ip_output.c revision 128003
1292920Sdim/* 2292920Sdim * Copyright (c) 1982, 1986, 1988, 1990, 1993 3292920Sdim * The Regents of the University of California. All rights reserved. 4292920Sdim * 5292920Sdim * Redistribution and use in source and binary forms, with or without 6292920Sdim * modification, are permitted provided that the following conditions 7292920Sdim * are met: 8292920Sdim * 1. Redistributions of source code must retain the above copyright 9292920Sdim * notice, this list of conditions and the following disclaimer. 10292920Sdim * 2. Redistributions in binary form must reproduce the above copyright 11292920Sdim * notice, this list of conditions and the following disclaimer in the 12292920Sdim * documentation and/or other materials provided with the distribution. 13292920Sdim * 3. All advertising materials mentioning features or use of this software 14292920Sdim * must display the following acknowledgement: 15292920Sdim * This product includes software developed by the University of 16292920Sdim * California, Berkeley and its contributors. 17292920Sdim * 4. Neither the name of the University nor the names of its contributors 18292920Sdim * may be used to endorse or promote products derived from this software 19292920Sdim * without specific prior written permission. 20292920Sdim * 21292920Sdim * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22292920Sdim * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23292920Sdim * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24292920Sdim * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25292920Sdim * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26292920Sdim * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27292920Sdim * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28292920Sdim * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29292920Sdim * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30292920Sdim * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31292920Sdim * SUCH DAMAGE. 32292920Sdim * 33292920Sdim * @(#)ip_output.c 8.3 (Berkeley) 1/21/94 34292920Sdim * $FreeBSD: head/sys/netinet/ip_output.c 128003 2004-04-07 10:01:39Z ru $ 35292920Sdim */ 36292920Sdim 37292920Sdim#include "opt_ipfw.h" 38292920Sdim#include "opt_ipdn.h" 39292920Sdim#include "opt_ipdivert.h" 40292920Sdim#include "opt_ipfilter.h" 41292920Sdim#include "opt_ipsec.h" 42292920Sdim#include "opt_mac.h" 43292920Sdim#include "opt_pfil_hooks.h" 44292920Sdim#include "opt_random_ip_id.h" 45292920Sdim#include "opt_mbuf_stress_test.h" 46292920Sdim 47292920Sdim#include <sys/param.h> 48292920Sdim#include <sys/systm.h> 49292920Sdim#include <sys/kernel.h> 50292920Sdim#include <sys/mac.h> 51292920Sdim#include <sys/malloc.h> 52292920Sdim#include <sys/mbuf.h> 53292920Sdim#include <sys/protosw.h> 54292920Sdim#include <sys/socket.h> 55292920Sdim#include <sys/socketvar.h> 56292920Sdim#include <sys/sysctl.h> 57292920Sdim 58292920Sdim#include <net/if.h> 59292920Sdim#include <net/route.h> 60292920Sdim 61292920Sdim#include <netinet/in.h> 62292920Sdim#include <netinet/in_systm.h> 63292920Sdim#include <netinet/ip.h> 64292920Sdim#include <netinet/in_pcb.h> 65292920Sdim#include <netinet/in_var.h> 66292920Sdim#include <netinet/ip_var.h> 67292920Sdim 68292920Sdim#ifdef PFIL_HOOKS 69292920Sdim#include <net/pfil.h> 70292920Sdim#endif 71292920Sdim 72292920Sdim#include <machine/in_cksum.h> 73292920Sdim 74292920Sdimstatic MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "internet multicast options"); 75292920Sdim 76292920Sdim#ifdef IPSEC 77292920Sdim#include <netinet6/ipsec.h> 78292920Sdim#include <netkey/key.h> 79292920Sdim#ifdef IPSEC_DEBUG 80292920Sdim#include <netkey/key_debug.h> 81292920Sdim#else 82292920Sdim#define KEYDEBUG(lev,arg) 83292920Sdim#endif 84292920Sdim#endif /*IPSEC*/ 85292920Sdim 86292920Sdim#ifdef FAST_IPSEC 87292920Sdim#include <netipsec/ipsec.h> 88292920Sdim#include <netipsec/xform.h> 89292920Sdim#include <netipsec/key.h> 90292920Sdim#endif /*FAST_IPSEC*/ 91292920Sdim 92292920Sdim#include <netinet/ip_fw.h> 93292920Sdim#include <netinet/ip_divert.h> 94292920Sdim#include <netinet/ip_dummynet.h> 95292920Sdim 96292920Sdim#define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 97292920Sdim x, (ntohl(a.s_addr)>>24)&0xFF,\ 98292920Sdim (ntohl(a.s_addr)>>16)&0xFF,\ 99292920Sdim (ntohl(a.s_addr)>>8)&0xFF,\ 100292920Sdim (ntohl(a.s_addr))&0xFF, y); 101292920Sdim 102292920Sdimu_short ip_id; 103292920Sdim 104292920Sdim#ifdef MBUF_STRESS_TEST 105292920Sdimint mbuf_frag_size = 0; 106292920SdimSYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 107292920Sdim &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 108292920Sdim#endif 109292920Sdim 110292920Sdimstatic struct mbuf *ip_insertoptions(struct mbuf *, struct mbuf *, int *); 111292920Sdimstatic struct ifnet *ip_multicast_if(struct in_addr *, int *); 112292920Sdimstatic void ip_mloopback 113292920Sdim (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 114292920Sdimstatic int ip_getmoptions 115292920Sdim (struct sockopt *, struct ip_moptions *); 116292920Sdimstatic int ip_pcbopts(int, struct mbuf **, struct mbuf *); 117292920Sdimstatic int ip_setmoptions 118292920Sdim (struct sockopt *, struct ip_moptions **); 119292920Sdim 120292920Sdimint ip_optcopy(struct ip *, struct ip *); 121292920Sdim 122292920Sdim 123292920Sdimextern struct protosw inetsw[]; 124292920Sdim 125292920Sdim/* 126292920Sdim * IP output. The packet in mbuf chain m contains a skeletal IP 127292920Sdim * header (with len, off, ttl, proto, tos, src, dst). 128292920Sdim * The mbuf chain containing the packet will be freed. 129292920Sdim * The mbuf opt, if present, will not be freed. 130292920Sdim * In the IP forwarding case, the packet will arrive with options already 131292920Sdim * inserted, so must have a NULL opt pointer. 132292920Sdim */ 133292920Sdimint 134292920Sdimip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, 135292920Sdim int flags, struct ip_moptions *imo, struct inpcb *inp) 136292920Sdim{ 137292920Sdim struct ip *ip; 138292920Sdim struct ifnet *ifp = NULL; /* keep compiler happy */ 139292920Sdim struct mbuf *m0; 140292920Sdim int hlen = sizeof (struct ip); 141292920Sdim int len, off, error = 0; 142292920Sdim struct sockaddr_in *dst = NULL; /* keep compiler happy */ 143292920Sdim struct in_ifaddr *ia = NULL; 144292920Sdim int isbroadcast, sw_csum; 145292920Sdim struct in_addr pkt_dst; 146292920Sdim struct route iproute; 147292920Sdim struct m_tag *mtag, *dummytag; 148292920Sdim#ifdef IPSEC 149292920Sdim struct secpolicy *sp = NULL; 150292920Sdim#endif 151292920Sdim#ifdef FAST_IPSEC 152292920Sdim struct secpolicy *sp = NULL; 153292920Sdim struct tdb_ident *tdbi; 154292920Sdim int s; 155292920Sdim#endif /* FAST_IPSEC */ 156292920Sdim struct ip_fw_args args; 157292920Sdim int src_was_INADDR_ANY = 0; /* as the name says... */ 158292920Sdim 159292920Sdim args.eh = NULL; 160292920Sdim args.rule = NULL; 161292920Sdim 162292920Sdim M_ASSERTPKTHDR(m); 163292920Sdim 164292920Sdim args.next_hop = ip_claim_next_hop(m); 165292920Sdim dummytag = m_tag_find(m, PACKET_TAG_DUMMYNET, NULL); 166292920Sdim if (dummytag != NULL) { 167292920Sdim struct dn_pkt_tag *dt = (struct dn_pkt_tag *)(dummytag+1); 168292920Sdim /* 169292920Sdim * Prevent lower layers from finding the tag 170292920Sdim * Cleanup and free is done below 171292920Sdim */ 172292920Sdim m_tag_unlink(m, dummytag); 173292920Sdim /* 174292920Sdim * the packet was already tagged, so part of the 175292920Sdim * processing was already done, and we need to go down. 176292920Sdim * Get parameters from the header. 177292920Sdim */ 178292920Sdim args.rule = dt->rule; 179292920Sdim ro = &(dt->ro); 180292920Sdim dst = dt->dn_dst; 181292920Sdim ifp = dt->ifp; 182292920Sdim } 183292920Sdim 184292920Sdim if (ro == NULL) { 185292920Sdim ro = &iproute; 186292920Sdim bzero(ro, sizeof (*ro)); 187292920Sdim } 188292920Sdim 189292920Sdim if (inp != NULL) 190292920Sdim INP_LOCK_ASSERT(inp); 191292920Sdim 192292920Sdim if (args.rule != NULL) { /* dummynet already saw us */ 193292920Sdim ip = mtod(m, struct ip *); 194292920Sdim hlen = ip->ip_hl << 2 ; 195292920Sdim if (ro->ro_rt) 196292920Sdim ia = ifatoia(ro->ro_rt->rt_ifa); 197292920Sdim goto sendit; 198292920Sdim } 199292920Sdim 200292920Sdim if (opt) { 201292920Sdim len = 0; 202292920Sdim m = ip_insertoptions(m, opt, &len); 203292920Sdim if (len != 0) 204292920Sdim hlen = len; 205292920Sdim } 206292920Sdim ip = mtod(m, struct ip *); 207292920Sdim pkt_dst = args.next_hop ? args.next_hop->sin_addr : ip->ip_dst; 208292920Sdim 209292920Sdim /* 210292920Sdim * Fill in IP header. If we are not allowing fragmentation, 211292920Sdim * then the ip_id field is meaningless, but we don't set it 212292920Sdim * to zero. Doing so causes various problems when devices along 213292920Sdim * the path (routers, load balancers, firewalls, etc.) illegally 214292920Sdim * disable DF on our packet. Note that a 16-bit counter 215292920Sdim * will wrap around in less than 10 seconds at 100 Mbit/s on a 216292920Sdim * medium with MTU 1500. See Steven M. Bellovin, "A Technique 217292920Sdim * for Counting NATted Hosts", Proc. IMW'02, available at 218292920Sdim * <http://www.research.att.com/~smb/papers/fnat.pdf>. 