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_ipsec.h" 37 #include "opt_route.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 #include <net/flowtable.h> 62 #ifdef RADIX_MPATH 63 #include <net/radix_mpath.h> 64 #endif 65 #include <net/vnet.h> 66 67 #include <netinet/in.h> 68 #include <netinet/in_systm.h> 69 #include <netinet/ip.h> 70 #include <netinet/in_pcb.h> 71 #include <netinet/in_var.h> 72 #include <netinet/ip_var.h> 73 #include <netinet/ip_options.h> 74 #include <netinet/vinet.h> 75 #ifdef SCTP 76 #include <netinet/sctp.h> 77 #include <netinet/sctp_crc32.h> 78 #endif 79 80 #ifdef IPSEC 81 #include <netinet/ip_ipsec.h> 82 #include <netipsec/ipsec.h> 83 #endif /* IPSEC*/ 84 85 #include <machine/in_cksum.h> 86 87 #include <security/mac/mac_framework.h> 88 89 #define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 90 x, (ntohl(a.s_addr)>>24)&0xFF,\ 91 (ntohl(a.s_addr)>>16)&0xFF,\ 92 (ntohl(a.s_addr)>>8)&0xFF,\ 93 (ntohl(a.s_addr))&0xFF, y); 94 95 #ifdef VIMAGE_GLOBALS 96 u_short ip_id; 97 #endif 98 99 #ifdef MBUF_STRESS_TEST 100 int mbuf_frag_size = 0; 101 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 102 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 103 #endif 104 105 static void ip_mloopback 106 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 107 108 109 extern int in_mcast_loop; 110 extern struct protosw inetsw[]; 111 112 /* 113 * IP output. The packet in mbuf chain m contains a skeletal IP 114 * header (with len, off, ttl, proto, tos, src, dst). 115 * The mbuf chain containing the packet will be freed. 116 * The mbuf opt, if present, will not be freed. 117 * In the IP forwarding case, the packet will arrive with options already 118 * inserted, so must have a NULL opt pointer. 119 */ 120 int 121 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, 122 struct ip_moptions *imo, struct inpcb *inp) 123 { 124 INIT_VNET_NET(curvnet); 125 INIT_VNET_INET(curvnet); 126 struct ip *ip; 127 struct ifnet *ifp = NULL; /* keep compiler happy */ 128 struct mbuf *m0; 129 int hlen = sizeof (struct ip); 130 int mtu; 131 int len, error = 0; 132 int nortfree = 0; 133 struct sockaddr_in *dst = NULL; /* keep compiler happy */ 134 struct in_ifaddr *ia = NULL; 135 int isbroadcast, sw_csum; 136 struct route iproute; 137 struct in_addr odst; 138 #ifdef IPFIREWALL_FORWARD 139 struct m_tag *fwd_tag = NULL; 140 #endif 141 #ifdef IPSEC 142 int no_route_but_check_spd = 0; 143 #endif 144 M_ASSERTPKTHDR(m); 145 146 if (inp != NULL) { 147 INP_LOCK_ASSERT(inp); 148 M_SETFIB(m, inp->inp_inc.inc_fibnum); 149 if (inp->inp_flags & (INP_HW_FLOWID|INP_SW_FLOWID)) { 150 m->m_pkthdr.flowid = inp->inp_flowid; 151 m->m_flags |= M_FLOWID; 152 } 153 } 154 155 if (ro == NULL) { 156 ro = &iproute; 157 bzero(ro, sizeof (*ro)); 158 159 /* 160 * The flow table returns route entries valid for up to 30 161 * seconds; we rely on the remainder of ip_output() taking no 162 * longer than that long for the stability of ro_rt. The 163 * flow ID assignment must have happened before this point. 164 */ 165 if (flowtable_lookup(ip_ft, m, ro) == 0) 166 nortfree = 1; 167 } 168 169 if (opt) { 170 len = 0; 171 m = ip_insertoptions(m, opt, &len); 172 if (len != 0) 173 hlen = len; 174 } 175 ip = mtod(m, struct ip *); 176 177 /* 178 * Fill in IP header. If we are not allowing fragmentation, 179 * then the ip_id field is meaningless, but we don't set it 180 * to zero. Doing so causes various problems when devices along 181 * the path (routers, load balancers, firewalls, etc.) illegally 182 * disable DF on our packet. Note that a 16-bit counter 183 * will wrap around in less than 10 seconds at 100 Mbit/s on a 184 * medium with MTU 1500. See Steven M. Bellovin, "A Technique 185 * for Counting NATted Hosts", Proc. IMW'02, available at 186 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>. 187 */ 188 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 189 ip->ip_v = IPVERSION; 190 ip->ip_hl = hlen >> 2; 191 ip->ip_id = ip_newid(); 192 IPSTAT_INC(ips_localout); 193 } else { 194 hlen = ip->ip_hl << 2; 195 } 196 197 dst = (struct sockaddr_in *)&ro->ro_dst; 198 again: 199 /* 200 * If there is a cached route, 201 * check that it is to the same destination 202 * and is still up. If not, free it and try again. 