1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)ip_output.c 8.3 (Berkeley) 1/21/94 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_ipfw.h" 36 #include "opt_inet.h" 37 #include "opt_ipsec.h" 38 #include "opt_mac.h" 39 #include "opt_mbuf_stress_test.h" 40 #include "opt_mpath.h" 41 #include "opt_sctp.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/kernel.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/priv.h> 49 #include <sys/proc.h> 50 #include <sys/protosw.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/sysctl.h> 54 #include <sys/ucred.h> 55 #include <sys/vimage.h> 56 57 #include <net/if.h> 58 #include <net/netisr.h> 59 #include <net/pfil.h> 60 #include <net/route.h> 61 #ifdef RADIX_MPATH 62 #include <net/radix_mpath.h> 63 #endif 64 #include <net/vnet.h> 65 66 #include <netinet/in.h> 67 #include <netinet/in_systm.h> 68 #include <netinet/ip.h> 69 #include <netinet/in_pcb.h> 70 #include <netinet/in_var.h> 71 #include <netinet/ip_var.h> 72 #include <netinet/ip_options.h> 73 #include <netinet/vinet.h> 74 #ifdef SCTP 75 #include <netinet/sctp.h> 76 #include <netinet/sctp_crc32.h> 77 #endif 78 79 #ifdef IPSEC 80 #include <netinet/ip_ipsec.h> 81 #include <netipsec/ipsec.h> 82 #endif /* IPSEC*/ 83 84 #include <machine/in_cksum.h> 85 86 #include <security/mac/mac_framework.h> 87 88 #define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 89 x, (ntohl(a.s_addr)>>24)&0xFF,\ 90 (ntohl(a.s_addr)>>16)&0xFF,\ 91 (ntohl(a.s_addr)>>8)&0xFF,\ 92 (ntohl(a.s_addr))&0xFF, y); 93 94 #ifdef VIMAGE_GLOBALS 95 u_short ip_id; 96 #endif 97 98 #ifdef MBUF_STRESS_TEST 99 int mbuf_frag_size = 0; 100 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 101 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 102 #endif 103 104 #if defined(IP_NONLOCALBIND) 105 static int ip_nonlocalok = 0; 106 SYSCTL_INT(_net_inet_ip, OID_AUTO, nonlocalok, 107 CTLFLAG_RW|CTLFLAG_SECURE, &ip_nonlocalok, 0, ""); 108 #endif 109 110 static void ip_mloopback 111 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 112 113 114 extern struct protosw inetsw[]; 115 116 /* 117 * IP output. The packet in mbuf chain m contains a skeletal IP 118 * header (with len, off, ttl, proto, tos, src, dst). 119 * The mbuf chain containing the packet will be freed. 120 * The mbuf opt, if present, will not be freed. 121 * In the IP forwarding case, the packet will arrive with options already 122 * inserted, so must have a NULL opt pointer. 123 */ 124 int 125 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, 126 struct ip_moptions *imo, struct inpcb *inp) 127 { 128 INIT_VNET_NET(curvnet); 129 INIT_VNET_INET(curvnet); 130 struct ip *ip; 131 struct ifnet *ifp = NULL; /* keep compiler happy */ 132 struct mbuf *m0; 133 int hlen = sizeof (struct ip); 134 int mtu; 135 int len, error = 0; 136 struct sockaddr_in *dst = NULL; /* keep compiler happy */ 137 struct in_ifaddr *ia = NULL; 138 int isbroadcast, sw_csum; 139 struct route iproute; 140 struct in_addr odst; 141 #ifdef IPFIREWALL_FORWARD 142 struct m_tag *fwd_tag = NULL; 143 #endif 144 M_ASSERTPKTHDR(m); 145 146 if (ro == NULL) { 147 ro = &iproute; 148 bzero(ro, sizeof (*ro)); 149 } 150 151 if (inp != NULL) { 152 M_SETFIB(m, inp->inp_inc.inc_fibnum); 153 INP_LOCK_ASSERT(inp); 154 } 155 156 if (opt) { 157 len = 0; 158 m = ip_insertoptions(m, opt, &len); 159 if (len != 0) 160 hlen = len; 161 } 162 ip = mtod(m, struct ip *); 163 164 /* 165 * Fill in IP header. If we are not allowing fragmentation, 166 * then the ip_id field is meaningless, but we don't set it 167 * to zero. Doing so causes various problems when devices along 168 * the path (routers, load balancers, firewalls, etc.) illegally 169 * disable DF on our packet. Note that a 16-bit counter 170 * will wrap around in less than 10 seconds at 100 Mbit/s on a 171 * medium with MTU 1500. See Steven M. Bellovin, "A Technique 172 * for Counting NATted Hosts", Proc. IMW'02, available at 173 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>. 