPOINT|IFF_BROADCAST)) { 659 case IFF_POINTOPOINT: 660 case IFF_BROADCAST: 661 if (tp->tun_if.if_flags & IFF_UP) { 662 error = EBUSY; 663 goto out; 664 } 665 tp->tun_if.if_flags &= 666 ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST); 667 tp->tun_if.if_flags |= *(int *)data; 668 break; 669 default: 670 error = EINVAL; 671 goto out; 672 } 673 break; 674 675 case TUNSLMODE: 676 if (*(int *)data) { 677 tp->tun_flags |= TUN_PREPADDR; 678 tp->tun_flags &= ~TUN_IFHEAD; 679 } else 680 tp->tun_flags &= ~TUN_PREPADDR; 681 break; 682 683 case TUNSIFHEAD: 684 if (*(int *)data) { 685 tp->tun_flags |= TUN_IFHEAD; 686 tp->tun_flags &= ~TUN_PREPADDR; 687 } else 688 tp->tun_flags &= ~TUN_IFHEAD; 689 break; 690 691 case TUNGIFHEAD: 692 *(int *)data = (tp->tun_flags & TUN_IFHEAD); 693 break; 694 695 case FIONBIO: 696 if (*(int *)data) 697 tp->tun_flags |= TUN_NBIO; 698 else 699 tp->tun_flags &= ~TUN_NBIO; 700 break; 701 702 case FIOASYNC: 703 if (*(int *)data) 704 tp->tun_flags |= TUN_ASYNC; 705 else 706 tp->tun_flags &= ~TUN_ASYNC; 707 break; 708 709 case FIONREAD: 710 if (tp->tun_if.if_snd.ifq_head) 711 *(int *)data = tp->tun_if.if_snd.ifq_head->m_pkthdr.len; 712 else 713 *(int *)data = 0; 714 break; 715 716 case TIOCSPGRP: 717 case FIOSETOWN: 718 error = fsetown(&tp->tun_pgid, cmd, data); 719 break; 720 721 case TIOCGPGRP: 722 case FIOGETOWN: 723 error = fgetown(tp->tun_pgid, cmd, data); 724 break; 725 726 default: 727 error = ENOTTY; 728 } 729 730 out: 731 simple_unlock(&tp->tun_lock); 732 out_nolock: 733 splx(s); 734 return (error); 735 } 736 737 /* 738 * The cdevsw read interface - reads a packet at a time, or at 739 * least as much of a packet as can be read. 740 */ 741 int 742 tunread(dev_t dev, struct uio *uio, int ioflag) 743 { 744 struct tun_softc *tp; 745 struct ifnet *ifp; 746 struct mbuf *m, *m0; 747 int error = 0, len, s, index; 748 749 s = splnet(); 750 tp = tun_find_unit(dev); 751 752 /* interface was "destroyed" already */ 753 if (tp == NULL) { 754 error = ENXIO; 755 goto out_nolock; 756 } 757 758 index = tp->tun_if.if_index; 759 ifp = &tp->tun_if; 760 761 TUNDEBUG ("%s: read\n", ifp->if_xname); 762 if ((tp->tun_flags & TUN_READY) != TUN_READY) { 763 TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags); 764 error = EHOSTDOWN; 765 goto out; 766 } 767 768 tp->tun_flags &= ~TUN_RWAIT; 769 770 do { 771 IFQ_DEQUEUE(&ifp->if_snd, m0); 772 if (m0 == 0) { 773 if (tp->tun_flags & TUN_NBIO) { 774 error = EWOULDBLOCK; 775 goto out; 776 } 777 tp->tun_flags |= TUN_RWAIT; 778 if (ltsleep((void *)tp, PZERO|PCATCH|PNORELOCK, 779 "tunread", 0, &tp->tun_lock) != 0) { 780 error = EINTR; 781 goto out_nolock; 782 } else { 783 /* 784 * Maybe the interface was destroyed while 785 * we were sleeping, so let's ensure that 786 * we're looking at the same (valid) tun 787 * interface before looping. 788 */ 789 tp = tun_find_unit(dev); 790 if (tp == NULL) { 791 error = ENXIO; 792 goto out_nolock; 793 } 794 if (tp->tun_if.if_index != index) { 795 error = ENXIO; 796 goto out; 797 } 798 } 799 } 800 } while (m0 == 0); 801 802 simple_unlock(&tp->tun_lock); 803 splx(s); 804 805 /* Copy the mbuf chain */ 806 while (m0 && uio->uio_resid > 0 && error == 0) { 807 len = min(uio->uio_resid, m0->m_len); 808 if (len != 0) 809 error = uiomove(mtod(m0, void *), len, uio); 810 MFREE(m0, m); 811 m0 = m; 812 } 813 814 if (m0) { 815 TUNDEBUG("Dropping mbuf\n"); 816 m_freem(m0); 817 } 818 if (error) 819 ifp->if_ierrors++; 820 821 