219292920Sdim */ 220292920Sdim if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 221292920Sdim ip->ip_v = IPVERSION; 222292920Sdim ip->ip_hl = hlen >> 2; 223292920Sdim#ifdef RANDOM_IP_ID 224292920Sdim ip->ip_id = ip_randomid(); 225292920Sdim#else 226292920Sdim ip->ip_id = htons(ip_id++); 227292920Sdim#endif 228292920Sdim ipstat.ips_localout++; 229292920Sdim } else { 230292920Sdim hlen = ip->ip_hl << 2; 231292920Sdim } 232292920Sdim 233292920Sdim dst = (struct sockaddr_in *)&ro->ro_dst; 234292920Sdim /* 235292920Sdim * If there is a cached route, 236292920Sdim * check that it is to the same destination 237292920Sdim * and is still up. If not, free it and try again. 238292920Sdim * The address family should also be checked in case of sharing the 239292920Sdim * cache with IPv6. 240292920Sdim */ 241292920Sdim if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 242292920Sdim dst->sin_family != AF_INET || 243292920Sdim dst->sin_addr.s_addr != pkt_dst.s_addr)) { 244292920Sdim RTFREE(ro->ro_rt); 245292920Sdim ro->ro_rt = (struct rtentry *)0; 246292920Sdim } 247292920Sdim if (ro->ro_rt == 0) { 248292920Sdim bzero(dst, sizeof(*dst)); 249292920Sdim dst->sin_family = AF_INET; 250292920Sdim dst->sin_len = sizeof(*dst); 251292920Sdim dst->sin_addr = pkt_dst; 252292920Sdim } 253292920Sdim /* 254292920Sdim * If routing to interface only, 255292920Sdim * short circuit routing lookup. 256292920Sdim */ 257292920Sdim if (flags & IP_ROUTETOIF) { 258292920Sdim if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == 0 && 259292920Sdim (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == 0) { 260292920Sdim ipstat.ips_noroute++; 261292920Sdim error = ENETUNREACH; 262292920Sdim goto bad; 263292920Sdim } 264292920Sdim ifp = ia->ia_ifp; 265292920Sdim ip->ip_ttl = 1; 266292920Sdim isbroadcast = in_broadcast(dst->sin_addr, ifp); 267292920Sdim } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 268292920Sdim imo != NULL && imo->imo_multicast_ifp != NULL) { 269292920Sdim /* 270292920Sdim * Bypass the normal routing lookup for multicast 271292920Sdim * packets if the interface is specified. 272292920Sdim */ 273292920Sdim ifp = imo->imo_multicast_ifp; 274292920Sdim IFP_TO_IA(ifp, ia); 275292920Sdim isbroadcast = 0; /* fool gcc */ 276292920Sdim } else { 277292920Sdim /* 278292920Sdim * We want to do any cloning requested by the link layer, 279292920Sdim * as this is probably required in all cases for correct 280292920Sdim * operation (as it is for ARP). 281292920Sdim */ 282292920Sdim if (ro->ro_rt == 0) 283292920Sdim rtalloc(ro); 284292920Sdim if (ro->ro_rt == 0) { 285292920Sdim ipstat.ips_noroute++; 286292920Sdim error = EHOSTUNREACH; 287292920Sdim goto bad; 288292920Sdim } 289292920Sdim ia = ifatoia(ro->ro_rt->rt_ifa); 290292920Sdim ifp = ro->ro_rt->rt_ifp; 291292920Sdim ro->ro_rt->rt_rmx.rmx_pksent++; 292292920Sdim if (ro->ro_rt->rt_flags & RTF_GATEWAY) 293292920Sdim dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 294292920Sdim if (ro->ro_rt->rt_flags & RTF_HOST) 295292920Sdim isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 296292920Sdim else 297292920Sdim isbroadcast = in_broadcast(dst->sin_addr, ifp); 298292920Sdim } 299292920Sdim if (IN_MULTICAST(ntohl(pkt_dst.s_addr))) { 300292920Sdim struct in_multi *inm; 301292920Sdim 302292920Sdim m->m_flags |= M_MCAST; 303292920Sdim /* 304292920Sdim * IP destination address is multicast. Make sure "dst" 305292920Sdim * still points to the address in "ro". (It may have been 306292920Sdim * changed to point to a gateway address, above.) 307292920Sdim */ 308292920Sdim dst = (struct sockaddr_in *)&ro->ro_dst; 309292920Sdim /* 310292920Sdim * See if the caller provided any multicast options 311292920Sdim */ 312292920Sdim if (imo != NULL) { 313292920Sdim ip->ip_ttl = imo->imo_multicast_ttl; 314292920Sdim if (imo->imo_multicast_vif != -1) 315292920Sdim ip->ip_src.s_addr = 316292920Sdim ip_mcast_src ? 317292920Sdim ip_mcast_src(imo->imo_multicast_vif) : 318292920Sdim INADDR_ANY; 319292920Sdim } else 320292920Sdim ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 321292920Sdim /* 322292920Sdim * Confirm that the outgoing interface supports multicast. 323292920Sdim */ 324292920Sdim if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 325292920Sdim if ((ifp->if_flags & IFF_MULTICAST) == 0) { 326292920Sdim ipstat.ips_noroute++; 327292920Sdim error = ENETUNREACH; 328292920Sdim goto bad; 329292920Sdim } 330292920Sdim } 331292920Sdim /* 332292920Sdim * If source address not specified yet, use address 333292920Sdim * of outgoing interface. 334292920Sdim */ 335292920Sdim if (ip->ip_src.s_addr == INADDR_ANY) { 336292920Sdim /* Interface may have no addresses. */ 337292920Sdim if (ia != NULL) 338292920Sdim ip->ip_src = IA_SIN(ia)->sin_addr; 339292920Sdim } 340292920Sdim 341292920Sdim IN_LOOKUP_MULTI(pkt_dst, ifp, inm); 342292920Sdim if (inm != NULL && 343292920Sdim (imo == NULL || imo->imo_multicast_loop)) { 344292920Sdim /* 345292920Sdim * If we belong to the destination multicast group 346292920Sdim * on the outgoing interface, and the caller did not 347292920Sdim * forbid loopback, loop back a copy. 348292920Sdim */ 349292920Sdim ip_mloopback(ifp, m, dst, hlen); 350292920Sdim } 351292920Sdim else { 352292920Sdim /* 353292920Sdim * If we are acting as a multicast router, perform 354292920Sdim * multicast forwarding as if the packet had just 355292920Sdim * arrived on the interface to which we are about 356292920Sdim * to send. The multicast forwarding function 357292920Sdim * recursively calls this function, using the 358292920Sdim * IP_FORWARDING flag to prevent infinite recursion. 359292920Sdim * 360292920Sdim * Multicasts that are looped back by ip_mloopback(), 361292920Sdim * above, will be forwarded by the ip_input() routine, 362292920Sdim * if necessary. 363292920Sdim */ 364292920Sdim if (ip_mrouter && (flags & IP_FORWARDING) == 0) { 365292920Sdim /* 366292920Sdim * If rsvp daemon is not running, do not 367292920Sdim * set ip_moptions. This ensures that the packet 368292920Sdim * is multicast and not just sent down one link 369292920Sdim * as prescribed by rsvpd. 370292920Sdim */ 371292920Sdim if (!rsvp_on) 372292920Sdim imo = NULL; 373292920Sdim if (ip_mforward && 374292920Sdim ip_mforward(ip, ifp, m, imo) != 0) { 375292920Sdim m_freem(m); 376292920Sdim goto done; 377292920Sdim } 378292920Sdim } 379292920Sdim } 380292920Sdim 381292920Sdim /* 382292920Sdim * Multicasts with a time-to-live of zero may be looped- 383292920Sdim * back, above, but must not be transmitted on a network. 384292920Sdim * Also, multicasts addressed to the loopback interface 385292920Sdim * are not sent -- the above call to ip_mloopback() will 386292920Sdim * loop back a copy if this host actually belongs to the 387292920Sdim * destination group on the loopback interface. 388292920Sdim */ 389292920Sdim if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 390292920Sdim m_freem(m); 391292920Sdim goto done; 392292920Sdim } 393292920Sdim 394292920Sdim goto sendit; 395292920Sdim } 396292920Sdim#ifndef notdef 397292920Sdim /* 398292920Sdim * If the source address is not specified yet, use the address 399292920Sdim * of the outoing interface. In case, keep note we did that, so 400292920Sdim * if the the firewall changes the next-hop causing the output 401292920Sdim * interface to change, we can fix that. 402292920Sdim */ 403292920Sdim if (ip->ip_src.s_addr == INADDR_ANY) { 404292920Sdim /* Interface may have no addresses. */ 405292920Sdim if (ia != NULL) { 406292920Sdim ip->ip_src = IA_SIN(ia)->sin_addr; 407292920Sdim src_was_INADDR_ANY = 1; 408292920Sdim } 409292920Sdim } 410292920Sdim#endif /* notdef */ 411292920Sdim /* 412292920Sdim * Verify that we have any chance at all of being able to queue 413292920Sdim * the packet or packet fragments 414292920Sdim */ 415292920Sdim if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >= 416292920Sdim ifp->if_snd.ifq_maxlen) { 417292920Sdim error = ENOBUFS; 418292920Sdim ipstat.ips_odropped++; 419292920Sdim goto bad; 420292920Sdim } 421292920Sdim 422292920Sdim /* 423292920Sdim * Look for broadcast address and 424292920Sdim * verify user is allowed to send 425292920Sdim * such a packet. 426292920Sdim */ 427292920Sdim if (isbroadcast) { 428292920Sdim if ((ifp->if_flags & IFF_BROADCAST) == 0) { 429292920Sdim error = EADDRNOTAVAIL; 430292920Sdim goto bad; 431292920Sdim } 432292920Sdim if ((flags & IP_ALLOWBROADCAST) == 0) { 433292920Sdim error = EACCES; 434292920Sdim goto bad; 435292920Sdim } 436292920Sdim /* don't allow broadcast messages to be fragmented */ 437292920Sdim if (ip->ip_len > ifp->if_mtu) { 438292920Sdim error = EMSGSIZE; 439292920Sdim goto bad; 440292920Sdim } 441292920Sdim if (flags & IP_SENDONES) 442292920Sdim ip->ip_dst.s_addr = INADDR_BROADCAST; 443292920Sdim m->m_flags |= M_BCAST; 444292920Sdim } else { 445292920Sdim m->m_flags &= ~M_BCAST; 446292920Sdim } 447292920Sdim 448292920Sdimsendit: 449292920Sdim#ifdef IPSEC 450292920Sdim /* get SP for this packet */ 451292920Sdim if (inp == NULL) 452292920Sdim sp = ipsec4_getpolicybyaddr(m, IPSEC_DIR_OUTBOUND, 453292920Sdim flags, &error); 454292920Sdim else 455292920Sdim sp = ipsec4_getpolicybypcb(m, IPSEC_DIR_OUTBOUND, inp, &error); 456292920Sdim 457292920Sdim if (sp == NULL) { 458292920Sdim ipsecstat.out_inval++; 459292920Sdim goto bad; 460292920Sdim } 461292920Sdim 462292920Sdim error = 0; 463292920Sdim 464292920Sdim /* check policy */ 465292920Sdim switch (sp->policy) { 466292920Sdim case IPSEC_POLICY_DISCARD: 467292920Sdim /* 468292920Sdim * This packet is just discarded. 