174 */ 175 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 176 ip->ip_v = IPVERSION; 177 ip->ip_hl = hlen >> 2; 178 ip->ip_id = ip_newid(); 179 V_ipstat.ips_localout++; 180 } else { 181 hlen = ip->ip_hl << 2; 182 } 183 184 dst = (struct sockaddr_in *)&ro->ro_dst; 185 again: 186 /* 187 * If there is a cached route, 188 * check that it is to the same destination 189 * and is still up. If not, free it and try again. 190 * The address family should also be checked in case of sharing the 191 * cache with IPv6. 192 */ 193 if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 194 dst->sin_family != AF_INET || 195 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 196 RTFREE(ro->ro_rt); 197 ro->ro_rt = (struct rtentry *)NULL; 198 } 199 #ifdef IPFIREWALL_FORWARD 200 if (ro->ro_rt == NULL && fwd_tag == NULL) { 201 #else 202 if (ro->ro_rt == NULL) { 203 #endif 204 bzero(dst, sizeof(*dst)); 205 dst->sin_family = AF_INET; 206 dst->sin_len = sizeof(*dst); 207 dst->sin_addr = ip->ip_dst; 208 } 209 /* 210 * If routing to interface only, short circuit routing lookup. 211 * The use of an all-ones broadcast address implies this; an 212 * interface is specified by the broadcast address of an interface, 213 * or the destination address of a ptp interface. 214 */ 215 if (flags & IP_SENDONES) { 216 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 217 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 218 V_ipstat.ips_noroute++; 219 error = ENETUNREACH; 220 goto bad; 221 } 222 ip->ip_dst.s_addr = INADDR_BROADCAST; 223 dst->sin_addr = ip->ip_dst; 224 ifp = ia->ia_ifp; 225 ip->ip_ttl = 1; 226 isbroadcast = 1; 227 } else if (flags & IP_ROUTETOIF) { 228 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 229 (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) { 230 V_ipstat.ips_noroute++; 231 error = ENETUNREACH; 232 goto bad; 233 } 234 ifp = ia->ia_ifp; 235 ip->ip_ttl = 1; 236 isbroadcast = in_broadcast(dst->sin_addr, ifp); 237 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 238 imo != NULL && imo->imo_multicast_ifp != NULL) { 239 /* 240 * Bypass the normal routing lookup for multicast 241 * packets if the interface is specified. 242 */ 243 ifp = imo->imo_multicast_ifp; 244 IFP_TO_IA(ifp, ia); 245 isbroadcast = 0; /* fool gcc */ 246 } else { 247 /* 248 * We want to do any cloning requested by the link layer, 249 * as this is probably required in all cases for correct 250 * operation (as it is for ARP). 251 */ 252 if (ro->ro_rt == NULL) 253 #ifdef RADIX_MPATH 254 rtalloc_mpath_fib(ro, 255 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 256 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 257 #else 258 in_rtalloc_ign(ro, 0, 259 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 260 #endif 261 if (ro->ro_rt == NULL) { 262 V_ipstat.ips_noroute++; 263 error = EHOSTUNREACH; 264 goto bad; 265 } 266 ia = ifatoia(ro->ro_rt->rt_ifa); 267 ifp = ro->ro_rt->rt_ifp; 268 ro->ro_rt->rt_rmx.rmx_pksent++; 269 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 270 dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 271 if (ro->ro_rt->rt_flags & RTF_HOST) 272 isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 273 else 274 isbroadcast = in_broadcast(dst->sin_addr, ifp); 275 } 276 /* 277 * Calculate MTU. If we have a route that is up, use that, 278 * otherwise use the interface's MTU. 279 */ 280 if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) { 281 /* 282 * This case can happen if the user changed the MTU 283 * of an interface after enabling IP on it. Because 284 * most netifs don't keep track of routes pointing to 285 * them, there is no way for one to update all its 286 * routes when the MTU is changed. 287 */ 288 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu) 289 ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 290 mtu = ro->ro_rt->rt_rmx.rmx_mtu; 291 } else { 292 mtu = ifp->if_mtu; 293 } 294 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 295 struct in_multi *inm; 296 297 m->m_flags |= M_MCAST; 298 /* 299 * IP destination address is multicast. Make sure "dst" 300 * still points to the address in "ro". (It may have been 301 * changed to point to a gateway address, above.) 