203 * The address family should also be checked in case of sharing the 204 * cache with IPv6. 205 */ 206 if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 207 dst->sin_family != AF_INET || 208 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 209 if (!nortfree) 210 RTFREE(ro->ro_rt); 211 ro->ro_rt = (struct rtentry *)NULL; 212 } 213 #ifdef IPFIREWALL_FORWARD 214 if (ro->ro_rt == NULL && fwd_tag == NULL) { 215 #else 216 if (ro->ro_rt == NULL) { 217 #endif 218 bzero(dst, sizeof(*dst)); 219 dst->sin_family = AF_INET; 220 dst->sin_len = sizeof(*dst); 221 dst->sin_addr = ip->ip_dst; 222 } 223 /* 224 * If routing to interface only, short circuit routing lookup. 225 * The use of an all-ones broadcast address implies this; an 226 * interface is specified by the broadcast address of an interface, 227 * or the destination address of a ptp interface. 228 */ 229 if (flags & IP_SENDONES) { 230 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 231 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 232 IPSTAT_INC(ips_noroute); 233 error = ENETUNREACH; 234 goto bad; 235 } 236 ip->ip_dst.s_addr = INADDR_BROADCAST; 237 dst->sin_addr = ip->ip_dst; 238 ifp = ia->ia_ifp; 239 ip->ip_ttl = 1; 240 isbroadcast = 1; 241 } else if (flags & IP_ROUTETOIF) { 242 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 243 (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) { 244 IPSTAT_INC(ips_noroute); 245 error = ENETUNREACH; 246 goto bad; 247 } 248 ifp = ia->ia_ifp; 249 ip->ip_ttl = 1; 250 isbroadcast = in_broadcast(dst->sin_addr, ifp); 251 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 252 imo != NULL && imo->imo_multicast_ifp != NULL) { 253 /* 254 * Bypass the normal routing lookup for multicast 255 * packets if the interface is specified. 256 */ 257 ifp = imo->imo_multicast_ifp; 258 IFP_TO_IA(ifp, ia); 259 isbroadcast = 0; /* fool gcc */ 260 } else { 261 /* 262 * We want to do any cloning requested by the link layer, 263 * as this is probably required in all cases for correct 264 * operation (as it is for ARP). 265 */ 266 if (ro->ro_rt == NULL) 267 #ifdef RADIX_MPATH 268 rtalloc_mpath_fib(ro, 269 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 270 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 271 #else 272 in_rtalloc_ign(ro, 0, 273 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 274 #endif 275 if (ro->ro_rt == NULL) { 276 #ifdef IPSEC 277 /* 278 * There is no route for this packet, but it is 279 * possible that a matching SPD entry exists. 280 */ 281 no_route_but_check_spd = 1; 282 mtu = 0; /* Silence GCC warning. */ 283 goto sendit; 284 #endif 285 IPSTAT_INC(ips_noroute); 286 error = EHOSTUNREACH; 287 goto bad; 288 } 289 ia = ifatoia(ro->ro_rt->rt_ifa); 290 ifp = ro->ro_rt->rt_ifp; 291 ro->ro_rt->rt_rmx.rmx_pksent++; 292 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 293 dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 294 if (ro->ro_rt->rt_flags & RTF_HOST) 295 isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 296 else 297 isbroadcast = in_broadcast(dst->sin_addr, ifp); 298 } 299 /* 300 * Calculate MTU. If we have a route that is up, use that, 301 * otherwise use the interface's MTU. 302 */ 303 if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) { 304 /* 305 * This case can happen if the user changed the MTU 306 * of an interface after enabling IP on it. Because 307 * most netifs don't keep track of routes pointing to 308 * them, there is no way for one to update all its 309 * routes when the MTU is changed. 310 */ 311 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu) 312 ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 313 mtu = ro->ro_rt->rt_rmx.rmx_mtu; 314 } else { 315 mtu = ifp->if_mtu; 316 } 317 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 318 m->m_flags |= M_MCAST; 319 /* 320 * IP destination address is multicast. Make sure "dst" 321 * still points to the address in "ro". (It may have been 322 * changed to point to a gateway address, above.) 