raw_ip.c revision 1.64
1 /* $NetBSD: raw_ip.c,v 1.64 2002/10/22 02:44:34 simonb Exp $ */ 2 3 /* 4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the project nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT 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 PROJECT 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 32 /* 33 * Copyright (c) 1982, 1986, 1988, 1993 34 * The Regents of the University of California. All rights reserved. 35 * 36 * Redistribution and use in source and binary forms, with or without 37 * modification, are permitted provided that the following conditions 38 * are met: 39 * 1. Redistributions of source code must retain the above copyright 40 * notice, this list of conditions and the following disclaimer. 41 * 2. Redistributions in binary form must reproduce the above copyright 42 * notice, this list of conditions and the following disclaimer in the 43 * documentation and/or other materials provided with the distribution. 44 * 3. All advertising materials mentioning features or use of this software 45 * must display the following acknowledgement: 46 * This product includes software developed by the University of 47 * California, Berkeley and its contributors. 48 * 4. Neither the name of the University nor the names of its contributors 49 * may be used to endorse or promote products derived from this software 50 * without specific prior written permission. 51 * 52 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 53 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 54 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 55 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 56 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 57 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 58 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 59 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 60 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 61 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 62 * SUCH DAMAGE. 63 * 64 * @(#)raw_ip.c 8.7 (Berkeley) 5/15/95 65 */ 66 67 #include <sys/cdefs.h> 68 __KERNEL_RCSID(0, "$NetBSD: raw_ip.c,v 1.64 2002/10/22 02:44:34 simonb Exp $"); 69 70 #include "opt_ipsec.h" 71 #include "opt_mrouting.h" 72 73 #include <sys/param.h> 74 #include <sys/malloc.h> 75 #include <sys/mbuf.h> 76 #include <sys/socket.h> 77 #include <sys/protosw.h> 78 #include <sys/socketvar.h> 79 #include <sys/errno.h> 80 #include <sys/systm.h> 81 #include <sys/proc.h> 82 83 #include <net/if.h> 84 #include <net/route.h> 85 86 #include <netinet/in.h> 87 #include <netinet/in_systm.h> 88 #include <netinet/ip.h> 89 #include <netinet/ip_var.h> 90 #include <netinet/ip_mroute.h> 91 #include <netinet/ip_icmp.h> 92 #include <netinet/in_pcb.h> 93 #include <netinet/in_var.h> 94 95 #include <machine/stdarg.h> 96 97 #ifdef IPSEC 98 #include <netinet6/ipsec.h> 99 #endif /*IPSEC*/ 100 101 struct inpcbtable rawcbtable; 102 103 int rip_pcbnotify __P((struct inpcbtable *, struct in_addr, 104 struct in_addr, int, int, void (*) __P((struct inpcb *, int)))); 105 int rip_bind __P((struct inpcb *, struct mbuf *)); 106 int rip_connect __P((struct inpcb *, struct mbuf *)); 107 void rip_disconnect __P((struct inpcb *)); 108 109 /* 110 * Nominal space allocated to a raw ip socket. 