302 */ 303 dst = (struct sockaddr_in *)&ro->ro_dst; 304 /* 305 * See if the caller provided any multicast options 306 */ 307 if (imo != NULL) { 308 ip->ip_ttl = imo->imo_multicast_ttl; 309 if (imo->imo_multicast_vif != -1) 310 ip->ip_src.s_addr = 311 ip_mcast_src ? 312 ip_mcast_src(imo->imo_multicast_vif) : 313 INADDR_ANY; 314 } else 315 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 316 /* 317 * Confirm that the outgoing interface supports multicast. 318 */ 319 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 320 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 321 V_ipstat.ips_noroute++; 322 error = ENETUNREACH; 323 goto bad; 324 } 325 } 326 /* 327 * If source address not specified yet, use address 328 * of outgoing interface. 329 */ 330 if (ip->ip_src.s_addr == INADDR_ANY) { 331 /* Interface may have no addresses. */ 332 if (ia != NULL) 333 ip->ip_src = IA_SIN(ia)->sin_addr; 334 } 335 336 IN_MULTI_LOCK(); 337 IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm); 338 if (inm != NULL && 339 (imo == NULL || imo->imo_multicast_loop)) { 340 IN_MULTI_UNLOCK(); 341 /* 342 * If we belong to the destination multicast group 343 * on the outgoing interface, and the caller did not 344 * forbid loopback, loop back a copy. 345 */ 346 ip_mloopback(ifp, m, dst, hlen); 347 } 348 else { 349 IN_MULTI_UNLOCK(); 350 /* 351 * If we are acting as a multicast router, perform 352 * multicast forwarding as if the packet had just 353 * arrived on the interface to which we are about 354 * to send. The multicast forwarding function 355 * recursively calls this function, using the 356 * IP_FORWARDING flag to prevent infinite recursion. 357 * 358 * Multicasts that are looped back by ip_mloopback(), 359 * above, will be forwarded by the ip_input() routine, 360 * if necessary. 361 */ 362 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 363 /* 364 * If rsvp daemon is not running, do not 365 * set ip_moptions. This ensures that the packet 366 * is multicast and not just sent down one link 367 * as prescribed by rsvpd. 368 */ 369 if (!V_rsvp_on) 370 imo = NULL; 371 if (ip_mforward && 372 ip_mforward(ip, ifp, m, imo) != 0) { 373 m_freem(m); 374 goto done; 375 } 376 } 377 } 378 379 /* 380 * Multicasts with a time-to-live of zero may be looped- 381 * back, above, but must not be transmitted on a network. 382 * Also, multicasts addressed to the loopback interface 383 * are not sent -- the above call to ip_mloopback() will 384 * loop back a copy if this host actually belongs to the 385 * destination group on the loopback interface. 386 */ 387 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 388 m_freem(m); 389 goto done; 390 } 391 392 goto sendit; 393 } 394 395 /* 396 * If the source address is not specified yet, use the address 397 * of the outoing interface. 398 */ 399 if (ip->ip_src.s_addr == INADDR_ANY) { 400 /* Interface may have no addresses. */ 401 if (ia != NULL) { 402 ip->ip_src = IA_SIN(ia)->sin_addr; 403 } 404 } 405 406 /* 407 * Verify that we have any chance at all of being able to queue the 408 * packet or packet fragments, unless ALTQ is enabled on the given 409 * interface in which case packetdrop should be done by queueing. 410 */ 411 #ifdef ALTQ 412 if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) && 413 ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 414 ifp->if_snd.ifq_maxlen)) 415 #else 416 if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 417 ifp->if_snd.ifq_maxlen) 418 #endif /* ALTQ */ 419 { 420 error = ENOBUFS; 421 V_ipstat.ips_odropped++; 422 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1); 423 goto bad; 424 } 425 426 /* 427 * Look for broadcast address and 428 * verify user is allowed to send 429 * such a packet. 