469292920Sdim */ 470292920Sdim ipsecstat.out_polvio++; 471292920Sdim goto bad; 472292920Sdim 473292920Sdim case IPSEC_POLICY_BYPASS: 474292920Sdim case IPSEC_POLICY_NONE: 475292920Sdim case IPSEC_POLICY_TCP: 476292920Sdim /* no need to do IPsec. */ 477292920Sdim goto skip_ipsec; 478292920Sdim 479292920Sdim case IPSEC_POLICY_IPSEC: 480292920Sdim if (sp->req == NULL) { 481292920Sdim /* acquire a policy */ 482292920Sdim error = key_spdacquire(sp); 483292920Sdim goto bad; 484292920Sdim } 485292920Sdim break; 486292920Sdim 487292920Sdim case IPSEC_POLICY_ENTRUST: 488292920Sdim default: 489292920Sdim printf("ip_output: Invalid policy found. %d\n", sp->policy); 490292920Sdim } 491292920Sdim { 492292920Sdim struct ipsec_output_state state; 493292920Sdim bzero(&state, sizeof(state)); 494292920Sdim state.m = m; 495292920Sdim if (flags & IP_ROUTETOIF) { 496292920Sdim state.ro = &iproute; 497292920Sdim bzero(&iproute, sizeof(iproute)); 498292920Sdim } else 499292920Sdim state.ro = ro; 500292920Sdim state.dst = (struct sockaddr *)dst; 501292920Sdim 502292920Sdim ip->ip_sum = 0; 503292920Sdim 504292920Sdim /* 505292920Sdim * XXX 506292920Sdim * delayed checksums are not currently compatible with IPsec 507292920Sdim */ 508292920Sdim if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 509292920Sdim in_delayed_cksum(m); 510292920Sdim m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 511292920Sdim } 512292920Sdim 513292920Sdim ip->ip_len = htons(ip->ip_len); 514292920Sdim ip->ip_off = htons(ip->ip_off); 515292920Sdim 516292920Sdim error = ipsec4_output(&state, sp, flags); 517292920Sdim 518292920Sdim m = state.m; 519292920Sdim if (flags & IP_ROUTETOIF) { 520292920Sdim /* 521292920Sdim * if we have tunnel mode SA, we may need to ignore 522292920Sdim * IP_ROUTETOIF. 523292920Sdim */ 524292920Sdim if (state.ro != &iproute || state.ro->ro_rt != NULL) { 525292920Sdim flags &= ~IP_ROUTETOIF; 526292920Sdim ro = state.ro; 527292920Sdim } 528292920Sdim } else 529292920Sdim ro = state.ro; 530292920Sdim dst = (struct sockaddr_in *)state.dst; 531292920Sdim if (error) { 532292920Sdim /* mbuf is already reclaimed in ipsec4_output. */ 533292920Sdim m = NULL; 534292920Sdim switch (error) { 535292920Sdim case EHOSTUNREACH: 536292920Sdim case ENETUNREACH: 537292920Sdim case EMSGSIZE: 538292920Sdim case ENOBUFS: 539292920Sdim case ENOMEM: 540292920Sdim break; 541292920Sdim default: 542292920Sdim printf("ip4_output (ipsec): error code %d\n", error); 543292920Sdim /*fall through*/ 544292920Sdim case ENOENT: 545292920Sdim /* don't show these error codes to the user */ 546292920Sdim error = 0; 547292920Sdim break; 548292920Sdim } 549292920Sdim goto bad; 550292920Sdim } 551292920Sdim 552292920Sdim /* be sure to update variables that are affected by ipsec4_output() */ 553292920Sdim ip = mtod(m, struct ip *); 554292920Sdim hlen = ip->ip_hl << 2; 555292920Sdim if (ro->ro_rt == NULL) { 556292920Sdim if ((flags & IP_ROUTETOIF) == 0) { 557292920Sdim printf("ip_output: " 558292920Sdim "can't update route after IPsec processing\n"); 559292920Sdim error = EHOSTUNREACH; /*XXX*/ 560292920Sdim goto bad; 561292920Sdim } 562292920Sdim } else { 563292920Sdim if (state.encap) { 564292920Sdim ia = ifatoia(ro->ro_rt->rt_ifa); 565292920Sdim ifp = ro->ro_rt->rt_ifp; 566292920Sdim } 567292920Sdim } 568292920Sdim } 569292920Sdim 570292920Sdim /* make it flipped, again. */ 571292920Sdim ip->ip_len = ntohs(ip->ip_len); 572292920Sdim ip->ip_off = ntohs(ip->ip_off); 573292920Sdimskip_ipsec: 574292920Sdim#endif /*IPSEC*/ 575292920Sdim#ifdef FAST_IPSEC 576292920Sdim /* 577292920Sdim * Check the security policy (SP) for the packet and, if 578292920Sdim * required, do IPsec-related processing. There are two 579292920Sdim * cases here; the first time a packet is sent through 580292920Sdim * it will be untagged and handled by ipsec4_checkpolicy. 581292920Sdim * If the packet is resubmitted to ip_output (e.g. after 582292920Sdim * AH, ESP, etc. processing), there will be a tag to bypass 583292920Sdim * the lookup and related policy checking. 584292920Sdim */ 585292920Sdim mtag = m_tag_find(m, PACKET_TAG_IPSEC_PENDING_TDB, NULL); 586292920Sdim s = splnet(); 587292920Sdim if (mtag != NULL) { 588292920Sdim tdbi = (struct tdb_ident *)(mtag + 1); 589292920Sdim sp = ipsec_getpolicy(tdbi, IPSEC_DIR_OUTBOUND); 590292920Sdim if (sp == NULL) 591292920Sdim error = -EINVAL; /* force silent drop */ 592292920Sdim m_tag_delete(m, mtag); 593292920Sdim } else { 594292920Sdim sp = ipsec4_checkpolicy(m, IPSEC_DIR_OUTBOUND, flags, 595292920Sdim &error, inp); 596292920Sdim } 597292920Sdim /* 598292920Sdim * There are four return cases: 599292920Sdim * sp != NULL apply IPsec policy 600292920Sdim * sp == NULL, error == 0 no IPsec handling needed 601292920Sdim * sp == NULL, error == -EINVAL discard packet w/o error 602292920Sdim * sp == NULL, error != 0 discard packet, report error 603292920Sdim */ 604292920Sdim if (sp != NULL) { 605292920Sdim /* Loop detection, check if ipsec processing already done */ 606292920Sdim KASSERT(sp->req != NULL, ("ip_output: no ipsec request")); 607292920Sdim for (mtag = m_tag_first(m); mtag != NULL; 608292920Sdim mtag = m_tag_next(m, mtag)) { 609292920Sdim if (mtag->m_tag_cookie != MTAG_ABI_COMPAT) 610292920Sdim continue; 611292920Sdim if (mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_DONE && 612292920Sdim mtag->m_tag_id != PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED) 613292920Sdim continue; 614292920Sdim /* 615292920Sdim * Check if policy has an SA associated with it. 616292920Sdim * This can happen when an SP has yet to acquire 617292920Sdim * an SA; e.g. on first reference. If it occurs, 618292920Sdim * then we let ipsec4_process_packet do its thing. 619292920Sdim */ 620292920Sdim if (sp->req->sav == NULL) 621292920Sdim break; 622292920Sdim tdbi = (struct tdb_ident *)(mtag + 1); 623292920Sdim if (tdbi->spi == sp->req->sav->spi && 624292920Sdim tdbi->proto == sp->req->sav->sah->saidx.proto && 625292920Sdim bcmp(&tdbi->dst, &sp->req->sav->sah->saidx.dst, 626292920Sdim sizeof (union sockaddr_union)) == 0) { 627292920Sdim /* 628292920Sdim * No IPsec processing is needed, free 629292920Sdim * reference to SP. 630292920Sdim * 631292920Sdim * NB: null pointer to avoid free at 632292920Sdim * done: below. 633292920Sdim */ 634292920Sdim KEY_FREESP(&sp), sp = NULL; 635292920Sdim splx(s); 636292920Sdim goto spd_done; 637292920Sdim } 638292920Sdim } 639292920Sdim 640292920Sdim /* 641292920Sdim * Do delayed checksums now because we send before 642292920Sdim * this is done in the normal processing path. 643292920Sdim */ 644292920Sdim if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 645292920Sdim in_delayed_cksum(m); 646292920Sdim m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 647292920Sdim } 648292920Sdim 649292920Sdim ip->ip_len = htons(ip->ip_len); 650292920Sdim ip->ip_off = htons(ip->ip_off); 651292920Sdim 652292920Sdim /* NB: callee frees mbuf */ 653292920Sdim error = ipsec4_process_packet(m, sp->req, flags, 0); 654292920Sdim /* 655292920Sdim * Preserve KAME behaviour: ENOENT can be returned 656292920Sdim * when an SA acquire is in progress. Don't propagate 657292920Sdim * this to user-level; it confuses applications. 658292920Sdim * 659292920Sdim * XXX this will go away when the SADB is redone. 660292920Sdim */ 661292920Sdim if (error == ENOENT) 662292920Sdim error = 0; 663292920Sdim splx(s); 664292920Sdim goto done; 665292920Sdim } else { 666292920Sdim splx(s); 667292920Sdim 668292920Sdim if (error != 0) { 669292920Sdim /* 670292920Sdim * Hack: -EINVAL is used to signal that a packet 671292920Sdim * should be silently discarded. This is typically 672292920Sdim * because we asked key management for an SA and 673292920Sdim * it was delayed (e.g. kicked up to IKE). 674292920Sdim */ 675292920Sdim if (error == -EINVAL) 676292920Sdim error = 0; 677292920Sdim goto bad; 678292920Sdim } else { 679292920Sdim /* No IPsec processing for this packet. */ 680292920Sdim } 681292920Sdim#ifdef notyet 682292920Sdim /* 683292920Sdim * If deferred crypto processing is needed, check that 684292920Sdim * the interface supports it. 685292920Sdim */ 686292920Sdim mtag = m_tag_find(m, PACKET_TAG_IPSEC_OUT_CRYPTO_NEEDED, NULL); 687292920Sdim if (mtag != NULL && (ifp->if_capenable & IFCAP_IPSEC) == 0) { 688292920Sdim /* notify IPsec to do its own crypto */ 689292920Sdim ipsp_skipcrypto_unmark((struct tdb_ident *)(mtag + 1)); 690292920Sdim error = EHOSTUNREACH; 691292920Sdim goto bad; 692292920Sdim } 693292920Sdim#endif 694292920Sdim } 695292920Sdimspd_done: 696292920Sdim#endif /* FAST_IPSEC */ 697292920Sdim 698292920Sdim /* 699292920Sdim * IpHack's section. 700292920Sdim * - Xlate: translate packet's addr/port (NAT). 701292920Sdim * - Firewall: deny/allow/etc. 702292920Sdim * - Wrap: fake packet's addr/port <unimpl.> 703292920Sdim * - Encapsulate: put it in another IP and send out. <unimp.