323 */ 324 dst = (struct sockaddr_in *)&ro->ro_dst; 325 /* 326 * See if the caller provided any multicast options 327 */ 328 if (imo != NULL) { 329 ip->ip_ttl = imo->imo_multicast_ttl; 330 if (imo->imo_multicast_vif != -1) 331 ip->ip_src.s_addr = 332 ip_mcast_src ? 333 ip_mcast_src(imo->imo_multicast_vif) : 334 INADDR_ANY; 335 } else 336 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 337 /* 338 * Confirm that the outgoing interface supports multicast. 339 */ 340 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 341 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 342 IPSTAT_INC(ips_noroute); 343 error = ENETUNREACH; 344 goto bad; 345 } 346 } 347 /* 348 * If source address not specified yet, use address 349 * of outgoing interface. 350 */ 351 if (ip->ip_src.s_addr == INADDR_ANY) { 352 /* Interface may have no addresses. */ 353 if (ia != NULL) 354 ip->ip_src = IA_SIN(ia)->sin_addr; 355 } 356 357 if ((imo == NULL && in_mcast_loop) || 358 (imo && imo->imo_multicast_loop)) { 359 /* 360 * Loop back multicast datagram if not expressly 361 * forbidden to do so, even if we are not a member 362 * of the group; ip_input() will filter it later, 363 * thus deferring a hash lookup and mutex acquisition 364 * at the expense of a cheap copy using m_copym(). 365 */ 366 ip_mloopback(ifp, m, dst, hlen); 367 } else { 368 /* 369 * If we are acting as a multicast router, perform 370 * multicast forwarding as if the packet had just 371 * arrived on the interface to which we are about 372 * to send. The multicast forwarding function 373 * recursively calls this function, using the 374 * IP_FORWARDING flag to prevent infinite recursion. 375 * 376 * Multicasts that are looped back by ip_mloopback(), 377 * above, will be forwarded by the ip_input() routine, 378 * if necessary. 379 */ 380 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 381 /* 382 * If rsvp daemon is not running, do not 383 * set ip_moptions. This ensures that the packet 384 * is multicast and not just sent down one link 385 * as prescribed by rsvpd. 386 */ 387 if (!V_rsvp_on) 388 imo = NULL; 389 if (ip_mforward && 390 ip_mforward(ip, ifp, m, imo) != 0) { 391 m_freem(m); 392 goto done; 393 } 394 } 395 } 396 397 /* 398 * Multicasts with a time-to-live of zero may be looped- 399 * back, above, but must not be transmitted on a network. 400 * Also, multicasts addressed to the loopback interface 401 * are not sent -- the above call to ip_mloopback() will 402 * loop back a copy. ip_input() will drop the copy if 403 * this host does not belong to the destination group on 404 * the loopback interface. 405 */ 406 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 407 m_freem(m); 408 goto done; 409 } 410 411 goto sendit; 412 } 413 414 /* 415 * If the source address is not specified yet, use the address 416 * of the outoing interface. 417 */ 418 if (ip->ip_src.s_addr == INADDR_ANY) { 419 /* Interface may have no addresses. */ 420 if (ia != NULL) { 421 ip->ip_src = IA_SIN(ia)->sin_addr; 422 } 423 } 424 425 /* 426 * Verify that we have any chance at all of being able to queue the 427 * packet or packet fragments, unless ALTQ is enabled on the given 428 * interface in which case packetdrop should be done by queueing. 429 */ 430 #ifdef ALTQ 431 if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) && 432 ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 433 ifp->if_snd.ifq_maxlen)) 434 #else 435 if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 436 ifp->if_snd.ifq_maxlen) 437 #endif /* ALTQ */ 438 { 439 error = ENOBUFS; 440 IPSTAT_INC(ips_odropped); 441 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1); 442 goto bad; 443 } 444 445 /* 446 * Look for broadcast address and 447 * verify user is allowed to send 448 * such a packet. 