430 */ 431 if (isbroadcast) { 432 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 433 error = EADDRNOTAVAIL; 434 goto bad; 435 } 436 if ((flags & IP_ALLOWBROADCAST) == 0) { 437 error = EACCES; 438 goto bad; 439 } 440 /* don't allow broadcast messages to be fragmented */ 441 if (ip->ip_len > mtu) { 442 error = EMSGSIZE; 443 goto bad; 444 } 445 m->m_flags |= M_BCAST; 446 } else { 447 m->m_flags &= ~M_BCAST; 448 } 449 450 sendit: 451 #ifdef IPSEC 452 switch(ip_ipsec_output(&m, inp, &flags, &error, &ro, &iproute, &dst, &ia, &ifp)) { 453 case 1: 454 goto bad; 455 case -1: 456 goto done; 457 case 0: 458 default: 459 break; /* Continue with packet processing. */ 460 } 461 /* Update variables that are affected by ipsec4_output(). */ 462 ip = mtod(m, struct ip *); 463 hlen = ip->ip_hl << 2; 464 #endif /* IPSEC */ 465 466 /* Jump over all PFIL processing if hooks are not active. */ 467 if (!PFIL_HOOKED(&inet_pfil_hook)) 468 goto passout; 469 470 /* Run through list of hooks for output packets. */ 471 odst.s_addr = ip->ip_dst.s_addr; 472 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 473 if (error != 0 || m == NULL) 474 goto done; 475 476 ip = mtod(m, struct ip *); 477 478 /* See if destination IP address was changed by packet filter. */ 479 if (odst.s_addr != ip->ip_dst.s_addr) { 480 m->m_flags |= M_SKIP_FIREWALL; 481 /* If destination is now ourself drop to ip_input(). */ 482 if (in_localip(ip->ip_dst)) { 483 m->m_flags |= M_FASTFWD_OURS; 484 if (m->m_pkthdr.rcvif == NULL) 485 m->m_pkthdr.rcvif = V_loif; 486 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 487 m->m_pkthdr.csum_flags |= 488 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 489 m->m_pkthdr.csum_data = 0xffff; 490 } 491 m->m_pkthdr.csum_flags |= 492 CSUM_IP_CHECKED | CSUM_IP_VALID; 493 #ifdef SCTP 494 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 495 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 496 #endif 497 error = netisr_queue(NETISR_IP, m); 498 goto done; 499 } else 500 goto again; /* Redo the routing table lookup. */ 501 } 502 503 #ifdef IPFIREWALL_FORWARD 504 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 505 if (m->m_flags & M_FASTFWD_OURS) { 506 if (m->m_pkthdr.rcvif == NULL) 507 m->m_pkthdr.rcvif = V_loif; 508 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 509 m->m_pkthdr.csum_flags |= 510 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 511 m->m_pkthdr.csum_data = 0xffff; 512 } 513 #ifdef SCTP 514 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 515 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 516 #endif 517 m->m_pkthdr.csum_flags |= 518 CSUM_IP_CHECKED | CSUM_IP_VALID; 519 520 error = netisr_queue(NETISR_IP, m); 521 goto done; 522 } 523 /* Or forward to some other address? */ 524 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 525 if (fwd_tag) { 526 dst = (struct sockaddr_in *)&ro->ro_dst; 527 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 528 m->m_flags |= M_SKIP_FIREWALL; 529 m_tag_delete(m, fwd_tag); 530 goto again; 531 } 532 #endif /* IPFIREWALL_FORWARD */ 533 534 passout: 535 /* 127/8 must not appear on wire - RFC1122. */ 536 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 537 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 538 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 539 V_ipstat.ips_badaddr++; 540 error = EADDRNOTAVAIL; 541 goto bad; 542 } 543 } 544 545 m->m_pkthdr.csum_flags |= CSUM_IP; 546 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 547 if (sw_csum & CSUM_DELAY_DATA) { 548 in_delayed_cksum(m); 549 sw_csum &= ~CSUM_DELAY_DATA; 550 } 551 #ifdef SCTP 552 if (sw_csum & CSUM_SCTP) { 553 sctp_delayed_cksum(m); 554 sw_csum &= ~CSUM_SCTP; 555 } 556 #endif 557 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 558 559 /* 560 * If small enough for interface, or the interface will take 561 * care of the fragmentation for us, we can just send directly. 562 */ 563 if (ip->ip_len <= mtu || 564 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 565 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 566 ip->ip_len = htons(ip->ip_len); 567 ip->ip_off = htons(ip->ip_off); 568 ip->ip_sum = 0; 569 if (sw_csum & CSUM_DELAY_IP) 570 ip->ip_sum = in_cksum(m, hlen); 571 572 /* 573 * Record statistics for this interface address. 574 * With CSUM_TSO the byte/packet count will be slightly 575 * incorrect because we count the IP+TCP headers only 576 * once instead of for every generated packet. 577 */ 578 if (!