111 */ 112 #define RIPSNDQ 8192 113 #define RIPRCVQ 8192 114 115 /* 116 * Raw interface to IP protocol. 117 */ 118 119 /* 120 * Initialize raw connection block q. 121 */ 122 void 123 rip_init() 124 { 125 126 in_pcbinit(&rawcbtable, 1, 1); 127 } 128 129 /* 130 * Setup generic address and protocol structures 131 * for raw_input routine, then pass them along with 132 * mbuf chain. 133 */ 134 void 135 #if __STDC__ 136 rip_input(struct mbuf *m, ...) 137 #else 138 rip_input(m, va_alist) 139 struct mbuf *m; 140 va_dcl 141 #endif 142 { 143 int proto; 144 struct ip *ip = mtod(m, struct ip *); 145 struct inpcb *inp; 146 struct inpcb *last = 0; 147 struct mbuf *opts = 0; 148 struct sockaddr_in ripsrc; 149 va_list ap; 150 151 va_start(ap, m); 152 (void)va_arg(ap, int); /* ignore value, advance ap */ 153 proto = va_arg(ap, int); 154 va_end(ap); 155 156 ripsrc.sin_family = AF_INET; 157 ripsrc.sin_len = sizeof(struct sockaddr_in); 158 ripsrc.sin_addr = ip->ip_src; 159 ripsrc.sin_port = 0; 160 bzero((caddr_t)ripsrc.sin_zero, sizeof(ripsrc.sin_zero)); 161 162 /* 163 * XXX Compatibility: programs using raw IP expect ip_len 164 * XXX to have the header length subtracted, and in host order. 165 * XXX ip_off is also expected to be host order. 166 */ 167 ip->ip_len = ntohs(ip->ip_len) - (ip->ip_hl << 2); 168 NTOHS(ip->ip_off); 169 170 CIRCLEQ_FOREACH(inp, &rawcbtable.inpt_queue, inp_queue) { 171 if (inp->inp_ip.ip_p && inp->inp_ip.ip_p != proto) 172 continue; 173 if (!in_nullhost(inp->inp_laddr) && 174 !in_hosteq(inp->inp_laddr, ip->ip_dst)) 175 continue; 176 if (!in_nullhost(inp->inp_faddr) && 177 !in_hosteq(inp->inp_faddr, ip->ip_src)) 178 continue; 179 if (last) { 180 struct mbuf *n; 181 182 #ifdef IPSEC 183 /* check AH/ESP integrity. */ 184 if (ipsec4_in_reject_so(m, last->inp_socket)) { 185 ipsecstat.in_polvio++; 186 /* do not inject data to pcb */ 187 } else 188 #endif /*IPSEC*/ 189 if ((n = m_copy(m, 0, (int)M_COPYALL)) != NULL) { 190 if (last->inp_flags & INP_CONTROLOPTS || 191 last->inp_socket->so_options & SO_TIMESTAMP) 192 ip_savecontrol(last, &opts, ip, n); 193 if (sbappendaddr(&last->inp_socket->so_rcv, 194 sintosa(&ripsrc), n, opts) == 0) { 195 /* should notify about lost packet */ 196 m_freem(n); 197 if (opts) 198 m_freem(opts); 199 } else 200 sorwakeup(last->inp_socket); 201 opts = NULL; 202 } 203 } 204 last = inp; 205 } 206 #ifdef IPSEC 207 /* check AH/ESP integrity. */ 208 if (last && ipsec4_in_reject_so(m, last->inp_socket)) { 209 m_freem(m); 210 ipsecstat.in_polvio++; 211 ipstat.ips_delivered--; 212 /* do not inject data to pcb */ 213 } else 214 #endif /*IPSEC*/ 215 if (last) { 216 if (last->inp_flags & INP_CONTROLOPTS || 217 last->inp_socket->so_options & SO_TIMESTAMP) 218 ip_savecontrol(last, &opts, ip, m); 219 if (sbappendaddr(&last->inp_socket->so_rcv, 220 sintosa(&ripsrc), m, opts) == 0) { 221 m_freem(m); 222 if (opts) 223 m_freem(opts); 224 } else 225 sorwakeup(last->inp_socket); 226 } else { 227 if (inetsw[ip_protox[ip->ip_p]].pr_input == rip_input) { 228 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PROTOCOL, 