449 */ 450 if (isbroadcast) { 451 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 452 error = EADDRNOTAVAIL; 453 goto bad; 454 } 455 if ((flags & IP_ALLOWBROADCAST) == 0) { 456 error = EACCES; 457 goto bad; 458 } 459 /* don't allow broadcast messages to be fragmented */ 460 if (ip->ip_len > mtu) { 461 error = EMSGSIZE; 462 goto bad; 463 } 464 m->m_flags |= M_BCAST; 465 } else { 466 m->m_flags &= ~M_BCAST; 467 } 468 469 sendit: 470 #ifdef IPSEC 471 switch(ip_ipsec_output(&m, inp, &flags, &error, &ro, &iproute, &dst, &ia, &ifp)) { 472 case 1: 473 goto bad; 474 case -1: 475 goto done; 476 case 0: 477 default: 478 break; /* Continue with packet processing. */ 479 } 480 /* 481 * Check if there was a route for this packet; return error if not. 482 */ 483 if (no_route_but_check_spd) { 484 IPSTAT_INC(ips_noroute); 485 error = EHOSTUNREACH; 486 goto bad; 487 } 488 /* Update variables that are affected by ipsec4_output(). */ 489 ip = mtod(m, struct ip *); 490 hlen = ip->ip_hl << 2; 491 #endif /* IPSEC */ 492 493 /* Jump over all PFIL processing if hooks are not active. */ 494 if (!PFIL_HOOKED(&inet_pfil_hook)) 495 goto passout; 496 497 /* Run through list of hooks for output packets. */ 498 odst.s_addr = ip->ip_dst.s_addr; 499 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 500 if (error != 0 || m == NULL) 501 goto done; 502 503 ip = mtod(m, struct ip *); 504 505 /* See if destination IP address was changed by packet filter. */ 506 if (odst.s_addr != ip->ip_dst.s_addr) { 507 m->m_flags |= M_SKIP_FIREWALL; 508 /* If destination is now ourself drop to ip_input(). */ 509 if (in_localip(ip->ip_dst)) { 510 m->m_flags |= M_FASTFWD_OURS; 511 if (m->m_pkthdr.rcvif == NULL) 512 m->m_pkthdr.rcvif = V_loif; 513 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 514 m->m_pkthdr.csum_flags |= 515 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 516 m->m_pkthdr.csum_data = 0xffff; 517 } 518 m->m_pkthdr.csum_flags |= 519 CSUM_IP_CHECKED | CSUM_IP_VALID; 520 #ifdef SCTP 521 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 522 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 523 #endif 524 error = netisr_queue(NETISR_IP, m); 525 goto done; 526 } else 527 goto again; /* Redo the routing table lookup. */ 528 } 529 530 #ifdef IPFIREWALL_FORWARD 531 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 532 if (m->m_flags & M_FASTFWD_OURS) { 533 if (m->m_pkthdr.rcvif == NULL) 534 m->m_pkthdr.rcvif = V_loif; 535 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 536 m->m_pkthdr.csum_flags |= 537 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 538 m->m_pkthdr.csum_data = 0xffff; 539 } 540 #ifdef SCTP 541 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 542 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 543 #endif 544 m->m_pkthdr.csum_flags |= 545 CSUM_IP_CHECKED | CSUM_IP_VALID; 546 547 error = netisr_queue(NETISR_IP, m); 548 goto done; 549 } 550 /* Or forward to some other address? */ 551 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 552 if (fwd_tag) { 553 dst = (struct sockaddr_in *)&ro->ro_dst; 554 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 555 m->m_flags |= M_SKIP_FIREWALL; 556 m_tag_delete(m, fwd_tag); 557 goto again; 558 } 559 #endif /* IPFIREWALL_FORWARD */ 560 561 passout: 562 /* 127/8 must not appear on wire - RFC1122. */ 563 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 564 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 565 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 566 IPSTAT_INC(ips_badaddr); 567 error = EADDRNOTAVAIL; 568 goto bad; 569 } 570 } 571 572 m->m_pkthdr.csum_flags |= CSUM_IP; 573 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 574 if (sw_csum & CSUM_DELAY_DATA) { 575 in_delayed_cksum(m); 576 sw_csum &= ~CSUM_DELAY_DATA; 577 } 578 #ifdef SCTP 579 if (sw_csum & CSUM_SCTP) { 580 sctp_delayed_cksum(m); 581 sw_csum &= ~CSUM_SCTP; 582 } 583 #endif 584 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 585 586 /* 587 * If small enough for interface, or the interface will take 588 * care of the fragmentation for us, we can just send directly. 589 */ 590 if (ip->ip_len <= mtu || 591 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 592 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 593 ip->ip_len = htons(ip->ip_len); 594 ip->ip_off = htons(ip->ip_off); 595 ip->ip_sum = 0; 596 if (sw_csum & CSUM_DELAY_IP) 597 ip->ip_sum = in_cksum(m, hlen); 598 599 /* 600 * Record statistics for this interface address. 601 * With CSUM_TSO the byte/packet count will be slightly 602 * incorrect because we count the IP+TCP headers only 603 * once instead of for every generated packet. 604 */ 605 if (!