(flags & IP_FORWARDING) && ia) { 579 if (m->m_pkthdr.csum_flags & CSUM_TSO) 580 ia->ia_ifa.if_opackets += 581 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 582 else 583 ia->ia_ifa.if_opackets++; 584 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 585 } 586 #ifdef MBUF_STRESS_TEST 587 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 588 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 589 #endif 590 /* 591 * Reset layer specific mbuf flags 592 * to avoid confusing lower layers. 593 */ 594 m->m_flags &= ~(M_PROTOFLAGS); 595 error = (*ifp->if_output)(ifp, m, 596 (struct sockaddr *)dst, ro->ro_rt); 597 goto done; 598 } 599 600 /* Balk when DF bit is set or the interface didn't support TSO. */ 601 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 602 error = EMSGSIZE; 603 V_ipstat.ips_cantfrag++; 604 goto bad; 605 } 606 607 /* 608 * Too large for interface; fragment if possible. If successful, 609 * on return, m will point to a list of packets to be sent. 610 */ 611 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 612 if (error) 613 goto bad; 614 for (; m; m = m0) { 615 m0 = m->m_nextpkt; 616 m->m_nextpkt = 0; 617 if (error == 0) { 618 /* Record statistics for this interface address. */ 619 if (ia != NULL) { 620 ia->ia_ifa.if_opackets++; 621 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 622 } 623 /* 624 * Reset layer specific mbuf flags 625 * to avoid confusing upper layers. 626 */ 627 m->m_flags &= ~(M_PROTOFLAGS); 628 629 error = (*ifp->if_output)(ifp, m, 630 (struct sockaddr *)dst, ro->ro_rt); 631 } else 632 m_freem(m); 633 } 634 635 if (error == 0) 636 V_ipstat.ips_fragmented++; 637 638 done: 639 if (ro == &iproute && ro->ro_rt) { 640 RTFREE(ro->ro_rt); 641 } 642 return (error); 643 bad: 644 m_freem(m); 645 goto done; 646 } 647 648 /* 649 * Create a chain of fragments which fit the given mtu. m_frag points to the 650 * mbuf to be fragmented; on return it points to the chain with the fragments. 651 * Return 0 if no error. If error, m_frag may contain a partially built 652 * chain of fragments that should be freed by the caller. 653 * 654 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 655 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 656 */ 657 int 658 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 659 u_long if_hwassist_flags, int sw_csum) 660 { 661 INIT_VNET_INET(curvnet); 662 int error = 0; 663 int hlen = ip->ip_hl << 2; 664 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 665 int off; 666 struct mbuf *m0 = *m_frag; /* the original packet */ 667 int firstlen; 668 struct mbuf **mnext; 669 int nfrags; 670 671 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 672 V_ipstat.ips_cantfrag++; 673 return EMSGSIZE; 674 } 675 676 /* 677 * Must be able to put at least 8 bytes per fragment. 678 */ 679 if (len < 8) 680 return EMSGSIZE; 681 682 /* 683 * If the interface will not calculate checksums on 684 * fragmented packets, then do it here. 685 */ 686 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 687 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 688 in_delayed_cksum(m0); 689 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 690 } 691 #ifdef SCTP 692 if (m0->m_pkthdr.csum_flags & CSUM_SCTP && 693 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 694 sctp_delayed_cksum(m0); 695 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 696 } 697 #endif 698 if (len > PAGE_SIZE) { 699 /* 700 * Fragment large datagrams such that each segment 701 * contains a multiple of PAGE_SIZE amount of data, 702 * plus headers. This enables a receiver to perform 703 * page-flipping zero-copy optimizations. 704 * 705 * XXX When does this help given that sender and receiver 706 * could have different page sizes, and also mtu could 707 * be less than the receiver's page size ? 708 */ 709 int newlen; 710 struct mbuf *m; 711 712 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 713 off += m->m_len; 714 715 /* 716 * firstlen (off - hlen) must be aligned on an 717 * 8-byte boundary 718 */ 719 if (off < hlen) 720 goto smart_frag_failure; 721 off = ((off - hlen) & ~7) + hlen; 722 newlen = (~PAGE_MASK) & mtu; 723 if ((newlen + sizeof (struct ip)) > mtu) { 724 /* we failed, go back the default */ 725 smart_frag_failure: 726 newlen = len; 727 off = hlen + len; 728 } 729 len = newlen; 730 731 } else { 732 off = hlen + len; 733 } 734 735 firstlen = off - hlen; 736 mnext = &m0->m_nextpkt; /* pointer to next packet */ 737 738 /* 739 * Loop through length of segment after first fragment, 740 * make new header and copy data of each part and link onto chain. 