> 704292920Sdim */ 705292920Sdim#ifdef PFIL_HOOKS 706292920Sdim /* 707292920Sdim * Run through list of hooks for output packets. 708292920Sdim */ 709292920Sdim error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT); 710292920Sdim if (error != 0 || m == NULL) 711292920Sdim goto done; 712292920Sdim ip = mtod(m, struct ip *); 713292920Sdim#endif /* PFIL_HOOKS */ 714292920Sdim 715292920Sdim /* 716292920Sdim * Check with the firewall... 717292920Sdim * but not if we are already being fwd'd from a firewall. 718292920Sdim */ 719292920Sdim if (fw_enable && IPFW_LOADED && !args.next_hop) { 720292920Sdim struct sockaddr_in *old = dst; 721292920Sdim 722292920Sdim args.m = m; 723292920Sdim args.next_hop = dst; 724292920Sdim args.oif = ifp; 725292920Sdim off = ip_fw_chk_ptr(&args); 726292920Sdim m = args.m; 727292920Sdim dst = args.next_hop; 728292920Sdim 729292920Sdim /* 730292920Sdim * On return we must do the following: 731292920Sdim * m == NULL -> drop the pkt (old interface, deprecated) 732292920Sdim * (off & IP_FW_PORT_DENY_FLAG) -> drop the pkt (new interface) 733292920Sdim * 1<=off<= 0xffff -> DIVERT 734292920Sdim * (off & IP_FW_PORT_DYNT_FLAG) -> send to a DUMMYNET pipe 735292920Sdim * (off & IP_FW_PORT_TEE_FLAG) -> TEE the packet 736292920Sdim * dst != old -> IPFIREWALL_FORWARD 737292920Sdim * off==0, dst==old -> accept 738292920Sdim * If some of the above modules are not compiled in, then 739292920Sdim * we should't have to check the corresponding condition 740292920Sdim * (because the ipfw control socket should not accept 741292920Sdim * unsupported rules), but better play safe and drop 742292920Sdim * packets in case of doubt. 743292920Sdim */ 744292920Sdim if ( (off & IP_FW_PORT_DENY_FLAG) || m == NULL) { 745292920Sdim if (m) 746292920Sdim m_freem(m); 747292920Sdim error = EACCES; 748292920Sdim goto done; 749292920Sdim } 750292920Sdim ip = mtod(m, struct ip *); 751292920Sdim if (off == 0 && dst == old) /* common case */ 752292920Sdim goto pass; 753292920Sdim if (DUMMYNET_LOADED && (off & IP_FW_PORT_DYNT_FLAG) != 0) { 754292920Sdim /* 755292920Sdim * pass the pkt to dummynet. Need to include 756292920Sdim * pipe number, m, ifp, ro, dst because these are 757292920Sdim * not recomputed in the next pass. 758292920Sdim * All other parameters have been already used and 759292920Sdim * so they are not needed anymore. 760292920Sdim * XXX note: if the ifp or ro entry are deleted 761292920Sdim * while a pkt is in dummynet, we are in trouble! 762292920Sdim */ 763292920Sdim args.ro = ro; 764292920Sdim args.dst = dst; 765292920Sdim args.flags = flags; 766292920Sdim 767292920Sdim error = ip_dn_io_ptr(m, off & 0xffff, DN_TO_IP_OUT, 768292920Sdim &args); 769292920Sdim goto done; 770292920Sdim } 771292920Sdim#ifdef IPDIVERT 772292920Sdim if (off != 0 && (off & IP_FW_PORT_DYNT_FLAG) == 0) { 773292920Sdim struct mbuf *clone; 774292920Sdim 775292920Sdim /* Clone packet if we're doing a 'tee' */ 776292920Sdim if ((off & IP_FW_PORT_TEE_FLAG) != 0) 777292920Sdim clone = divert_clone(m); 778292920Sdim else 779292920Sdim clone = NULL; 780292920Sdim 781292920Sdim /* 782292920Sdim * XXX 783292920Sdim * delayed checksums are not currently compatible 784292920Sdim * with divert sockets. 785292920Sdim */ 786292920Sdim if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 787292920Sdim in_delayed_cksum(m); 788292920Sdim m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 789292920Sdim } 790292920Sdim 791292920Sdim /* Restore packet header fields to original values */ 792292920Sdim ip->ip_len = htons(ip->ip_len); 793292920Sdim ip->ip_off = htons(ip->ip_off); 794292920Sdim 795292920Sdim /* Deliver packet to divert input routine */ 796292920Sdim divert_packet(m, 0); 797292920Sdim 798292920Sdim /* If 'tee', continue with original packet */ 799292920Sdim if (clone != NULL) { 800292920Sdim m = clone; 801292920Sdim ip = mtod(m, struct ip *); 802292920Sdim goto pass; 803292920Sdim } 804292920Sdim goto done; 805292920Sdim } 806292920Sdim#endif 807292920Sdim 808292920Sdim /* IPFIREWALL_FORWARD */ 809292920Sdim /* 810292920Sdim * Check dst to make sure it is directly reachable on the 811292920Sdim * interface we previously thought it was. 812292920Sdim * If it isn't (which may be likely in some situations) we have 813292920Sdim * to re-route it (ie, find a route for the next-hop and the 814292920Sdim * associated interface) and set them here. This is nested 815292920Sdim * forwarding which in most cases is undesirable, except where 816292920Sdim * such control is nigh impossible. So we do it here. 817292920Sdim * And I'm babbling. 818292920Sdim */ 819292920Sdim if (off == 0 && old != dst) { /* FORWARD, dst has changed */ 820292920Sdim#if 0 821292920Sdim /* 822292920Sdim * XXX To improve readability, this block should be 823292920Sdim * changed into a function call as below: 824292920Sdim */ 825292920Sdim error = ip_ipforward(&m, &dst, &ifp); 826292920Sdim if (error) 827292920Sdim goto bad; 828292920Sdim if (m == NULL) /* ip_input consumed the mbuf */ 829292920Sdim goto done; 830292920Sdim#else 831292920Sdim struct in_ifaddr *ia; 832292920Sdim 833292920Sdim /* 834292920Sdim * XXX sro_fwd below is static, and a pointer 835292920Sdim * to it gets passed to routines downstream. 836292920Sdim * This could have surprisingly bad results in 837292920Sdim * practice, because its content is overwritten 838292920Sdim * by subsequent packets. 839292920Sdim */ 840292920Sdim /* There must be a better way to do this next line... */ 841292920Sdim static struct route sro_fwd; 842292920Sdim struct route *ro_fwd = &sro_fwd; 843292920Sdim 844292920Sdim#if 0 845292920Sdim print_ip("IPFIREWALL_FORWARD: New dst ip: ", 846292920Sdim dst->sin_addr, "\n"); 847292920Sdim#endif 848292920Sdim 849292920Sdim /* 850292920Sdim * We need to figure out if we have been forwarded 851292920Sdim * to a local socket. If so, then we should somehow 852292920Sdim * "loop back" to ip_input, and get directed to the 853292920Sdim * PCB as if we had received this packet. This is 854292920Sdim * because it may be dificult to identify the packets 855292920Sdim * you want to forward until they are being output 856292920Sdim * and have selected an interface. (e.g. locally 857292920Sdim * initiated packets) If we used the loopback inteface, 858292920Sdim * we would not be able to control what happens 859292920Sdim * as the packet runs through ip_input() as 860292920Sdim * it is done through an ISR. 861292920Sdim */ 862292920Sdim LIST_FOREACH(ia, 863292920Sdim INADDR_HASH(dst->sin_addr.s_addr), ia_hash) { 864292920Sdim /* 865292920Sdim * If the addr to forward to is one 866292920Sdim * of ours, we pretend to 867292920Sdim * be the destination for this packet. 868292920Sdim */ 869292920Sdim if (IA_SIN(ia)->sin_addr.s_addr == 870292920Sdim dst->sin_addr.s_addr) 871292920Sdim break; 872292920Sdim } 873292920Sdim if (ia) { /* tell ip_input "dont filter" */ 874292920Sdim mtag = m_tag_get( 875292920Sdim PACKET_TAG_IPFORWARD, 876292920Sdim sizeof(struct sockaddr_in *), M_NOWAIT); 877292920Sdim if (mtag == NULL) { 878292920Sdim error = ENOBUFS; 879292920Sdim goto bad; 880292920Sdim } 881292920Sdim *(struct sockaddr_in **)(mtag+1) = 882292920Sdim args.next_hop; 883292920Sdim m_tag_prepend(m, mtag); 884292920Sdim 885292920Sdim if (m->m_pkthdr.rcvif == NULL) 886292920Sdim m->m_pkthdr.rcvif = ifunit("lo0"); 887292920Sdim if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 888292920Sdim m->m_pkthdr.csum_flags |= 889292920Sdim CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 890292920Sdim m->m_pkthdr.csum_data = 0xffff; 891292920Sdim } 892292920Sdim m->m_pkthdr.csum_flags |= 893292920Sdim CSUM_IP_CHECKED | CSUM_IP_VALID; 894292920Sdim ip->ip_len = htons(ip->ip_len); 895292920Sdim ip->ip_off = htons(ip->ip_off); 896292920Sdim ip_input(m); 897292920Sdim goto done; 898292920Sdim } 899292920Sdim /* 900292920Sdim * Some of the logic for this was 901292920Sdim * nicked from above. 