(flags & IP_FORWARDING) && ia) { 606 if (m->m_pkthdr.csum_flags & CSUM_TSO) 607 ia->ia_ifa.if_opackets += 608 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 609 else 610 ia->ia_ifa.if_opackets++; 611 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 612 } 613 #ifdef MBUF_STRESS_TEST 614 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 615 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 616 #endif 617 /* 618 * Reset layer specific mbuf flags 619 * to avoid confusing lower layers. 620 */ 621 m->m_flags &= ~(M_PROTOFLAGS); 622 error = (*ifp->if_output)(ifp, m, 623 (struct sockaddr *)dst, ro); 624 goto done; 625 } 626 627 /* Balk when DF bit is set or the interface didn't support TSO. */ 628 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 629 error = EMSGSIZE; 630 IPSTAT_INC(ips_cantfrag); 631 goto bad; 632 } 633 634 /* 635 * Too large for interface; fragment if possible. If successful, 636 * on return, m will point to a list of packets to be sent. 637 */ 638 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 639 if (error) 640 goto bad; 641 for (; m; m = m0) { 642 m0 = m->m_nextpkt; 643 m->m_nextpkt = 0; 644 if (error == 0) { 645 /* Record statistics for this interface address. */ 646 if (ia != NULL) { 647 ia->ia_ifa.if_opackets++; 648 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 649 } 650 /* 651 * Reset layer specific mbuf flags 652 * to avoid confusing upper layers. 653 */ 654 m->m_flags &= ~(M_PROTOFLAGS); 655 656 error = (*ifp->if_output)(ifp, m, 657 (struct sockaddr *)dst, ro); 658 } else 659 m_freem(m); 660 } 661 662 if (error == 0) 663 IPSTAT_INC(ips_fragmented); 664 665 done: 666 if (ro == &iproute && ro->ro_rt && !nortfree) { 667 RTFREE(ro->ro_rt); 668 } 669 return (error); 670 bad: 671 m_freem(m); 672 goto done; 673 } 674 675 /* 676 * Create a chain of fragments which fit the given mtu. m_frag points to the 677 * mbuf to be fragmented; on return it points to the chain with the fragments. 678 * Return 0 if no error. If error, m_frag may contain a partially built 679 * chain of fragments that should be freed by the caller. 680 * 681 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 682 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 683 */ 684 int 685 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 686 u_long if_hwassist_flags, int sw_csum) 687 { 688 INIT_VNET_INET(curvnet); 689 int error = 0; 690 int hlen = ip->ip_hl << 2; 691 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 692 int off; 693 struct mbuf *m0 = *m_frag; /* the original packet */ 694 int firstlen; 695 struct mbuf **mnext; 696 int nfrags; 697 698 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 699 IPSTAT_INC(ips_cantfrag); 700 return EMSGSIZE; 701 } 702 703 /* 704 * Must be able to put at least 8 bytes per fragment. 705 */ 706 if (len < 8) 707 return EMSGSIZE; 708 709 /* 710 * If the interface will not calculate checksums on 711 * fragmented packets, then do it here. 712 */ 713 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 714 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 715 in_delayed_cksum(m0); 716 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 717 } 718 #ifdef SCTP 719 if (m0->m_pkthdr.csum_flags & CSUM_SCTP && 720 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 721 sctp_delayed_cksum(m0); 722 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 723 } 724 #endif 725 if (len > PAGE_SIZE) { 726 /* 727 * Fragment large datagrams such that each segment 728 * contains a multiple of PAGE_SIZE amount of data, 729 * plus headers. This enables a receiver to perform 730 * page-flipping zero-copy optimizations. 731 * 732 * XXX When does this help given that sender and receiver 733 * could have different page sizes, and also mtu could 734 * be less than the receiver's page size ? 