741 * Here, m0 is the original packet, m is the fragment being created. 742 * The fragments are linked off the m_nextpkt of the original 743 * packet, which after processing serves as the first fragment. 744 */ 745 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 746 struct ip *mhip; /* ip header on the fragment */ 747 struct mbuf *m; 748 int mhlen = sizeof (struct ip); 749 750 MGETHDR(m, M_DONTWAIT, MT_DATA); 751 if (m == NULL) { 752 error = ENOBUFS; 753 V_ipstat.ips_odropped++; 754 goto done; 755 } 756 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 757 /* 758 * In the first mbuf, leave room for the link header, then 759 * copy the original IP header including options. The payload 760 * goes into an additional mbuf chain returned by m_copy(). 761 */ 762 m->m_data += max_linkhdr; 763 mhip = mtod(m, struct ip *); 764 *mhip = *ip; 765 if (hlen > sizeof (struct ip)) { 766 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 767 mhip->ip_v = IPVERSION; 768 mhip->ip_hl = mhlen >> 2; 769 } 770 m->m_len = mhlen; 771 /* XXX do we need to add ip->ip_off below ? */ 772 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 773 if (off + len >= ip->ip_len) { /* last fragment */ 774 len = ip->ip_len - off; 775 m->m_flags |= M_LASTFRAG; 776 } else 777 mhip->ip_off |= IP_MF; 778 mhip->ip_len = htons((u_short)(len + mhlen)); 779 m->m_next = m_copy(m0, off, len); 780 if (m->m_next == NULL) { /* copy failed */ 781 m_free(m); 782 error = ENOBUFS; /* ??? */ 783 V_ipstat.ips_odropped++; 784 goto done; 785 } 786 m->m_pkthdr.len = mhlen + len; 787 m->m_pkthdr.rcvif = NULL; 788 #ifdef MAC 789 mac_netinet_fragment(m0, m); 790 #endif 791 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 792 mhip->ip_off = htons(mhip->ip_off); 793 mhip->ip_sum = 0; 794 if (sw_csum & CSUM_DELAY_IP) 795 mhip->ip_sum = in_cksum(m, mhlen); 796 *mnext = m; 797 mnext = &m->m_nextpkt; 798 } 799 V_ipstat.ips_ofragments += nfrags; 800 801 /* set first marker for fragment chain */ 802 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 803 m0->m_pkthdr.csum_data = nfrags; 804 805 /* 806 * Update first fragment by trimming what's been copied out 807 * and updating header. 808 */ 809 m_adj(m0, hlen + firstlen - ip->ip_len); 810 m0->m_pkthdr.len = hlen + firstlen; 811 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 812 ip->ip_off |= IP_MF; 813 ip->ip_off = htons(ip->ip_off); 814 ip->ip_sum = 0; 815 if (sw_csum & CSUM_DELAY_IP) 816 ip->ip_sum = in_cksum(m0, hlen); 817 818 done: 819 *m_frag = m0; 820 return error; 821 } 822 823 void 824 in_delayed_cksum(struct mbuf *m) 825 { 826 struct ip *ip; 827 u_short csum, offset; 828 829 ip = mtod(m, struct ip *); 830 offset = ip->ip_hl << 2 ; 831 csum = in_cksum_skip(m, ip->ip_len, offset); 832 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 833 csum = 0xffff; 834 offset += m->m_pkthdr.csum_data; /* checksum offset */ 835 836 if (offset + sizeof(u_short) > m->m_len) { 837 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 838 m->m_len, offset, ip->ip_p); 839 /* 840 * XXX 841 * this shouldn't happen, but if it does, the 842 * correct behavior may be to insert the checksum 843 * in the appropriate next mbuf in the chain. 844 */ 845 return; 846 } 847 *(u_short *)(m->m_data + offset) = csum; 848 } 849 850 /* 851 * IP socket option processing. 