902292920Sdim */ 903292920Sdim bcopy(dst, &ro_fwd->ro_dst, sizeof(*dst)); 904292920Sdim 905292920Sdim ro_fwd->ro_rt = 0; 906292920Sdim rtalloc_ign(ro_fwd, RTF_CLONING); 907292920Sdim 908292920Sdim if (ro_fwd->ro_rt == 0) { 909292920Sdim ipstat.ips_noroute++; 910292920Sdim error = EHOSTUNREACH; 911292920Sdim goto bad; 912292920Sdim } 913292920Sdim 914292920Sdim ia = ifatoia(ro_fwd->ro_rt->rt_ifa); 915292920Sdim ifp = ro_fwd->ro_rt->rt_ifp; 916292920Sdim ro_fwd->ro_rt->rt_rmx.rmx_pksent++; 917292920Sdim if (ro_fwd->ro_rt->rt_flags & RTF_GATEWAY) 918292920Sdim dst = (struct sockaddr_in *) 919292920Sdim ro_fwd->ro_rt->rt_gateway; 920292920Sdim if (ro_fwd->ro_rt->rt_flags & RTF_HOST) 921292920Sdim isbroadcast = 922292920Sdim (ro_fwd->ro_rt->rt_flags & RTF_BROADCAST); 923292920Sdim else 924292920Sdim isbroadcast = in_broadcast(dst->sin_addr, ifp); 925292920Sdim if (ro->ro_rt) 926292920Sdim RTFREE(ro->ro_rt); 927292920Sdim ro->ro_rt = ro_fwd->ro_rt; 928292920Sdim dst = (struct sockaddr_in *)&ro_fwd->ro_dst; 929292920Sdim 930292920Sdim#endif /* ... block to be put into a function */ 931292920Sdim /* 932292920Sdim * If we added a default src ip earlier, 933292920Sdim * which would have been gotten from the-then 934292920Sdim * interface, do it again, from the new one. 935292920Sdim */ 936292920Sdim if (src_was_INADDR_ANY) 937292920Sdim ip->ip_src = IA_SIN(ia)->sin_addr; 938292920Sdim goto pass ; 939292920Sdim } 940292920Sdim 941292920Sdim /* 942292920Sdim * if we get here, none of the above matches, and 943292920Sdim * we have to drop the pkt 944292920Sdim */ 945292920Sdim m_freem(m); 946292920Sdim error = EACCES; /* not sure this is the right error msg */ 947292920Sdim goto done; 948292920Sdim } 949292920Sdim 950292920Sdimpass: 951292920Sdim /* 127/8 must not appear on wire - RFC1122. */ 952292920Sdim if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 953292920Sdim (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 954292920Sdim if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 955292920Sdim ipstat.ips_badaddr++; 956292920Sdim error = EADDRNOTAVAIL; 957292920Sdim goto bad; 958292920Sdim } 959292920Sdim } 960292920Sdim 961292920Sdim m->m_pkthdr.csum_flags |= CSUM_IP; 962292920Sdim sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 963292920Sdim if (sw_csum & CSUM_DELAY_DATA) { 964292920Sdim in_delayed_cksum(m); 965292920Sdim sw_csum &= ~CSUM_DELAY_DATA; 966292920Sdim } 967292920Sdim m->m_pkthdr.csum_flags &= ifp->if_hwassist; 968292920Sdim 969292920Sdim /* 970292920Sdim * If small enough for interface, or the interface will take 971292920Sdim * care of the fragmentation for us, can just send directly. 972292920Sdim */ 973292920Sdim if (ip->ip_len <= ifp->if_mtu || (ifp->if_hwassist & CSUM_FRAGMENT && 974292920Sdim ((ip->ip_off & IP_DF) == 0))) { 975292920Sdim ip->ip_len = htons(ip->ip_len); 976292920Sdim ip->ip_off = htons(ip->ip_off); 977292920Sdim ip->ip_sum = 0; 978292920Sdim if (sw_csum & CSUM_DELAY_IP) 979292920Sdim ip->ip_sum = in_cksum(m, hlen); 980292920Sdim 981292920Sdim /* Record statistics for this interface address. */ 982292920Sdim if (!(flags & IP_FORWARDING) && ia) { 983292920Sdim ia->ia_ifa.if_opackets++; 984292920Sdim ia->ia_ifa.if_obytes += m->m_pkthdr.len; 985292920Sdim } 986292920Sdim 987292920Sdim#ifdef IPSEC 988292920Sdim /* clean ipsec history once it goes out of the node */ 989292920Sdim ipsec_delaux(m); 990292920Sdim#endif 991292920Sdim 992292920Sdim#ifdef MBUF_STRESS_TEST 993292920Sdim if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 994292920Sdim m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 995292920Sdim#endif 996292920Sdim error = (*ifp->if_output)(ifp, m, 997292920Sdim (struct sockaddr *)dst, ro->ro_rt); 998292920Sdim goto done; 999292920Sdim } 1000292920Sdim 1001292920Sdim if (ip->ip_off & IP_DF) { 1002292920Sdim error = EMSGSIZE; 1003292920Sdim /* 1004292920Sdim * This case can happen if the user changed the MTU 1005292920Sdim * of an interface after enabling IP on it. Because 1006292920Sdim * most netifs don't keep track of routes pointing to 1007292920Sdim * them, there is no way for one to update all its 1008292920Sdim * routes when the MTU is changed. 1009292920Sdim */ 1010292920Sdim if ((ro->ro_rt->rt_flags & (RTF_UP | RTF_HOST)) && 1011292920Sdim (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu)) { 1012292920Sdim ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 1013292920Sdim } 1014292920Sdim ipstat.ips_cantfrag++; 1015292920Sdim goto bad; 1016292920Sdim } 1017292920Sdim 1018292920Sdim /* 1019292920Sdim * Too large for interface; fragment if possible. If successful, 1020292920Sdim * on return, m will point to a list of packets to be sent. 1021292920Sdim */ 1022292920Sdim error = ip_fragment(ip, &m, ifp->if_mtu, ifp->if_hwassist, sw_csum); 1023292920Sdim if (error) 1024292920Sdim goto bad; 1025292920Sdim for (; m; m = m0) { 1026292920Sdim m0 = m->m_nextpkt; 1027292920Sdim m->m_nextpkt = 0; 1028292920Sdim#ifdef IPSEC 1029292920Sdim /* clean ipsec history once it goes out of the node */ 1030292920Sdim ipsec_delaux(m); 1031292920Sdim#endif 1032292920Sdim if (error == 0) { 1033292920Sdim /* Record statistics for this interface address. */ 1034292920Sdim if (ia != NULL) { 1035292920Sdim ia->ia_ifa.if_opackets++; 1036292920Sdim ia->ia_ifa.if_obytes += m->m_pkthdr.len; 1037292920Sdim } 1038292920Sdim 1039292920Sdim error = (*ifp->if_output)(ifp, m, 1040292920Sdim (struct sockaddr *)dst, ro->ro_rt); 1041292920Sdim } else 1042292920Sdim m_freem(m); 1043292920Sdim } 1044292920Sdim 1045292920Sdim if (error == 0) 1046292920Sdim ipstat.ips_fragmented++; 1047292920Sdim 1048292920Sdimdone: 1049292920Sdim if (ro == &iproute && ro->ro_rt) { 1050292920Sdim RTFREE(ro->ro_rt); 1051292920Sdim ro->ro_rt = NULL; 1052292920Sdim } 1053292920Sdim if (dummytag) { 1054292920Sdim struct dn_pkt_tag *dt = (struct dn_pkt_tag *)(dummytag+1); 1055292920Sdim if (dt->ro.ro_rt) 1056292920Sdim RTFREE(dt->ro.ro_rt); 1057292920Sdim m_tag_free(dummytag); 1058292920Sdim } 1059292920Sdim#ifdef IPSEC 1060292920Sdim if (sp != NULL) { 1061292920Sdim KEYDEBUG(KEYDEBUG_IPSEC_STAMP, 1062292920Sdim printf("DP ip_output call free SP:%p\n", sp)); 1063292920Sdim key_freesp(sp); 1064292920Sdim } 1065292920Sdim#endif 1066292920Sdim#ifdef FAST_IPSEC 1067292920Sdim if (sp != NULL) 1068292920Sdim KEY_FREESP(&sp); 1069292920Sdim#endif 1070292920Sdim return (error); 1071292920Sdimbad: 1072292920Sdim m_freem(m); 1073292920Sdim goto done; 1074292920Sdim} 1075292920Sdim 1076292920Sdim/* 1077292920Sdim * Create a chain of fragments which fit the given mtu. m_frag points to the 1078292920Sdim * mbuf to be fragmented; on return it points to the chain with the fragments. 1079292920Sdim * Return 0 if no error. If error, m_frag may contain a partially built 1080292920Sdim * chain of fragments that should be freed by the caller. 1081292920Sdim * 1082292920Sdim * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 1083292920Sdim * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 1084292920Sdim */ 1085292920Sdimint 1086292920Sdimip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 1087292920Sdim u_long if_hwassist_flags, int sw_csum) 1088292920Sdim{ 1089292920Sdim int error = 0; 1090292920Sdim int hlen = ip->ip_hl << 2; 1091292920Sdim int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 1092292920Sdim int off; 1093292920Sdim struct mbuf *m0 = *m_frag; /* the original packet */ 1094292920Sdim int firstlen; 1095292920Sdim struct mbuf **mnext; 1096292920Sdim int nfrags; 1097292920Sdim 1098292920Sdim if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 1099292920Sdim ipstat.ips_cantfrag++; 1100292920Sdim return EMSGSIZE; 1101292920Sdim } 1102292920Sdim 1103292920Sdim /* 1104292920Sdim * Must be able to put at least 8 bytes per fragment. 1105292920Sdim */ 1106292920Sdim if (len < 8) 1107292920Sdim return EMSGSIZE; 1108292920Sdim 1109292920Sdim /* 1110292920Sdim * If the interface will not calculate checksums on 1111292920Sdim * fragmented packets, then do it here. 1112292920Sdim */ 1113292920Sdim if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 1114292920Sdim (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 1115292920Sdim in_delayed_cksum(m0); 1116292920Sdim m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1117292920Sdim } 1118292920Sdim 1119292920Sdim if (len > PAGE_SIZE) { 1120292920Sdim /* 1121292920Sdim * Fragment large datagrams such that each segment 1122292920Sdim * contains a multiple of PAGE_SIZE amount of data, 1123292920Sdim * plus headers. This enables a receiver to perform 1124292920Sdim * page-flipping zero-copy optimizations. 1125292920Sdim * 1126292920Sdim * XXX When does this help given that sender and receiver 1127292920Sdim * could have different page sizes, and also mtu could 1128292920Sdim * be less than the receiver's page size ? 1129292920Sdim */ 1130292920Sdim int newlen; 1131292920Sdim struct mbuf *m; 1132292920Sdim 1133292920Sdim for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 1134292920Sdim off += m->m_len; 1135292920Sdim 1136292920Sdim /* 1137292920Sdim * firstlen (off - hlen) must be aligned on an 1138292920Sdim * 8-byte boundary 1139292920Sdim */ 1140292920Sdim if (off < hlen) 1141292920Sdim goto smart_frag_failure; 1142292920Sdim off = ((off - hlen) & ~7) + hlen; 1143292920Sdim newlen = (~PAGE_MASK) & mtu; 1144292920Sdim if ((newlen + sizeof (struct ip)) > mtu) { 1145292920Sdim /* we failed, go back the default */ 1146292920Sdimsmart_frag_failure: 1147292920Sdim newlen = len; 1148292920Sdim off = hlen + len; 1149292920Sdim } 1150292920Sdim len = newlen; 1151292920Sdim 1152292920Sdim } else { 1153292920Sdim off = hlen + len; 1154292920Sdim } 1155292920Sdim 1156292920Sdim firstlen = off - hlen; 1157292920Sdim mnext = &m0->m_nextpkt; /* pointer to next packet */ 1158292920Sdim 1159292920Sdim /* 1160292920Sdim * Loop through length of segment after first fragment, 1161292920Sdim * make new header and copy data of each part and link onto chain. 1162292920Sdim * Here, m0 is the original packet, m is the fragment being created. 1163292920Sdim * The fragments are linked off the m_nextpkt of the original 1164292920Sdim * packet, which after processing serves as the first fragment. 