735 */ 736 int newlen; 737 struct mbuf *m; 738 739 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 740 off += m->m_len; 741 742 /* 743 * firstlen (off - hlen) must be aligned on an 744 * 8-byte boundary 745 */ 746 if (off < hlen) 747 goto smart_frag_failure; 748 off = ((off - hlen) & ~7) + hlen; 749 newlen = (~PAGE_MASK) & mtu; 750 if ((newlen + sizeof (struct ip)) > mtu) { 751 /* we failed, go back the default */ 752 smart_frag_failure: 753 newlen = len; 754 off = hlen + len; 755 } 756 len = newlen; 757 758 } else { 759 off = hlen + len; 760 } 761 762 firstlen = off - hlen; 763 mnext = &m0->m_nextpkt; /* pointer to next packet */ 764 765 /* 766 * Loop through length of segment after first fragment, 767 * make new header and copy data of each part and link onto chain. 768 * Here, m0 is the original packet, m is the fragment being created. 769 * The fragments are linked off the m_nextpkt of the original 770 * packet, which after processing serves as the first fragment. 771 */ 772 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 773 struct ip *mhip; /* ip header on the fragment */ 774 struct mbuf *m; 775 int mhlen = sizeof (struct ip); 776 777 MGETHDR(m, M_DONTWAIT, MT_DATA); 778 if (m == NULL) { 779 error = ENOBUFS; 780 IPSTAT_INC(ips_odropped); 781 goto done; 782 } 783 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 784 /* 785 * In the first mbuf, leave room for the link header, then 786 * copy the original IP header including options. The payload 787 * goes into an additional mbuf chain returned by m_copym(). 788 */ 789 m->m_data += max_linkhdr; 790 mhip = mtod(m, struct ip *); 791 *mhip = *ip; 792 if (hlen > sizeof (struct ip)) { 793 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 794 mhip->ip_v = IPVERSION; 795 mhip->ip_hl = mhlen >> 2; 796 } 797 m->m_len = mhlen; 798 /* XXX do we need to add ip->ip_off below ? */ 799 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 800 if (off + len >= ip->ip_len) { /* last fragment */ 801 len = ip->ip_len - off; 802 m->m_flags |= M_LASTFRAG; 803 } else 804 mhip->ip_off |= IP_MF; 805 mhip->ip_len = htons((u_short)(len + mhlen)); 806 m->m_next = m_copym(m0, off, len, M_DONTWAIT); 807 if (m->m_next == NULL) { /* copy failed */ 808 m_free(m); 809 error = ENOBUFS; /* ??? */ 810 IPSTAT_INC(ips_odropped); 811 goto done; 812 } 813 m->m_pkthdr.len = mhlen + len; 814 m->m_pkthdr.rcvif = NULL; 815 #ifdef MAC 816 mac_netinet_fragment(m0, m); 817 #endif 818 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 819 mhip->ip_off = htons(mhip->ip_off); 820 mhip->ip_sum = 0; 821 if (sw_csum & CSUM_DELAY_IP) 822 mhip->ip_sum = in_cksum(m, mhlen); 823 *mnext = m; 824 mnext = &m->m_nextpkt; 825 } 826 IPSTAT_ADD(ips_ofragments, nfrags); 827 828 /* set first marker for fragment chain */ 829 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 830 m0->m_pkthdr.csum_data = nfrags; 831 832 /* 833 * Update first fragment by trimming what's been copied out 834 * and updating header. 835 */ 836 m_adj(m0, hlen + firstlen - ip->ip_len); 837 m0->m_pkthdr.len = hlen + firstlen; 838 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 839 ip->ip_off |= IP_MF; 840 ip->ip_off = htons(ip->ip_off); 841 ip->ip_sum = 0; 842 if (sw_csum & CSUM_DELAY_IP) 843 ip->ip_sum = in_cksum(m0, hlen); 844 845 done: 846 *m_frag = m0; 847 return error; 848 } 849 850 void 851 in_delayed_cksum(struct mbuf *m) 852 { 853 struct ip *ip; 854 u_short csum, offset; 855 856 ip = mtod(m, struct ip *); 857 offset = ip->ip_hl << 2 ; 858 csum = in_cksum_skip(m, ip->ip_len, offset); 859 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 860 csum = 0xffff; 861 offset += m->m_pkthdr.csum_data; /* checksum offset */ 862 863 if (offset + sizeof(u_short) > m->m_len) { 864 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 865 m->m_len, offset, ip->ip_p); 866 /* 867 * XXX 868 * this shouldn't happen, but if it does, the 869 * correct behavior may be to insert the checksum 870 * in the appropriate next mbuf in the chain. 871 */ 872 return; 873 } 874 *(u_short *)(m->m_data + offset) = csum; 875 } 876 877 /* 878 * IP socket option processing. 