229 0, 0); 230 ipstat.ips_noproto++; 231 ipstat.ips_delivered--; 232 } else 233 m_freem(m); 234 } 235 return; 236 } 237 238 int 239 rip_pcbnotify(table, faddr, laddr, proto, errno, notify) 240 struct inpcbtable *table; 241 struct in_addr faddr, laddr; 242 int proto; 243 int errno; 244 void (*notify) __P((struct inpcb *, int)); 245 { 246 struct inpcb *inp, *ninp; 247 int nmatch; 248 249 nmatch = 0; 250 for (inp = CIRCLEQ_FIRST(&table->inpt_queue); 251 inp != (struct inpcb *)&table->inpt_queue; 252 inp = ninp) { 253 ninp = inp->inp_queue.cqe_next; 254 if (inp->inp_ip.ip_p && inp->inp_ip.ip_p != proto) 255 continue; 256 if (in_hosteq(inp->inp_faddr, faddr) && 257 in_hosteq(inp->inp_laddr, laddr)) { 258 (*notify)(inp, errno); 259 nmatch++; 260 } 261 } 262 263 return nmatch; 264 } 265 266 void * 267 rip_ctlinput(cmd, sa, v) 268 int cmd; 269 struct sockaddr *sa; 270 void *v; 271 { 272 struct ip *ip = v; 273 void (*notify) __P((struct inpcb *, int)) = in_rtchange; 274 int errno; 275 276 if (sa->sa_family != AF_INET || 277 sa->sa_len != sizeof(struct sockaddr_in)) 278 return NULL; 279 if ((unsigned)cmd >= PRC_NCMDS) 280 return NULL; 281 errno = inetctlerrmap[cmd]; 282 if (PRC_IS_REDIRECT(cmd)) 283 notify = in_rtchange, ip = 0; 284 else if (cmd == PRC_HOSTDEAD) 285 ip = 0; 286 else if (errno == 0) 287 return NULL; 288 if (ip) { 289 rip_pcbnotify(&rawcbtable, satosin(sa)->sin_addr, 290 ip->ip_src, ip->ip_p, errno, notify); 291 292 /* XXX mapped address case */ 293 } else 294 in_pcbnotifyall(&rawcbtable, satosin(sa)->sin_addr, errno, 295 notify); 296 return NULL; 297 } 298 299 /* 300 * Generate IP header and pass packet to ip_output. 301 * Tack on options user may have setup with control call. 302 */ 303 int 304 #if __STDC__ 305 rip_output(struct mbuf *m, ...) 306 #else 307 rip_output(m, va_alist) 308 struct mbuf *m; 309 va_dcl 310 #endif 311 { 312 struct inpcb *inp; 313 struct ip *ip; 314 struct mbuf *opts; 315 int flags; 316 va_list ap; 317 318 va_start(ap, m); 319 inp = va_arg(ap, struct inpcb *); 320 va_end(ap); 321 322 flags = 323 (inp->inp_socket->so_options & SO_DONTROUTE) | IP_ALLOWBROADCAST 324 | IP_RETURNMTU; 325 326 /* 327 * If the user handed us a complete IP packet, use it. 328 * Otherwise, allocate an mbuf for a header and fill it in. 329 */ 330 if ((inp->inp_flags & INP_HDRINCL) == 0) { 331 if ((m->m_pkthdr.len + sizeof(struct ip)) > IP_MAXPACKET) { 332 m_freem(m); 333 return (EMSGSIZE); 334 } 335 M_PREPEND(m, sizeof(struct ip), M_WAIT); 336 ip = mtod(m, struct ip *); 337 ip->ip_tos = 0; 338 ip->ip_off = htons(0); 339 ip->ip_p = inp->inp_ip.ip_p; 340 ip->ip_len = htons(m->m_pkthdr.len); 341 ip->ip_src = inp->inp_laddr; 342 ip->ip_dst = inp->inp_faddr; 343 ip->ip_ttl = MAXTTL; 344 opts = inp->inp_options; 345 } else { 346 if (m->m_pkthdr.len > IP_MAXPACKET) { 347 m_freem(m); 348 return (EMSGSIZE); 349 } 350 ip = mtod(m, struct ip *); 351 /* XXX userland passes ip_len and ip_off in host order */ 352 if (m->m_pkthdr.len != ip->ip_len) { 353 m_freem(m); 354 return (EINVAL); 355 } 356 HTONS(ip->ip_len); 357 HTONS(ip->ip_off); 358 if (ip->ip_id == 0) 359 ip->ip_id = htons(ip_id++); 360 opts = NULL; 361 /* XXX prevent ip_output from overwriting header fields */ 362 flags |= IP_RAWOUTPUT; 363 ipstat.ips_rawout++; 364 } 365 #ifdef IPSEC 366 if (ipsec_setsocket(m, inp->inp_socket) != 0) { 367 m_freem(m); 368 return ENOBUFS; 369 } 370 #endif /*IPSEC*/ 371 return (ip_output(m, opts, &inp->inp_route, flags, inp->inp_moptions, 372 &inp->inp_errormtu)); 373 } 374 375 /* 376 * Raw IP socket option processing. 