1165292920Sdim */ 1166292920Sdim for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 1167292920Sdim struct ip *mhip; /* ip header on the fragment */ 1168292920Sdim struct mbuf *m; 1169292920Sdim int mhlen = sizeof (struct ip); 1170292920Sdim 1171292920Sdim MGETHDR(m, M_DONTWAIT, MT_HEADER); 1172292920Sdim if (m == 0) { 1173292920Sdim error = ENOBUFS; 1174292920Sdim ipstat.ips_odropped++; 1175292920Sdim goto done; 1176292920Sdim } 1177292920Sdim m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 1178292920Sdim /* 1179292920Sdim * In the first mbuf, leave room for the link header, then 1180292920Sdim * copy the original IP header including options. The payload 1181292920Sdim * goes into an additional mbuf chain returned by m_copy(). 1182292920Sdim */ 1183292920Sdim m->m_data += max_linkhdr; 1184292920Sdim mhip = mtod(m, struct ip *); 1185292920Sdim *mhip = *ip; 1186292920Sdim if (hlen > sizeof (struct ip)) { 1187292920Sdim mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 1188292920Sdim mhip->ip_v = IPVERSION; 1189292920Sdim mhip->ip_hl = mhlen >> 2; 1190292920Sdim } 1191292920Sdim m->m_len = mhlen; 1192292920Sdim /* XXX do we need to add ip->ip_off below ? */ 1193292920Sdim mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 1194292920Sdim if (off + len >= ip->ip_len) { /* last fragment */ 1195292920Sdim len = ip->ip_len - off; 1196292920Sdim m->m_flags |= M_LASTFRAG; 1197292920Sdim } else 1198292920Sdim mhip->ip_off |= IP_MF; 1199292920Sdim mhip->ip_len = htons((u_short)(len + mhlen)); 1200292920Sdim m->m_next = m_copy(m0, off, len); 1201292920Sdim if (m->m_next == 0) { /* copy failed */ 1202 m_free(m); 1203 error = ENOBUFS; /* ??? */ 1204 ipstat.ips_odropped++; 1205 goto done; 1206 } 1207 m->m_pkthdr.len = mhlen + len; 1208 m->m_pkthdr.rcvif = (struct ifnet *)0; 1209#ifdef MAC 1210 mac_create_fragment(m0, m); 1211#endif 1212 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 1213 mhip->ip_off = htons(mhip->ip_off); 1214 mhip->ip_sum = 0; 1215 if (sw_csum & CSUM_DELAY_IP) 1216 mhip->ip_sum = in_cksum(m, mhlen); 1217 *mnext = m; 1218 mnext = &m->m_nextpkt; 1219 } 1220 ipstat.ips_ofragments += nfrags; 1221 1222 /* set first marker for fragment chain */ 1223 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 1224 m0->m_pkthdr.csum_data = nfrags; 1225 1226 /* 1227 * Update first fragment by trimming what's been copied out 1228 * and updating header. 1229 */ 1230 m_adj(m0, hlen + firstlen - ip->ip_len); 1231 m0->m_pkthdr.len = hlen + firstlen; 1232 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 1233 ip->ip_off |= IP_MF; 1234 ip->ip_off = htons(ip->ip_off); 1235 ip->ip_sum = 0; 1236 if (sw_csum & CSUM_DELAY_IP) 1237 ip->ip_sum = in_cksum(m0, hlen); 1238 1239done: 1240 *m_frag = m0; 1241 return error; 1242} 1243 1244void 1245in_delayed_cksum(struct mbuf *m) 1246{ 1247 struct ip *ip; 1248 u_short csum, offset; 1249 1250 ip = mtod(m, struct ip *); 1251 offset = ip->ip_hl << 2 ; 1252 csum = in_cksum_skip(m, ip->ip_len, offset); 1253 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 1254 csum = 0xffff; 1255 offset += m->m_pkthdr.csum_data; /* checksum offset */ 1256 1257 if (offset + sizeof(u_short) > m->m_len) { 1258 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 1259 m->m_len, offset, ip->ip_p); 1260 /* 1261 * XXX 1262 * this shouldn't happen, but if it does, the 1263 * correct behavior may be to insert the checksum 1264 * in the existing chain instead of rearranging it. 1265 */ 1266 m = m_pullup(m, offset + sizeof(u_short)); 1267 } 1268 *(u_short *)(m->m_data + offset) = csum; 1269} 1270 1271/* 1272 * Insert IP options into preformed packet. 1273 * Adjust IP destination as required for IP source routing, 1274 * as indicated by a non-zero in_addr at the start of the options. 1275 * 1276 * XXX This routine assumes that the packet has no options in place. 1277 */ 1278static struct mbuf * 1279ip_insertoptions(m, opt, phlen) 1280 register struct mbuf *m; 1281 struct mbuf *opt; 1282 int *phlen; 1283{ 1284 register struct ipoption *p = mtod(opt, struct ipoption *); 1285 struct mbuf *n; 1286 register struct ip *ip = mtod(m, struct ip *); 1287 unsigned optlen; 1288 1289 optlen = opt->m_len - sizeof(p->ipopt_dst); 1290 if (optlen + ip->ip_len > IP_MAXPACKET) { 1291 *phlen = 0; 1292 return (m); /* XXX should fail */ 1293 } 1294 if (p->ipopt_dst.s_addr) 1295 ip->ip_dst = p->ipopt_dst; 1296 if (m->m_flags & M_EXT || m->m_data - optlen < m->m_pktdat) { 1297 MGETHDR(n, M_DONTWAIT, MT_HEADER); 1298 if (n == 0) { 1299 *phlen = 0; 1300 return (m); 1301 } 1302 n->m_pkthdr.rcvif = (struct ifnet *)0; 1303#ifdef MAC 1304 mac_create_mbuf_from_mbuf(m, n); 1305#endif 1306 n->m_pkthdr.len = m->m_pkthdr.len + optlen; 1307 m->m_len -= sizeof(struct ip); 1308 m->m_data += sizeof(struct ip); 1309 n->m_next = m; 1310 m = n; 1311 m->m_len = optlen + sizeof(struct ip); 1312 m->m_data += max_linkhdr; 1313 bcopy(ip, mtod(m, void *), sizeof(struct ip)); 1314 } else { 1315 m->m_data -= optlen; 1316 m->m_len += optlen; 1317 m->m_pkthdr.len += optlen; 1318 bcopy(ip, mtod(m, void *), sizeof(struct ip)); 1319 } 1320 ip = mtod(m, struct ip *); 1321 bcopy(p->ipopt_list, ip + 1, optlen); 1322 *phlen = sizeof(struct ip) + optlen; 1323 ip->ip_v = IPVERSION; 1324 ip->ip_hl = *phlen >> 2; 1325 ip->ip_len += optlen; 1326 return (m); 1327} 1328 1329/* 1330 * Copy options from ip to jp, 1331 * omitting those not copied during fragmentation. 1332 */ 1333int 1334ip_optcopy(ip, jp) 1335 struct ip *ip, *jp; 1336{ 1337 register u_char *cp, *dp; 1338 int opt, optlen, cnt; 1339 1340 cp = (u_char *)(ip + 1); 1341 dp = (u_char *)(jp + 1); 1342 cnt = (ip->ip_hl << 2) - sizeof (struct ip); 1343 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1344 opt = cp[0]; 1345 if (opt == IPOPT_EOL) 1346 break; 1347 if (opt == IPOPT_NOP) { 1348 /* Preserve for IP mcast tunnel's LSRR alignment. */ 1349 *dp++ = IPOPT_NOP; 1350 optlen = 1; 1351 continue; 1352 } 1353 1354 KASSERT(cnt >= IPOPT_OLEN + sizeof(*cp), 1355 ("ip_optcopy: malformed ipv4 option")); 1356 optlen = cp[IPOPT_OLEN]; 1357 KASSERT(optlen >= IPOPT_OLEN + sizeof(*cp) && optlen <= cnt, 1358 ("ip_optcopy: malformed ipv4 option")); 1359 1360 /* bogus lengths should have been caught by ip_dooptions */ 1361 if (optlen > cnt) 1362 optlen = cnt; 1363 if (IPOPT_COPIED(opt)) { 1364 bcopy(cp, dp, optlen); 1365 dp += optlen; 1366 } 1367 } 1368 for (optlen = dp - (u_char *)(jp+1); optlen & 0x3; optlen++) 1369 *dp++ = IPOPT_EOL; 1370 return (optlen); 1371} 1372 1373/* 1374 * IP socket option processing. 1375 */ 1376int 1377ip_ctloutput(so, sopt) 1378 struct socket *so; 1379 struct sockopt *sopt; 1380{ 1381 struct inpcb *inp = sotoinpcb(so); 1382 int error, optval; 1383 1384 error = optval = 0; 1385 if (sopt->sopt_level != IPPROTO_IP) { 1386 return (EINVAL); 1387 } 1388 1389 switch (sopt->sopt_dir) { 1390 case SOPT_SET: 1391 switch (sopt->sopt_name) { 1392 case IP_OPTIONS: 1393#ifdef notyet 1394 case IP_RETOPTS: 1395#endif 1396 { 1397 struct mbuf *m; 1398 if (sopt->sopt_valsize > MLEN) { 1399 error = EMSGSIZE; 1400 break; 1401 } 1402 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_HEADER); 1403 if (m == 0) { 1404 error = ENOBUFS; 1405 break; 1406 } 1407 m->m_len = sopt->sopt_valsize; 1408 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 1409 m->m_len); 1410 1411 return (ip_pcbopts(sopt->sopt_name, &inp->inp_options, 1412 m)); 1413 } 1414 1415 case IP_TOS: 1416 case IP_TTL: 1417 case IP_RECVOPTS: 1418 case IP_RECVRETOPTS: 1419 case IP_RECVDSTADDR: 1420 case IP_RECVTTL: 1421 case IP_RECVIF: 1422 case IP_FAITH: 1423 case IP_ONESBCAST: 1424 error = sooptcopyin(sopt, &optval, sizeof optval, 1425 sizeof optval); 1426 if (error) 1427 break; 1428 1429 switch (sopt->sopt_name) { 1430 case IP_TOS: 1431 inp->inp_ip_tos = optval; 1432 break; 1433 1434 case IP_TTL: 1435 inp->inp_ip_ttl = optval; 1436 break; 1437#define OPTSET(bit) \ 1438 if (optval) \ 1439 inp->inp_flags |= bit; \ 1440 else \ 1441 inp->inp_flags &= ~bit; 1442 1443 case IP_RECVOPTS: 1444 OPTSET(INP_RECVOPTS); 1445 break; 1446 1447 case IP_RECVRETOPTS: 1448 OPTSET(INP_RECVRETOPTS); 1449 break; 1450 1451 case IP_RECVDSTADDR: 1452 OPTSET(INP_RECVDSTADDR); 1453 break; 1454 1455 case IP_RECVTTL: 1456 OPTSET(INP_RECVTTL); 1457 break; 1458 1459 case IP_RECVIF: 1460 OPTSET(INP_RECVIF); 1461 break; 1462 1463 case IP_FAITH: 1464 OPTSET(INP_FAITH); 1465 break; 1466 1467 case IP_ONESBCAST: 1468 OPTSET(INP_ONESBCAST); 1469 break; 1470 } 1471 break; 1472#undef OPTSET 1473 1474 case IP_MULTICAST_IF: 1475 case IP_MULTICAST_VIF: 1476 case IP_MULTICAST_TTL: 1477 case IP_MULTICAST_LOOP: 1478 case IP_ADD_MEMBERSHIP: 1479 case IP_DROP_MEMBERSHIP: 1480 error = ip_setmoptions(sopt, &inp->inp_moptions); 1481 break; 1482 1483 case IP_PORTRANGE: 1484 error = sooptcopyin(sopt, &optval, sizeof optval, 1485 sizeof optval); 1486 if (error) 1487 break; 1488 1489 switch (optval) { 1490 case IP_PORTRANGE_DEFAULT: 1491 inp->inp_flags &= ~(INP_LOWPORT); 1492 inp->inp_flags &= ~(INP_HIGHPORT); 1493 break; 1494 1495 case IP_PORTRANGE_HIGH: 1496 inp->inp_flags &= ~(INP_LOWPORT); 1497 inp->inp_flags |= INP_HIGHPORT; 1498 break; 1499 1500 case IP_PORTRANGE_LOW: 1501 inp->inp_flags &= ~(INP_HIGHPORT); 1502 inp->inp_flags |= INP_LOWPORT; 1503 break; 1504 1505 default: 1506 error = EINVAL; 1507 break; 1508 } 1509 break; 1510 1511#if defined(IPSEC) || defined(FAST_IPSEC) 1512 case IP_IPSEC_POLICY: 1513 { 1514 caddr_t req; 1515 size_t len = 0; 1516 int priv; 1517 struct mbuf *m; 1518 int optname; 1519 1520 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1521 break; 1522 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1523 break; 1524 priv = (sopt->sopt_td != NULL && 1525 suser(sopt->sopt_td) != 0) ? 