852 */ 853 int 854 ip_ctloutput(struct socket *so, struct sockopt *sopt) 855 { 856 struct inpcb *inp = sotoinpcb(so); 857 int error, optval; 858 859 error = optval = 0; 860 if (sopt->sopt_level != IPPROTO_IP) { 861 if ((sopt->sopt_level == SOL_SOCKET) && 862 (sopt->sopt_name == SO_SETFIB)) { 863 inp->inp_inc.inc_fibnum = so->so_fibnum; 864 return (0); 865 } 866 return (EINVAL); 867 } 868 869 switch (sopt->sopt_dir) { 870 case SOPT_SET: 871 switch (sopt->sopt_name) { 872 case IP_OPTIONS: 873 #ifdef notyet 874 case IP_RETOPTS: 875 #endif 876 { 877 struct mbuf *m; 878 if (sopt->sopt_valsize > MLEN) { 879 error = EMSGSIZE; 880 break; 881 } 882 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 883 if (m == NULL) { 884 error = ENOBUFS; 885 break; 886 } 887 m->m_len = sopt->sopt_valsize; 888 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 889 m->m_len); 890 if (error) { 891 m_free(m); 892 break; 893 } 894 INP_WLOCK(inp); 895 error = ip_pcbopts(inp, sopt->sopt_name, m); 896 INP_WUNLOCK(inp); 897 return (error); 898 } 899 900 #if defined(IP_NONLOCALBIND) 901 case IP_NONLOCALOK: 902 if (! ip_nonlocalok) { 903 error = ENOPROTOOPT; 904 break; 905 } 906 /* FALLTHROUGH */ 907 #endif 908 case IP_TOS: 909 case IP_TTL: 910 case IP_MINTTL: 911 case IP_RECVOPTS: 912 case IP_RECVRETOPTS: 913 case IP_RECVDSTADDR: 914 case IP_RECVTTL: 915 case IP_RECVIF: 916 case IP_FAITH: 917 case IP_ONESBCAST: 918 case IP_DONTFRAG: 919 error = sooptcopyin(sopt, &optval, sizeof optval, 920 sizeof optval); 921 if (error) 922 break; 923 924 switch (sopt->sopt_name) { 925 case IP_TOS: 926 inp->inp_ip_tos = optval; 927 break; 928 929 case IP_TTL: 930 inp->inp_ip_ttl = optval; 931 break; 932 933 case IP_MINTTL: 934 if (optval >= 0 && optval <= MAXTTL) 935 inp->inp_ip_minttl = optval; 936 else 937 error = EINVAL; 938 break; 939 940 #define OPTSET(bit) do { \ 941 INP_WLOCK(inp); \ 942 if (optval) \ 943 inp->inp_flags |= bit; \ 944 else \ 945 inp->inp_flags &= ~bit; \ 946 INP_WUNLOCK(inp); \ 947 } while (0) 948 949 case IP_RECVOPTS: 950 OPTSET(INP_RECVOPTS); 951 break; 952 953 case IP_RECVRETOPTS: 954 OPTSET(INP_RECVRETOPTS); 955 break; 956 957 case IP_RECVDSTADDR: 958 OPTSET(INP_RECVDSTADDR); 959 break; 960 961 case IP_RECVTTL: 962 OPTSET(INP_RECVTTL); 963 break; 964 965 case IP_RECVIF: 966 OPTSET(INP_RECVIF); 967 break; 968 969 case IP_FAITH: 970 OPTSET(INP_FAITH); 971 break; 972 973 case IP_ONESBCAST: 974 OPTSET(INP_ONESBCAST); 975 break; 976 case IP_DONTFRAG: 977 OPTSET(INP_DONTFRAG); 978 break; 979 #if defined(IP_NONLOCALBIND) 980 case IP_NONLOCALOK: 981 OPTSET(INP_NONLOCALOK); 982 break; 983 #endif 984 } 985 break; 986 #undef OPTSET 987 988 /* 989 * Multicast socket options are processed by the in_mcast 990 * module. 991 */ 992 case IP_MULTICAST_IF: 993 case IP_MULTICAST_VIF: 994 case IP_MULTICAST_TTL: 995 case IP_MULTICAST_LOOP: 996 case IP_ADD_MEMBERSHIP: 997 case IP_DROP_MEMBERSHIP: 998 case IP_ADD_SOURCE_MEMBERSHIP: 999 case IP_DROP_SOURCE_MEMBERSHIP: 1000 case IP_BLOCK_SOURCE: 1001 case IP_UNBLOCK_SOURCE: 1002 case IP_MSFILTER: 1003 case MCAST_JOIN_GROUP: 1004 case MCAST_LEAVE_GROUP: 1005 case MCAST_JOIN_SOURCE_GROUP: 1006 case MCAST_LEAVE_SOURCE_GROUP: 1007 case MCAST_BLOCK_SOURCE: 1008 case MCAST_UNBLOCK_SOURCE: 1009 error = inp_setmoptions(inp, sopt); 1010 break; 1011 1012 case IP_PORTRANGE: 1013 error = sooptcopyin(sopt, &optval, sizeof optval, 1014 sizeof optval); 1015 if (error) 1016 break; 1017 1018 INP_WLOCK(inp); 1019 switch (optval) { 1020 case IP_PORTRANGE_DEFAULT: 1021 inp->inp_flags &= ~(INP_LOWPORT); 1022 inp->inp_flags &= ~(INP_HIGHPORT); 1023 break; 1024 1025 case IP_PORTRANGE_HIGH: 1026 inp->inp_flags &= ~(INP_LOWPORT); 1027 inp->inp_flags |= INP_HIGHPORT; 1028 break; 1029 1030 case IP_PORTRANGE_LOW: 1031 inp->inp_flags &= ~(INP_HIGHPORT); 1032 inp->inp_flags |= INP_LOWPORT; 1033 break; 1034 1035 default: 1036 error = EINVAL; 1037 break; 1038 } 1039 INP_WUNLOCK(inp); 1040 break; 1041 1042 #ifdef IPSEC 1043 case IP_IPSEC_POLICY: 1044 { 1045 caddr_t req; 1046 struct mbuf *m; 1047 1048 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1049 break; 1050 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1051 break; 1052 req = mtod(m, caddr_t); 1053 error = ipsec_set_policy(inp, sopt->sopt_name, req, 1054 m->m_len, (sopt->sopt_td != NULL) ? 