377 */ 378 int 379 rip_ctloutput(op, so, level, optname, m) 380 int op; 381 struct socket *so; 382 int level, optname; 383 struct mbuf **m; 384 { 385 struct inpcb *inp = sotoinpcb(so); 386 int error = 0; 387 388 if (level != IPPROTO_IP) { 389 error = ENOPROTOOPT; 390 if (op == PRCO_SETOPT && *m != 0) 391 (void) m_free(*m); 392 } else switch (op) { 393 394 case PRCO_SETOPT: 395 switch (optname) { 396 case IP_HDRINCL: 397 if (*m == 0 || (*m)->m_len < sizeof (int)) 398 error = EINVAL; 399 else { 400 if (*mtod(*m, int *)) 401 inp->inp_flags |= INP_HDRINCL; 402 else 403 inp->inp_flags &= ~INP_HDRINCL; 404 } 405 if (*m != 0) 406 (void) m_free(*m); 407 break; 408 409 #ifdef MROUTING 410 case MRT_INIT: 411 case MRT_DONE: 412 case MRT_ADD_VIF: 413 case MRT_DEL_VIF: 414 case MRT_ADD_MFC: 415 case MRT_DEL_MFC: 416 case MRT_ASSERT: 417 error = ip_mrouter_set(so, optname, m); 418 break; 419 #endif 420 421 default: 422 error = ip_ctloutput(op, so, level, optname, m); 423 break; 424 } 425 break; 426 427 case PRCO_GETOPT: 428 switch (optname) { 429 case IP_HDRINCL: 430 *m = m_get(M_WAIT, M_SOOPTS); 431 (*m)->m_len = sizeof (int); 432 *mtod(*m, int *) = inp->inp_flags & INP_HDRINCL ? 1 : 0; 433 break; 434 435 #ifdef MROUTING 436 case MRT_VERSION: 437 case MRT_ASSERT: 438 error = ip_mrouter_get(so, optname, m); 439 break; 440 #endif 441 442 default: 443 error = ip_ctloutput(op, so, level, optname, m); 444 break; 445 } 446 break; 447 } 448 return (error); 449 } 450 451 int 452 rip_bind(inp, nam) 453 struct inpcb *inp; 454 struct mbuf *nam; 455 { 456 struct sockaddr_in *addr = mtod(nam, struct sockaddr_in *); 457 458 if (nam->m_len != sizeof(*addr)) 459 return (EINVAL); 460 if (TAILQ_FIRST(&ifnet) == 0) 461 return (EADDRNOTAVAIL); 462 if (addr->sin_family != AF_INET && 463 addr->sin_family != AF_IMPLINK) 464 return (EAFNOSUPPORT); 465 if (!in_nullhost(addr->sin_addr) && 466 ifa_ifwithaddr(sintosa(addr)) == 0) 467 return (EADDRNOTAVAIL); 468 inp->inp_laddr = addr->sin_addr; 469 return (0); 470 } 471 472 int 473 rip_connect(inp, nam) 474 struct inpcb *inp; 475 struct mbuf *nam; 476 { 477 struct sockaddr_in *addr = mtod(nam, struct sockaddr_in *); 478 479 if (nam->m_len != sizeof(*addr)) 480 return (EINVAL); 481 if (TAILQ_FIRST(&ifnet) == 0) 482 return (EADDRNOTAVAIL); 483 if (addr->sin_family != AF_INET && 484 addr->sin_family != AF_IMPLINK) 485 return (EAFNOSUPPORT); 486 inp->inp_faddr = addr->sin_addr; 487 return (0); 488 } 489 490 void 491 rip_disconnect(inp) 492 struct inpcb *inp; 493 { 494 495 inp->inp_faddr = zeroin_addr; 496 } 497 498 u_long rip_sendspace = RIPSNDQ; 499 u_long rip_recvspace = RIPRCVQ; 500 501 /*ARGSUSED*/ 502 int 503 rip_usrreq(so, req, m, nam, control, p) 504 struct socket *so; 505 int req; 506 struct mbuf *m, *nam, *control; 507 struct