0 : 1; 1526 req = mtod(m, caddr_t); 1527 len = m->m_len; 1528 optname = sopt->sopt_name; 1529 error = ipsec4_set_policy(inp, optname, req, len, priv); 1530 m_freem(m); 1531 break; 1532 } 1533#endif /*IPSEC*/ 1534 1535 default: 1536 error = ENOPROTOOPT; 1537 break; 1538 } 1539 break; 1540 1541 case SOPT_GET: 1542 switch (sopt->sopt_name) { 1543 case IP_OPTIONS: 1544 case IP_RETOPTS: 1545 if (inp->inp_options) 1546 error = sooptcopyout(sopt, 1547 mtod(inp->inp_options, 1548 char *), 1549 inp->inp_options->m_len); 1550 else 1551 sopt->sopt_valsize = 0; 1552 break; 1553 1554 case IP_TOS: 1555 case IP_TTL: 1556 case IP_RECVOPTS: 1557 case IP_RECVRETOPTS: 1558 case IP_RECVDSTADDR: 1559 case IP_RECVTTL: 1560 case IP_RECVIF: 1561 case IP_PORTRANGE: 1562 case IP_FAITH: 1563 case IP_ONESBCAST: 1564 switch (sopt->sopt_name) { 1565 1566 case IP_TOS: 1567 optval = inp->inp_ip_tos; 1568 break; 1569 1570 case IP_TTL: 1571 optval = inp->inp_ip_ttl; 1572 break; 1573 1574#define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1575 1576 case IP_RECVOPTS: 1577 optval = OPTBIT(INP_RECVOPTS); 1578 break; 1579 1580 case IP_RECVRETOPTS: 1581 optval = OPTBIT(INP_RECVRETOPTS); 1582 break; 1583 1584 case IP_RECVDSTADDR: 1585 optval = OPTBIT(INP_RECVDSTADDR); 1586 break; 1587 1588 case IP_RECVTTL: 1589 optval = OPTBIT(INP_RECVTTL); 1590 break; 1591 1592 case IP_RECVIF: 1593 optval = OPTBIT(INP_RECVIF); 1594 break; 1595 1596 case IP_PORTRANGE: 1597 if (inp->inp_flags & INP_HIGHPORT) 1598 optval = IP_PORTRANGE_HIGH; 1599 else if (inp->inp_flags & INP_LOWPORT) 1600 optval = IP_PORTRANGE_LOW; 1601 else 1602 optval = 0; 1603 break; 1604 1605 case IP_FAITH: 1606 optval = OPTBIT(INP_FAITH); 1607 break; 1608 1609 case IP_ONESBCAST: 1610 optval = OPTBIT(INP_ONESBCAST); 1611 break; 1612 } 1613 error = sooptcopyout(sopt, &optval, sizeof optval); 1614 break; 1615 1616 case IP_MULTICAST_IF: 1617 case IP_MULTICAST_VIF: 1618 case IP_MULTICAST_TTL: 1619 case IP_MULTICAST_LOOP: 1620 case IP_ADD_MEMBERSHIP: 1621 case IP_DROP_MEMBERSHIP: 1622 error = ip_getmoptions(sopt, inp->inp_moptions); 1623 break; 1624 1625#if defined(IPSEC) || defined(FAST_IPSEC) 1626 case IP_IPSEC_POLICY: 1627 { 1628 struct mbuf *m = NULL; 1629 caddr_t req = NULL; 1630 size_t len = 0; 1631 1632 if (m != 0) { 1633 req = mtod(m, caddr_t); 1634 len = m->m_len; 1635 } 1636 error = ipsec4_get_policy(sotoinpcb(so), req, len, &m); 1637 if (error == 0) 1638 error = soopt_mcopyout(sopt, m); /* XXX */ 1639 if (error == 0) 1640 m_freem(m); 1641 break; 1642 } 1643#endif /*IPSEC*/ 1644 1645 default: 1646 error = ENOPROTOOPT; 1647 break; 1648 } 1649 break; 1650 } 1651 return (error); 1652} 1653 1654/* 1655 * Set up IP options in pcb for insertion in output packets. 1656 * Store in mbuf with pointer in pcbopt, adding pseudo-option 1657 * with destination address if source routed. 1658 */ 1659static int 1660ip_pcbopts(optname, pcbopt, m) 1661 int optname; 1662 struct mbuf **pcbopt; 1663 register struct mbuf *m; 1664{ 1665 register int cnt, optlen; 1666 register u_char *cp; 1667 u_char opt; 1668 1669 /* turn off any old options */ 1670 if (*pcbopt) 1671 (void)m_free(*pcbopt); 1672 *pcbopt = 0; 1673 if (m == (struct mbuf *)0 || m->m_len == 0) { 1674 /* 1675 * Only turning off any previous options. 1676 */ 1677 if (m) 1678 (void)m_free(m); 1679 return (0); 1680 } 1681 1682 if (m->m_len % sizeof(int32_t)) 1683 goto bad; 1684 /* 1685 * IP first-hop destination address will be stored before 1686 * actual options; move other options back 1687 * and clear it when none present. 1688 */ 1689 if (m->m_data + m->m_len + sizeof(struct in_addr) >= &m->m_dat[MLEN]) 1690 goto bad; 1691 cnt = m->m_len; 1692 m->m_len += sizeof(struct in_addr); 1693 cp = mtod(m, u_char *) + sizeof(struct in_addr); 1694 bcopy(mtod(m, void *), cp, (unsigned)cnt); 1695 bzero(mtod(m, void *), sizeof(struct in_addr)); 1696 1697 for (; cnt > 0; cnt -= optlen, cp += optlen) { 1698 opt = cp[IPOPT_OPTVAL]; 1699 if (opt == IPOPT_EOL) 1700 break; 1701 if (opt == IPOPT_NOP) 1702 optlen = 1; 1703 else { 1704 if (cnt < IPOPT_OLEN + sizeof(*cp)) 1705 goto bad; 1706 optlen = cp[IPOPT_OLEN]; 1707 if (optlen < IPOPT_OLEN + sizeof(*cp) || optlen > cnt) 1708 goto bad; 1709 } 1710 switch (opt) { 1711 1712 default: 1713 break; 1714 1715 case IPOPT_LSRR: 1716 case IPOPT_SSRR: 1717 /* 1718 * user process specifies route as: 1719 * ->A->B->C->D 1720 * D must be our final destination (but we can't 1721 * check that since we may not have connected yet). 1722 * A is first hop destination, which doesn't appear in 1723 * actual IP option, but is stored before the options. 1724 */ 1725 if (optlen < IPOPT_MINOFF - 1 + sizeof(struct in_addr)) 1726 goto bad; 1727 m->m_len -= sizeof(struct in_addr); 1728 cnt -= sizeof(struct in_addr); 1729 optlen -= sizeof(struct in_addr); 1730 cp[IPOPT_OLEN] = optlen; 1731 /* 1732 * Move first hop before start of options. 1733 */ 1734 bcopy((caddr_t)&cp[IPOPT_OFFSET+1], mtod(m, caddr_t), 1735 sizeof(struct in_addr)); 1736 /* 1737 * Then copy rest of options back 1738 * to close up the deleted entry. 1739 */ 1740 bcopy((&cp[IPOPT_OFFSET+1] + sizeof(struct in_addr)), 1741 &cp[IPOPT_OFFSET+1], 1742 (unsigned)cnt + sizeof(struct in_addr)); 1743 break; 1744 } 1745 } 1746 if (m->m_len > MAX_IPOPTLEN + sizeof(struct in_addr)) 1747 goto bad; 1748 *pcbopt = m; 1749 return (0); 1750 1751bad: 1752 (void)m_free(m); 1753 return (EINVAL); 1754} 1755 1756/* 1757 * XXX 1758 * The whole multicast option thing needs to be re-thought. 1759 * Several of these options are equally applicable to non-multicast 1760 * transmission, and one (IP_MULTICAST_TTL) totally duplicates a 1761 * standard option (IP_TTL). 1762 */ 1763 1764/* 1765 * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index. 1766 */ 1767static struct ifnet * 1768ip_multicast_if(a, ifindexp) 1769 struct in_addr *a; 1770 int *ifindexp; 1771{ 1772 int ifindex; 1773 struct ifnet *ifp; 1774 1775 if (ifindexp) 1776 *ifindexp = 0; 1777 if (ntohl(a->s_addr) >> 24 == 0) { 1778 ifindex = ntohl(a->s_addr) & 0xffffff; 1779 if (ifindex < 0 || if_index < ifindex) 1780 return NULL; 1781 ifp = ifnet_byindex(ifindex); 1782 if (ifindexp) 1783 *ifindexp = ifindex; 1784 } else { 1785 INADDR_TO_IFP(*a, ifp); 1786 } 1787 return ifp; 1788} 1789 1790/* 1791 * Set the IP multicast options in response to user setsockopt(). 1792 */ 1793static int 1794ip_setmoptions(sopt, imop) 1795 struct sockopt *sopt; 1796 struct ip_moptions **imop; 1797{ 1798 int error = 0; 1799 int i; 1800 struct in_addr addr; 1801 struct ip_mreq mreq; 1802 struct ifnet *ifp; 1803 struct ip_moptions *imo = *imop; 1804 struct route ro; 1805 struct sockaddr_in *dst; 1806 int ifindex; 1807 int s; 1808 1809 if (imo == NULL) { 1810 /* 1811 * No multicast option buffer attached to the pcb; 1812 * allocate one and initialize to default values. 1813 */ 1814 imo = (struct ip_moptions*)malloc(sizeof(*imo), M_IPMOPTS, 1815 M_WAITOK); 1816 1817 if (imo == NULL) 1818 return (ENOBUFS); 1819 *imop = imo; 1820 imo->imo_multicast_ifp = NULL; 1821 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1822 imo->imo_multicast_vif = -1; 1823 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1824 imo->imo_multicast_loop = IP_DEFAULT_MULTICAST_LOOP; 1825 imo->imo_num_memberships = 0; 1826 } 1827 1828 switch (sopt->sopt_name) { 1829 /* store an index number for the vif you wanna use in the send */ 1830 case IP_MULTICAST_VIF: 1831 if (legal_vif_num == 0) { 1832 error = EOPNOTSUPP; 1833 break; 1834 } 1835 error = sooptcopyin(sopt, &i, sizeof i, sizeof i); 1836 if (error) 1837 break; 1838 if (!legal_vif_num(i) && (i != -1)) { 1839 error = EINVAL; 1840 break; 1841 } 1842 imo->imo_multicast_vif = i; 1843 break; 1844 1845 case IP_MULTICAST_IF: 1846 /* 1847 * Select the interface for outgoing multicast packets. 1848 */ 1849 error = sooptcopyin(sopt, &addr, sizeof addr, sizeof addr); 1850 if (error) 1851 break; 1852 /* 1853 * INADDR_ANY is used to remove a previous selection. 1854 * When no interface is selected, a default one is 1855 * chosen every time a multicast packet is sent. 1856 */ 1857 if (addr.s_addr == INADDR_ANY) { 1858 imo->imo_multicast_ifp = NULL; 1859 break; 1860 } 1861 /* 1862 * The selected interface is identified by its local 1863 * IP address. Find the interface and confirm that 1864 * it supports multicasting. 1865 */ 1866 s = splimp(); 1867 ifp = ip_multicast_if(&addr, &ifindex); 1868 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) { 1869 splx(s); 1870 error = EADDRNOTAVAIL; 1871 break; 1872 } 1873 imo->imo_multicast_ifp = ifp; 1874 if (ifindex) 1875 imo->imo_multicast_addr = addr; 1876 else 1877 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1878 splx(s); 1879 break; 1880 1881 case IP_MULTICAST_TTL: 1882 /* 1883 * Set the IP time-to-live for outgoing multicast packets. 