879 */ 880 int 881 ip_ctloutput(struct socket *so, struct sockopt *sopt) 882 { 883 struct inpcb *inp = sotoinpcb(so); 884 int error, optval; 885 886 error = optval = 0; 887 if (sopt->sopt_level != IPPROTO_IP) { 888 if ((sopt->sopt_level == SOL_SOCKET) && 889 (sopt->sopt_name == SO_SETFIB)) { 890 inp->inp_inc.inc_fibnum = so->so_fibnum; 891 return (0); 892 } 893 return (EINVAL); 894 } 895 896 switch (sopt->sopt_dir) { 897 case SOPT_SET: 898 switch (sopt->sopt_name) { 899 case IP_OPTIONS: 900 #ifdef notyet 901 case IP_RETOPTS: 902 #endif 903 { 904 struct mbuf *m; 905 if (sopt->sopt_valsize > MLEN) { 906 error = EMSGSIZE; 907 break; 908 } 909 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 910 if (m == NULL) { 911 error = ENOBUFS; 912 break; 913 } 914 m->m_len = sopt->sopt_valsize; 915 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 916 m->m_len); 917 if (error) { 918 m_free(m); 919 break; 920 } 921 INP_WLOCK(inp); 922 error = ip_pcbopts(inp, sopt->sopt_name, m); 923 INP_WUNLOCK(inp); 924 return (error); 925 } 926 927 case IP_BINDANY: 928 if (sopt->sopt_td != NULL) { 929 error = priv_check(sopt->sopt_td, 930 PRIV_NETINET_BINDANY); 931 if (error) 932 break; 933 } 934 /* FALLTHROUGH */ 935 case IP_TOS: 936 case IP_TTL: 937 case IP_MINTTL: 938 case IP_RECVOPTS: 939 case IP_RECVRETOPTS: 940 case IP_RECVDSTADDR: 941 case IP_RECVTTL: 942 case IP_RECVIF: 943 case IP_FAITH: 944 case IP_ONESBCAST: 945 case IP_DONTFRAG: 946 error = sooptcopyin(sopt, &optval, sizeof optval, 947 sizeof optval); 948 if (error) 949 break; 950 951 switch (sopt->sopt_name) { 952 case IP_TOS: 953 inp->inp_ip_tos = optval; 954 break; 955 956 case IP_TTL: 957 inp->inp_ip_ttl = optval; 958 break; 959 960 case IP_MINTTL: 961 if (optval >= 0 && optval <= MAXTTL) 962 inp->inp_ip_minttl = optval; 963 else 964 error = EINVAL; 965 break; 966 967 #define OPTSET(bit) do { \ 968 INP_WLOCK(inp); \ 969 if (optval) \ 970 inp->inp_flags |= bit; \ 971 else \ 972 inp->inp_flags &= ~bit; \ 973 INP_WUNLOCK(inp); \ 974 } while (0) 975 976 case IP_RECVOPTS: 977 OPTSET(INP_RECVOPTS); 978 break; 979 980 case IP_RECVRETOPTS: 981 OPTSET(INP_RECVRETOPTS); 982 break; 983 984 case IP_RECVDSTADDR: 985 OPTSET(INP_RECVDSTADDR); 986 break; 987 988 case IP_RECVTTL: 989 OPTSET(INP_RECVTTL); 990 break; 991 992 case IP_RECVIF: 993 OPTSET(INP_RECVIF); 994 break; 995 996 case IP_FAITH: 997 OPTSET(INP_FAITH); 998 break; 999 1000 case IP_ONESBCAST: 1001 OPTSET(INP_ONESBCAST); 1002 break; 1003 case IP_DONTFRAG: 1004 OPTSET(INP_DONTFRAG); 1005 break; 1006 case IP_BINDANY: 1007 OPTSET(INP_BINDANY); 1008 break; 1009 } 1010 break; 1011 #undef OPTSET 1012 1013 /* 1014 * Multicast socket options are processed by the in_mcast 1015 * module. 1016 */ 1017 case IP_MULTICAST_IF: 1018 case IP_MULTICAST_VIF: 1019 case IP_MULTICAST_TTL: 1020 case IP_MULTICAST_LOOP: 1021 case IP_ADD_MEMBERSHIP: 1022 case IP_DROP_MEMBERSHIP: 1023 case IP_ADD_SOURCE_MEMBERSHIP: 1024 case IP_DROP_SOURCE_MEMBERSHIP: 1025 case IP_BLOCK_SOURCE: 1026 case IP_UNBLOCK_SOURCE: 1027 case IP_MSFILTER: 1028 case MCAST_JOIN_GROUP: 1029 case MCAST_LEAVE_GROUP: 1030 case MCAST_JOIN_SOURCE_GROUP: 1031 case MCAST_LEAVE_SOURCE_GROUP: 1032 case MCAST_BLOCK_SOURCE: 1033 case MCAST_UNBLOCK_SOURCE: 1034 error = inp_setmoptions(inp, sopt); 1035 break; 1036 1037 case IP_PORTRANGE: 1038 error = sooptcopyin(sopt, &optval, sizeof optval, 1039 sizeof optval); 1040 if (error) 1041 break; 1042 1043 INP_WLOCK(inp); 1044 switch (optval) { 1045 case IP_PORTRANGE_DEFAULT: 1046 inp->inp_flags &= ~(INP_LOWPORT); 1047 inp->inp_flags &= ~(INP_HIGHPORT); 1048 break; 1049 1050 case IP_PORTRANGE_HIGH: 1051 inp->inp_flags &= ~(INP_LOWPORT); 1052 inp->inp_flags |= INP_HIGHPORT; 1053 break; 1054 1055 case IP_PORTRANGE_LOW: 1056 inp->inp_flags &= ~(INP_HIGHPORT); 1057 inp->inp_flags |= INP_LOWPORT; 1058 break; 1059 1060 default: 1061 error = EINVAL; 1062 break; 1063 } 1064 INP_WUNLOCK(inp); 1065 break; 1066 1067 #ifdef IPSEC 1068 case IP_IPSEC_POLICY: 1069 { 1070 caddr_t req; 1071 struct mbuf *m; 1072 1073 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1074 break; 1075 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1076 break; 1077 req = mtod(m, caddr_t); 1078 error = ipsec_set_policy(inp, sopt->sopt_name, req, 1079 m->m_len, (sopt->sopt_td != NULL) ? 