proc *p; 508 { 509 struct inpcb *inp; 510 int s; 511 int error = 0; 512 #ifdef MROUTING 513 extern struct socket *ip_mrouter; 514 #endif 515 516 if (req == PRU_CONTROL) 517 return (in_control(so, (long)m, (caddr_t)nam, 518 (struct ifnet *)control, p)); 519 520 if (req == PRU_PURGEIF) { 521 in_pcbpurgeif0(&rawcbtable, (struct ifnet *)control); 522 in_purgeif((struct ifnet *)control); 523 in_pcbpurgeif(&rawcbtable, (struct ifnet *)control); 524 return (0); 525 } 526 527 s = splsoftnet(); 528 inp = sotoinpcb(so); 529 #ifdef DIAGNOSTIC 530 if (req != PRU_SEND && req != PRU_SENDOOB && control) 531 panic("rip_usrreq: unexpected control mbuf"); 532 #endif 533 if (inp == 0 && req != PRU_ATTACH) { 534 error = EINVAL; 535 goto release; 536 } 537 538 switch (req) { 539 540 case PRU_ATTACH: 541 if (inp != 0) { 542 error = EISCONN; 543 break; 544 } 545 if (p == 0 || (error = suser(p->p_ucred, &p->p_acflag))) { 546 error = EACCES; 547 break; 548 } 549 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 550 error = soreserve(so, rip_sendspace, rip_recvspace); 551 if (error) 552 break; 553 } 554 error = in_pcballoc(so, &rawcbtable); 555 if (error) 556 break; 557 inp = sotoinpcb(so); 558 inp->inp_ip.ip_p = (long)nam; 559 break; 560 561 case PRU_DETACH: 562 #ifdef MROUTING 563 if (so == ip_mrouter) 564 ip_mrouter_done(); 565 #endif 566 in_pcbdetach(inp); 567 break; 568 569 case PRU_BIND: 570 error = rip_bind(inp, nam); 571 break; 572 573 case PRU_LISTEN: 574 error = EOPNOTSUPP; 575 break; 576 577 case PRU_CONNECT: 578 error = rip_connect(inp, nam); 579 if (error) 580 break; 581 soisconnected(so); 582 break; 583 584 case PRU_CONNECT2: 585 error = EOPNOTSUPP; 586 break; 587 588 case PRU_DISCONNECT: 589 soisdisconnected(so); 590 rip_disconnect(inp); 591 break; 592 593 /* 594 * Mark the connection as being incapable of further input. 595 */ 596 case PRU_SHUTDOWN: 597 socantsendmore(so); 598 break; 599 600 case PRU_RCVD: 601 error = EOPNOTSUPP; 602 break; 603 604 /* 605 * Ship a packet out. The appropriate raw output 606 * routine handles any massaging necessary. 607 */ 608 case PRU_SEND: 609 if (control && control->m_len) { 610 m_freem(control); 611 m_freem(m); 612 error = EINVAL; 613 break; 614 } 615 { 616 if (nam) { 617 if ((so->so_state & SS_ISCONNECTED) != 0) { 618 error = EISCONN; 619 goto die; 620 } 621 error = rip_connect(inp, nam); 622 if (error) { 623 die: 624 m_freem(m); 625 break; 626 } 627 } else { 628 if ((so->so_state & SS_ISCONNECTED) == 0) { 629 error = ENOTCONN; 630 goto die; 631 } 632 } 633 error = rip_output(m, inp); 634 if (nam) 635 rip_disconnect(inp); 636 } 637 break; 638 639 case PRU_SENSE: 640 /* 641 * stat: don't bother with a blocksize. 642 */ 643 splx(s); 644 return (0); 645 646 case PRU_RCVOOB: 647 error = EOPNOTSUPP; 648 break; 649 650 case PRU_SENDOOB: 651 m_freem(control); 652 m_freem(m); 653 error = EOPNOTSUPP; 654 break; 655 656 case PRU_SOCKADDR: 657 in_setsockaddr(inp, nam); 658 break; 659 660 case PRU_PEERADDR: 661 in_setpeeraddr(inp, nam); 662 break; 663 664 default: 665 panic("rip_usrreq"); 666 } 667 668 release: 669 splx(s); 670 return (error); 671 } 672
Indexes created Mon Sep 29 21:09:56 GMT 2025