return (error); 822 823 out: 824 simple_unlock(&tp->tun_lock); 825 out_nolock: 826 splx(s); 827 return (error); 828 } 829 830 /* 831 * the cdevsw write interface - an atomic write is a packet - or else! 832 */ 833 int 834 tunwrite(dev_t dev, struct uio *uio, int ioflag) 835 { 836 struct tun_softc *tp; 837 struct ifnet *ifp; 838 struct mbuf *top, **mp, *m; 839 struct ifqueue *ifq; 840 struct sockaddr dst; 841 int isr, error = 0, s, tlen, mlen; 842 uint32_t family; 843 844 s = splnet(); 845 tp = tun_find_unit(dev); 846 847 /* interface was "destroyed" already */ 848 if (tp == NULL) { 849 error = ENXIO; 850 goto out_nolock; 851 } 852 853 /* Unlock until we've got the data */ 854 simple_unlock(&tp->tun_lock); 855 splx(s); 856 857 ifp = &tp->tun_if; 858 859 TUNDEBUG("%s: tunwrite\n", ifp->if_xname); 860 861 if (tp->tun_flags & TUN_PREPADDR) { 862 if (uio->uio_resid < sizeof(dst)) { 863 error = EIO; 864 goto out0; 865 } 866 error = uiomove((void *)&dst, sizeof(dst), uio); 867 if (dst.sa_len > sizeof(dst)) { 868 /* Duh.. */ 869 char discard; 870 int n = dst.sa_len - sizeof(dst); 871 while (n--) 872 if ((error = uiomove(&discard, 1, uio)) != 0) { 873 goto out0; 874 } 875 } 876 } else if (tp->tun_flags & TUN_IFHEAD) { 877 if (uio->uio_resid < sizeof(family)){ 878 error = EIO; 879 goto out0; 880 } 881 error = uiomove((void *)&family, sizeof(family), uio); 882 dst.sa_family = ntohl(family); 883 } else { 884 #ifdef INET 885 dst.sa_family = AF_INET; 886 #endif 887 } 888 889 if (uio->uio_resid > TUNMTU) { 890 TUNDEBUG("%s: len=%lu!\n", ifp->if_xname, 891 (unsigned long)uio->uio_resid); 892 error = EIO; 893 goto out0; 894 } 895 896 switch (dst.sa_family) { 897 #ifdef INET 898 case AF_INET: 899 ifq = &ipintrq; 900 isr = NETISR_IP; 901 break; 902 #endif 903 #ifdef INET6 904 case AF_INET6: 905 ifq = &ip6intrq; 906 isr = NETISR_IPV6; 907 break; 908 #endif 909 default: 910 error = EAFNOSUPPORT; 911 goto out0; 912 } 913 914 tlen = uio->uio_resid; 915 916 /* get a header mbuf */ 917 MGETHDR(m, M_DONTWAIT, MT_DATA); 918 if (m == NULL) { 919 error = ENOBUFS; 920 goto out0; 921 } 922 mlen = MHLEN; 923 924 top = NULL; 925 mp = ⊤ 926 while (error == 0 && uio->uio_resid > 0) { 927 m->m_len = min(mlen, uio->uio_resid); 928 error = uiomove(mtod(m, void *), m->m_len, uio); 929 *mp = m; 930 mp = &m->m_next; 931 if (error == 0 && uio->uio_resid > 0) { 932 MGET(m, M_DONTWAIT, MT_DATA); 933 if (m == NULL) { 934 error = ENOBUFS; 935 break; 936 } 937 mlen = MLEN; 938 } 939 } 940 if (error) { 941 if (top != NULL) 942 m_freem (top); 943 ifp->if_ierrors++; 944 goto out0; 945 } 946 947 top->m_pkthdr.len = tlen; 948 top->m_pkthdr.rcvif = ifp; 949 950 #if NBPFILTER > 0 951 if (ifp->if_bpf) 952 bpf_mtap_af(ifp->if_bpf, dst.sa_family, top); 953 #endif 954 955 s = splnet(); 956 simple_lock(&tp->tun_lock); 957 if ((tp->tun_flags & TUN_INITED) == 0) { 958 /* Interface was destroyed */ 959 error = ENXIO; 960 goto out; 961 } 962 if (IF_QFULL(ifq)) { 963 IF_DROP(ifq); 964 ifp->if_collisions++; 965 m_freem(top); 966 error = ENOBUFS; 967 goto out; 968 } 969 970 IF_ENQUEUE(ifq, top); 971 ifp->if_ipackets++; 972 ifp->if_ibytes += tlen; 973 schednetisr(isr); 974 out: 975 simple_unlock(&tp->tun_lock); 976 out_nolock: 977 splx(s); 978 out0: 979 return (error); 980 } 981 982 #ifdef ALTQ 983 /* 984 * Start packet transmission on the interface. 985 * when the interface queue is rate-limited by ALTQ or TBR, 986 * if_start is needed to drain packets from the queue in order 987 * to notify readers when outgoing packets become ready. 