1080 sopt->sopt_td->td_ucred : NULL); 1081 m_freem(m); 1082 break; 1083 } 1084 #endif /* IPSEC */ 1085 1086 default: 1087 error = ENOPROTOOPT; 1088 break; 1089 } 1090 break; 1091 1092 case SOPT_GET: 1093 switch (sopt->sopt_name) { 1094 case IP_OPTIONS: 1095 case IP_RETOPTS: 1096 if (inp->inp_options) 1097 error = sooptcopyout(sopt, 1098 mtod(inp->inp_options, 1099 char *), 1100 inp->inp_options->m_len); 1101 else 1102 sopt->sopt_valsize = 0; 1103 break; 1104 1105 case IP_TOS: 1106 case IP_TTL: 1107 case IP_MINTTL: 1108 case IP_RECVOPTS: 1109 case IP_RECVRETOPTS: 1110 case IP_RECVDSTADDR: 1111 case IP_RECVTTL: 1112 case IP_RECVIF: 1113 case IP_PORTRANGE: 1114 case IP_FAITH: 1115 case IP_ONESBCAST: 1116 case IP_DONTFRAG: 1117 switch (sopt->sopt_name) { 1118 1119 case IP_TOS: 1120 optval = inp->inp_ip_tos; 1121 break; 1122 1123 case IP_TTL: 1124 optval = inp->inp_ip_ttl; 1125 break; 1126 1127 case IP_MINTTL: 1128 optval = inp->inp_ip_minttl; 1129 break; 1130 1131 #define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1132 1133 case IP_RECVOPTS: 1134 optval = OPTBIT(INP_RECVOPTS); 1135 break; 1136 1137 case IP_RECVRETOPTS: 1138 optval = OPTBIT(INP_RECVRETOPTS); 1139 break; 1140 1141 case IP_RECVDSTADDR: 1142 optval = OPTBIT(INP_RECVDSTADDR); 1143 break; 1144 1145 case IP_RECVTTL: 1146 optval = OPTBIT(INP_RECVTTL); 1147 break; 1148 1149 case IP_RECVIF: 1150 optval = OPTBIT(INP_RECVIF); 1151 break; 1152 1153 case IP_PORTRANGE: 1154 if (inp->inp_flags & INP_HIGHPORT) 1155 optval = IP_PORTRANGE_HIGH; 1156 else if (inp->inp_flags & INP_LOWPORT) 1157 optval = IP_PORTRANGE_LOW; 1158 else 1159 optval = 0; 1160 break; 1161 1162 case IP_FAITH: 1163 optval = OPTBIT(INP_FAITH); 1164 break; 1165 1166 case IP_ONESBCAST: 1167 optval = OPTBIT(INP_ONESBCAST); 1168 break; 1169 case IP_DONTFRAG: 1170 optval = OPTBIT(INP_DONTFRAG); 1171 break; 1172 } 1173 error = sooptcopyout(sopt, &optval, sizeof optval); 1174 break; 1175 1176 /* 1177 * Multicast socket options are processed by the in_mcast 1178 * module. 1179 */ 1180 case IP_MULTICAST_IF: 1181 case IP_MULTICAST_VIF: 1182 case IP_MULTICAST_TTL: 1183 case IP_MULTICAST_LOOP: 1184 case IP_MSFILTER: 1185 error = inp_getmoptions(inp, sopt); 1186 break; 1187 1188 #ifdef IPSEC 1189 case IP_IPSEC_POLICY: 1190 { 1191 struct mbuf *m = NULL; 1192 caddr_t req = NULL; 1193 size_t len = 0; 1194 1195 if (m != 0) { 1196 req = mtod(m, caddr_t); 1197 len = m->m_len; 1198 } 1199 error = ipsec_get_policy(sotoinpcb(so), req, len, &m); 1200 if (error == 0) 1201 error = soopt_mcopyout(sopt, m); /* XXX */ 1202 if (error == 0) 1203 m_freem(m); 1204 break; 1205 } 1206 #endif /* IPSEC */ 1207 1208 default: 1209 error = ENOPROTOOPT; 1210 break; 1211 } 1212 break; 1213 } 1214 return (error); 1215 } 1216 1217 /* 1218 * Routine called from ip_output() to loop back a copy of an IP multicast 1219 * packet to the input queue of a specified interface. Note that this 1220 * calls the output routine of the loopback "driver", but with an interface 1221 * pointer that might NOT be a loopback interface -- evil, but easier than 1222 * replicating that code here. 1223 */ 1224 static void 1225 ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1226 int hlen) 1227 { 1228 register struct ip *ip; 1229 struct mbuf *copym; 1230 1231 /* 1232 * Make a deep copy of the packet because we're going to 1233 * modify the pack in order to generate checksums. 1234 */ 1235 copym = m_dup(m, M_DONTWAIT); 1236 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1237 copym = m_pullup(copym, hlen); 1238 if (copym != NULL) { 1239 /* If needed, compute the checksum and mark it as valid. */ 1240 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1241 in_delayed_cksum(copym); 1242 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1243 copym->m_pkthdr.csum_flags |= 1244 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1245 copym->m_pkthdr.csum_data = 0xffff; 1246 } 1247 /* 1248 * We don't bother to fragment if the IP length is greater 1249 * than the interface's MTU. Can this possibly matter? 1250 */ 1251 ip = mtod(copym, struct ip *); 1252 ip->ip_len = htons(ip->ip_len); 1253 ip->ip_off = htons(ip->ip_off); 1254 ip->ip_sum = 0; 1255 ip->ip_sum = in_cksum(copym, hlen); 1256 #if 1 /* XXX */ 1257 if (dst->sin_family != AF_INET) { 1258 printf("ip_mloopback: bad address family %d\n", 1259 dst->sin_family); 1260 dst->sin_family = AF_INET; 1261 } 1262 #endif 1263 if_simloop(ifp, copym, dst->sin_family, 0); 1264 } 1265 } 1266