1884 * The original multicast API required a char argument, 1885 * which is inconsistent with the rest of the socket API. 1886 * We allow either a char or an int. 1887 */ 1888 if (sopt->sopt_valsize == 1) { 1889 u_char ttl; 1890 error = sooptcopyin(sopt, &ttl, 1, 1); 1891 if (error) 1892 break; 1893 imo->imo_multicast_ttl = ttl; 1894 } else { 1895 u_int ttl; 1896 error = sooptcopyin(sopt, &ttl, sizeof ttl, 1897 sizeof ttl); 1898 if (error) 1899 break; 1900 if (ttl > 255) 1901 error = EINVAL; 1902 else 1903 imo->imo_multicast_ttl = ttl; 1904 } 1905 break; 1906 1907 case IP_MULTICAST_LOOP: 1908 /* 1909 * Set the loopback flag for outgoing multicast packets. 1910 * Must be zero or one. The original multicast API required a 1911 * char argument, which is inconsistent with the rest 1912 * of the socket API. We allow either a char or an int. 1913 */ 1914 if (sopt->sopt_valsize == 1) { 1915 u_char loop; 1916 error = sooptcopyin(sopt, &loop, 1, 1); 1917 if (error) 1918 break; 1919 imo->imo_multicast_loop = !!loop; 1920 } else { 1921 u_int loop; 1922 error = sooptcopyin(sopt, &loop, sizeof loop, 1923 sizeof loop); 1924 if (error) 1925 break; 1926 imo->imo_multicast_loop = !!loop; 1927 } 1928 break; 1929 1930 case IP_ADD_MEMBERSHIP: 1931 /* 1932 * Add a multicast group membership. 1933 * Group must be a valid IP multicast address. 1934 */ 1935 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq); 1936 if (error) 1937 break; 1938 1939 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) { 1940 error = EINVAL; 1941 break; 1942 } 1943 s = splimp(); 1944 /* 1945 * If no interface address was provided, use the interface of 1946 * the route to the given multicast address. 1947 */ 1948 if (mreq.imr_interface.s_addr == INADDR_ANY) { 1949 bzero((caddr_t)&ro, sizeof(ro)); 1950 dst = (struct sockaddr_in *)&ro.ro_dst; 1951 dst->sin_len = sizeof(*dst); 1952 dst->sin_family = AF_INET; 1953 dst->sin_addr = mreq.imr_multiaddr; 1954 rtalloc_ign(&ro, RTF_CLONING); 1955 if (ro.ro_rt == NULL) { 1956 error = EADDRNOTAVAIL; 1957 splx(s); 1958 break; 1959 } 1960 ifp = ro.ro_rt->rt_ifp; 1961 RTFREE(ro.ro_rt); 1962 } 1963 else { 1964 ifp = ip_multicast_if(&mreq.imr_interface, NULL); 1965 } 1966 1967 /* 1968 * See if we found an interface, and confirm that it 1969 * supports multicast. 1970 */ 1971 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) { 1972 error = EADDRNOTAVAIL; 1973 splx(s); 1974 break; 1975 } 1976 /* 1977 * See if the membership already exists or if all the 1978 * membership slots are full. 1979 */ 1980 for (i = 0; i < imo->imo_num_memberships; ++i) { 1981 if (imo->imo_membership[i]->inm_ifp == ifp && 1982 imo->imo_membership[i]->inm_addr.s_addr 1983 == mreq.imr_multiaddr.s_addr) 1984 break; 1985 } 1986 if (i < imo->imo_num_memberships) { 1987 error = EADDRINUSE; 1988 splx(s); 1989 break; 1990 } 1991 if (i == IP_MAX_MEMBERSHIPS) { 1992 error = ETOOMANYREFS; 1993 splx(s); 1994 break; 1995 } 1996 /* 1997 * Everything looks good; add a new record to the multicast 1998 * address list for the given interface. 1999 */ 2000 if ((imo->imo_membership[i] = 2001 in_addmulti(&mreq.imr_multiaddr, ifp)) == NULL) { 2002 error = ENOBUFS; 2003 splx(s); 2004 break; 2005 } 2006 ++imo->imo_num_memberships; 2007 splx(s); 2008 break; 2009 2010 case IP_DROP_MEMBERSHIP: 2011 /* 2012 * Drop a multicast group membership. 2013 * Group must be a valid IP multicast address. 2014 */ 2015 error = sooptcopyin(sopt, &mreq, sizeof mreq, sizeof mreq); 2016 if (error) 2017 break; 2018 2019 if (!IN_MULTICAST(ntohl(mreq.imr_multiaddr.s_addr))) { 2020 error = EINVAL; 2021 break; 2022 } 2023 2024 s = splimp(); 2025 /* 2026 * If an interface address was specified, get a pointer 2027 * to its ifnet structure. 2028 */ 2029 if (mreq.imr_interface.s_addr == INADDR_ANY) 2030 ifp = NULL; 2031 else { 2032 ifp = ip_multicast_if(&mreq.imr_interface, NULL); 2033 if (ifp == NULL) { 2034 error = EADDRNOTAVAIL; 2035 splx(s); 2036 break; 2037 } 2038 } 2039 /* 2040 * Find the membership in the membership array. 2041 */ 2042 for (i = 0; i < imo->imo_num_memberships; ++i) { 2043 if ((ifp == NULL || 2044 imo->imo_membership[i]->inm_ifp == ifp) && 2045 imo->imo_membership[i]->inm_addr.s_addr == 2046 mreq.imr_multiaddr.s_addr) 2047 break; 2048 } 2049 if (i == imo->imo_num_memberships) { 2050 error = EADDRNOTAVAIL; 2051 splx(s); 2052 break; 2053 } 2054 /* 2055 * Give up the multicast address record to which the 2056 * membership points. 2057 */ 2058 in_delmulti(imo->imo_membership[i]); 2059 /* 2060 * Remove the gap in the membership array. 2061 */ 2062 for (++i; i < imo->imo_num_memberships; ++i) 2063 imo->imo_membership[i-1] = imo->imo_membership[i]; 2064 --imo->imo_num_memberships; 2065 splx(s); 2066 break; 2067 2068 default: 2069 error = EOPNOTSUPP; 2070 break; 2071 } 2072 2073 /* 2074 * If all options have default values, no need to keep the mbuf. 2075 */ 2076 if (imo->imo_multicast_ifp == NULL && 2077 imo->imo_multicast_vif == -1 && 2078 imo->imo_multicast_ttl == IP_DEFAULT_MULTICAST_TTL && 2079 imo->imo_multicast_loop == IP_DEFAULT_MULTICAST_LOOP && 2080 imo->imo_num_memberships == 0) { 2081 free(*imop, M_IPMOPTS); 2082 *imop = NULL; 2083 } 2084 2085 return (error); 2086} 2087 2088/* 2089 * Return the IP multicast options in response to user getsockopt(). 2090 */ 2091static int 2092ip_getmoptions(sopt, imo) 2093 struct sockopt *sopt; 2094 register struct ip_moptions *imo; 2095{ 2096 struct in_addr addr; 2097 struct in_ifaddr *ia; 2098 int error, optval; 2099 u_char coptval; 2100 2101 error = 0; 2102 switch (sopt->sopt_name) { 2103 case IP_MULTICAST_VIF: 2104 if (imo != NULL) 2105 optval = imo->imo_multicast_vif; 2106 else 2107 optval = -1; 2108 error = sooptcopyout(sopt, &optval, sizeof optval); 2109 break; 2110 2111 case IP_MULTICAST_IF: 2112 if (imo == NULL || imo->imo_multicast_ifp == NULL) 2113 addr.s_addr = INADDR_ANY; 2114 else if (imo->imo_multicast_addr.s_addr) { 2115 /* return the value user has set */ 2116 addr = imo->imo_multicast_addr; 2117 } else { 2118 IFP_TO_IA(imo->imo_multicast_ifp, ia); 2119 addr.s_addr = (ia == NULL) ? INADDR_ANY 2120 : IA_SIN(ia)->sin_addr.s_addr; 2121 } 2122 error = sooptcopyout(sopt, &addr, sizeof addr); 2123 break; 2124 2125 case IP_MULTICAST_TTL: 2126 if (imo == 0) 2127 optval = coptval = IP_DEFAULT_MULTICAST_TTL; 2128 else 2129 optval = coptval = imo->imo_multicast_ttl; 2130 if (sopt->sopt_valsize == 1) 2131 error = sooptcopyout(sopt, &coptval, 1); 2132 else 2133 error = sooptcopyout(sopt, &optval, sizeof optval); 2134 break; 2135 2136 case IP_MULTICAST_LOOP: 2137 if (imo == 0) 2138 optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 2139 else 2140 optval = coptval = imo->imo_multicast_loop; 2141 if (sopt->sopt_valsize == 1) 2142 error = sooptcopyout(sopt, &coptval, 1); 2143 else 2144 error = sooptcopyout(sopt, &optval, sizeof optval); 2145 break; 2146 2147 default: 2148 error = ENOPROTOOPT; 2149 break; 2150 } 2151 return (error); 2152} 2153 2154/* 2155 * Discard the IP multicast options. 2156 */ 2157void 2158ip_freemoptions(imo) 2159 register struct ip_moptions *imo; 2160{ 2161 register int i; 2162 2163 if (imo != NULL) { 2164 for (i = 0; i < imo->imo_num_memberships; ++i) 2165 in_delmulti(imo->imo_membership[i]); 2166 free(imo, M_IPMOPTS); 2167 } 2168} 2169 2170/* 2171 * Routine called from ip_output() to loop back a copy of an IP multicast 2172 * packet to the input queue of a specified interface. Note that this 2173 * calls the output routine of the loopback "driver", but with an interface 2174 * pointer that might NOT be a loopback interface -- evil, but easier than 2175 * replicating that code here. 2176 */ 2177static void 2178ip_mloopback(ifp, m, dst, hlen) 2179 struct ifnet *ifp; 2180 register struct mbuf *m; 2181 register struct sockaddr_in *dst; 2182 int hlen; 2183{ 2184 register struct ip *ip; 2185 struct mbuf *copym; 2186 2187 copym = m_copy(m, 0, M_COPYALL); 2188 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 2189 copym = m_pullup(copym, hlen); 2190 if (copym != NULL) { 2191 /* If needed, compute the checksum and mark it as valid. */ 2192 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 2193 in_delayed_cksum(copym); 2194 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 2195 copym->m_pkthdr.csum_flags |= 2196 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 2197 copym->m_pkthdr.csum_data = 0xffff; 2198 } 2199 /* 2200 * We don't bother to fragment if the IP length is greater 2201 * than the interface's MTU. Can this possibly matter? 2202 */ 2203 ip = mtod(copym, struct ip *); 2204 ip->ip_len = htons(ip->ip_len); 2205 ip->ip_off = htons(ip->ip_off); 2206 ip->ip_sum = 0; 2207 ip->ip_sum = in_cksum(copym, hlen); 2208 /* 2209 * NB: 2210 * It's not clear whether there are any lingering 2211 * reentrancy problems in other areas which might 2212 * be exposed by using ip_input directly (in 2213 * particular, everything which modifies the packet 2214 * in-place). Yet another option is using the 2215 * protosw directly to deliver the looped back 2216 * packet. For the moment, we'll err on the side 2217 * of safety by using if_simloop(). 2218 */ 2219#if 1 /* XXX */ 2220 if (dst->sin_family != AF_INET) { 2221 printf("ip_mloopback: bad address family %d\n", 2222 dst->sin_family); 2223 dst->sin_family = AF_INET; 2224 } 2225#endif 2226 2227#ifdef notdef 2228 copym->m_pkthdr.rcvif = ifp; 2229 ip_input(copym); 2230#else 2231 if_simloop(ifp, copym, dst->sin_family, 0); 2232#endif 2233 } 2234} 2235