988 * 989 * Should be called at splnet. 990 */ 991 static void 992 tunstart(struct ifnet *ifp) 993 { 994 struct tun_softc *tp = ifp->if_softc; 995 996 if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd)) 997 return; 998 999 simple_lock(&tp->tun_lock); 1000 if (!IF_IS_EMPTY(&ifp->if_snd)) { 1001 if (tp->tun_flags & TUN_RWAIT) { 1002 tp->tun_flags &= ~TUN_RWAIT; 1003 wakeup((void *)tp); 1004 } 1005 if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid) 1006 softint_schedule(tp->tun_osih); 1007 1008 selnotify(&tp->tun_rsel, 0, 0); 1009 } 1010 simple_unlock(&tp->tun_lock); 1011 } 1012 #endif /* ALTQ */ 1013 /* 1014 * tunpoll - the poll interface, this is only useful on reads 1015 * really. The write detect always returns true, write never blocks 1016 * anyway, it either accepts the packet or drops it. 1017 */ 1018 int 1019 tunpoll(dev_t dev, int events, struct lwp *l) 1020 { 1021 struct tun_softc *tp; 1022 struct ifnet *ifp; 1023 int s, revents = 0; 1024 1025 s = splnet(); 1026 tp = tun_find_unit(dev); 1027 1028 /* interface was "destroyed" already */ 1029 if (tp == NULL) 1030 goto out_nolock; 1031 1032 ifp = &tp->tun_if; 1033 1034 TUNDEBUG("%s: tunpoll\n", ifp->if_xname); 1035 1036 if (events & (POLLIN | POLLRDNORM)) { 1037 if (!IFQ_IS_EMPTY(&ifp->if_snd)) { 1038 TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname, 1039 ifp->if_snd.ifq_len); 1040 revents |= events & (POLLIN | POLLRDNORM); 1041 } else { 1042 TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname); 1043 selrecord(l, &tp->tun_rsel); 1044 } 1045 } 1046 1047 if (events & (POLLOUT | POLLWRNORM)) 1048 revents |= events & (POLLOUT | POLLWRNORM); 1049 1050 simple_unlock(&tp->tun_lock); 1051 out_nolock: 1052 splx(s); 1053 return (revents); 1054 } 1055 1056 static void 1057 filt_tunrdetach(struct knote *kn) 1058 { 1059 struct tun_softc *tp = kn->kn_hook; 1060 int s; 1061 1062 s = splnet(); 1063 SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext); 1064 splx(s); 1065 } 1066 1067 static int 1068 filt_tunread(struct knote *kn, long hint) 1069 { 1070 struct tun_softc *tp = kn->kn_hook; 1071 struct ifnet *ifp = &tp->tun_if; 1072 struct mbuf *m; 1073 int s; 1074 1075 s = splnet(); 1076 IF_POLL(&ifp->if_snd, m); 1077 if (m == NULL) { 1078 splx(s); 1079 return (0); 1080 } 1081 1082 for (kn->kn_data = 0; m != NULL; m = m->m_next) 1083 kn->kn_data += m->m_len; 1084 1085 splx(s); 1086 return (1); 1087 } 1088 1089 static const struct filterops tunread_filtops = 1090 { 1, NULL, filt_tunrdetach, filt_tunread }; 1091 1092 static const struct filterops tun_seltrue_filtops = 1093 { 1, NULL, filt_tunrdetach, filt_seltrue }; 1094 1095 int 1096 tunkqfilter(dev_t dev, struct knote *kn) 1097 { 1098 struct tun_softc *tp; 1099 struct klist *klist; 1100 int rv = 0, s; 1101 1102 s = splnet(); 1103 tp = tun_find_unit(dev); 1104 if (tp == NULL) 1105 goto out_nolock; 1106 1107 switch (kn->kn_filter) { 1108 case EVFILT_READ: 1109 klist = &tp->tun_rsel.sel_klist; 1110 kn->kn_fop = &tunread_filtops; 1111 break; 1112 1113 case EVFILT_WRITE: 1114 klist = &tp->tun_rsel.sel_klist; 1115 kn->kn_fop = &tun_seltrue_filtops; 1116 break; 1117 1118 default: 1119 rv = EINVAL; 1120 goto out; 1121 } 1122 1123 kn->kn_hook = tp; 1124 1125 SLIST_INSERT_HEAD(klist, kn, kn_selnext); 1126 1127 out: 1128 simple_unlock(&tp->tun_lock); 1129 out_nolock: 1130 splx(s); 1131 return (rv); 1132 } 1133