1055 sopt->sopt_td->td_ucred : NULL); 1056 m_freem(m); 1057 break; 1058 } 1059 #endif /* IPSEC */ 1060 1061 default: 1062 error = ENOPROTOOPT; 1063 break; 1064 } 1065 break; 1066 1067 case SOPT_GET: 1068 switch (sopt->sopt_name) { 1069 case IP_OPTIONS: 1070 case IP_RETOPTS: 1071 if (inp->inp_options) 1072 error = sooptcopyout(sopt, 1073 mtod(inp->inp_options, 1074 char *), 1075 inp->inp_options->m_len); 1076 else 1077 sopt->sopt_valsize = 0; 1078 break; 1079 1080 case IP_TOS: 1081 case IP_TTL: 1082 case IP_MINTTL: 1083 case IP_RECVOPTS: 1084 case IP_RECVRETOPTS: 1085 case IP_RECVDSTADDR: 1086 case IP_RECVTTL: 1087 case IP_RECVIF: 1088 case IP_PORTRANGE: 1089 case IP_FAITH: 1090 case IP_ONESBCAST: 1091 case IP_DONTFRAG: 1092 switch (sopt->sopt_name) { 1093 1094 case IP_TOS: 1095 optval = inp->inp_ip_tos; 1096 break; 1097 1098 case IP_TTL: 1099 optval = inp->inp_ip_ttl; 1100 break; 1101 1102 case IP_MINTTL: 1103 optval = inp->inp_ip_minttl; 1104 break; 1105 1106 #define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1107 1108 case IP_RECVOPTS: 1109 optval = OPTBIT(INP_RECVOPTS); 1110 break; 1111 1112 case IP_RECVRETOPTS: 1113 optval = OPTBIT(INP_RECVRETOPTS); 1114 break; 1115 1116 case IP_RECVDSTADDR: 1117 optval = OPTBIT(INP_RECVDSTADDR); 1118 break; 1119 1120 case IP_RECVTTL: 1121 optval = OPTBIT(INP_RECVTTL); 1122 break; 1123 1124 case IP_RECVIF: 1125 optval = OPTBIT(INP_RECVIF); 1126 break; 1127 1128 case IP_PORTRANGE: 1129 if (inp->inp_flags & INP_HIGHPORT) 1130 optval = IP_PORTRANGE_HIGH; 1131 else if (inp->inp_flags & INP_LOWPORT) 1132 optval = IP_PORTRANGE_LOW; 1133 else 1134 optval = 0; 1135 break; 1136 1137 case IP_FAITH: 1138 optval = OPTBIT(INP_FAITH); 1139 break; 1140 1141 case IP_ONESBCAST: 1142 optval = OPTBIT(INP_ONESBCAST); 1143 break; 1144 case IP_DONTFRAG: 1145 optval = OPTBIT(INP_DONTFRAG); 1146 break; 1147 } 1148 error = sooptcopyout(sopt, &optval, sizeof optval); 1149 break; 1150 1151 /* 1152 * Multicast socket options are processed by the in_mcast 1153 * module. 1154 */ 1155 case IP_MULTICAST_IF: 1156 case IP_MULTICAST_VIF: 1157 case IP_MULTICAST_TTL: 1158 case IP_MULTICAST_LOOP: 1159 case IP_MSFILTER: 1160 error = inp_getmoptions(inp, sopt); 1161 break; 1162 1163 #ifdef IPSEC 1164 case IP_IPSEC_POLICY: 1165 { 1166 struct mbuf *m = NULL; 1167 caddr_t req = NULL; 1168 size_t len = 0; 1169 1170 if (m != 0) { 1171 req = mtod(m, caddr_t); 1172 len = m->m_len; 1173 } 1174 error = ipsec_get_policy(sotoinpcb(so), req, len, &m); 1175 if (error == 0) 1176 error = soopt_mcopyout(sopt, m); /* XXX */ 1177 if (error == 0) 1178 m_freem(m); 1179 break; 1180 } 1181 #endif /* IPSEC */ 1182 1183 default: 1184 error = ENOPROTOOPT; 1185 break; 1186 } 1187 break; 1188 } 1189 return (error); 1190 } 1191 1192 /* 1193 * Routine called from ip_output() to loop back a copy of an IP multicast 1194 * packet to the input queue of a specified interface. Note that this 1195 * calls the output routine of the loopback "driver", but with an interface 1196 * pointer that might NOT be a loopback interface -- evil, but easier than 1197 * replicating that code here. 1198 */ 1199 static void 1200 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1201 int hlen) 1202 { 1203 register struct ip *ip; 1204 struct mbuf *copym; 1205 1206 /* 1207 * Make a deep copy of the packet because we're going to 1208 * modify the pack in order to generate checksums. 1209 */ 1210 copym = m_dup(m, M_DONTWAIT); 1211 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1212 copym = m_pullup(copym, hlen); 1213 if (copym != NULL) { 1214 /* If needed, compute the checksum and mark it as valid. */ 1215 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1216 in_delayed_cksum(copym); 1217 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1218 copym->m_pkthdr.csum_flags |= 1219 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1220 copym->m_pkthdr.csum_data = 0xffff; 1221 } 1222 /* 1223 * We don't bother to fragment if the IP length is greater 1224 * than the interface's MTU. Can this possibly matter? 1225 */ 1226 ip = mtod(copym, struct ip *); 1227 ip->ip_len = htons(ip->ip_len); 1228 ip->ip_off = htons(ip->ip_off); 1229 ip->ip_sum = 0; 1230 ip->ip_sum = in_cksum(copym, hlen); 1231 #if 1 /* XXX */ 1232 if (dst->sin_family != AF_INET) { 1233 printf("ip_mloopback: bad address family %d\n", 1234 dst->sin_family); 1235 dst->sin_family = AF_INET; 1236 } 1237 #endif 1238 if_simloop(ifp, copym, dst->sin_family, 0); 1239 } 1240 } 1241