ip_output.c revision 194076
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: head/sys/netinet/ip_output.c 194076 2009-06-12 20:46:36Z bz $"); 34 35#include "opt_ipfw.h" 36#include "opt_ipsec.h" 37#include "opt_route.h" 38#include "opt_mbuf_stress_test.h" 39#include "opt_mpath.h" 40#include "opt_sctp.h" 41 42#include <sys/param.h> 43#include <sys/systm.h> 44#include <sys/kernel.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/priv.h> 48#include <sys/proc.h> 49#include <sys/protosw.h> 50#include <sys/socket.h> 51#include <sys/socketvar.h> 52#include <sys/sysctl.h> 53#include <sys/ucred.h> 54#include <sys/vimage.h> 55 56#include <net/if.h> 57#include <net/netisr.h> 58#include <net/pfil.h> 59#include <net/route.h> 60#include <net/flowtable.h> 61#ifdef RADIX_MPATH 62#include <net/radix_mpath.h> 63#endif 64#include <net/vnet.h> 65 66#include <netinet/in.h> 67#include <netinet/in_systm.h> 68#include <netinet/ip.h> 69#include <netinet/in_pcb.h> 70#include <netinet/in_var.h> 71#include <netinet/ip_var.h> 72#include <netinet/ip_options.h> 73#include <netinet/vinet.h> 74#ifdef SCTP 75#include <netinet/sctp.h> 76#include <netinet/sctp_crc32.h> 77#endif 78 79#ifdef IPSEC 80#include <netinet/ip_ipsec.h> 81#include <netipsec/ipsec.h> 82#endif /* IPSEC*/ 83 84#include <machine/in_cksum.h> 85 86#include <security/mac/mac_framework.h> 87 88#define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 89 x, (ntohl(a.s_addr)>>24)&0xFF,\ 90 (ntohl(a.s_addr)>>16)&0xFF,\ 91 (ntohl(a.s_addr)>>8)&0xFF,\ 92 (ntohl(a.s_addr))&0xFF, y); 93 94#ifdef VIMAGE_GLOBALS 95u_short ip_id; 96#endif 97 98#ifdef MBUF_STRESS_TEST 99int mbuf_frag_size = 0; 100SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 101 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 102#endif 103 104static void ip_mloopback 105 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 106 107 108extern int in_mcast_loop; 109extern struct protosw inetsw[]; 110 111/* 112 * IP output. The packet in mbuf chain m contains a skeletal IP 113 * header (with len, off, ttl, proto, tos, src, dst). 114 * The mbuf chain containing the packet will be freed. 115 * The mbuf opt, if present, will not be freed. 116 * In the IP forwarding case, the packet will arrive with options already 117 * inserted, so must have a NULL opt pointer. 118 */ 119int 120ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, 121 struct ip_moptions *imo, struct inpcb *inp) 122{ 123 INIT_VNET_NET(curvnet); 124 INIT_VNET_INET(curvnet); 125 struct ip *ip; 126 struct ifnet *ifp = NULL; /* keep compiler happy */ 127 struct mbuf *m0; 128 int hlen = sizeof (struct ip); 129 int mtu; 130 int len, error = 0; 131 int nortfree = 0; 132 struct sockaddr_in *dst = NULL; /* keep compiler happy */ 133 struct in_ifaddr *ia = NULL; 134 int isbroadcast, sw_csum; 135 struct route iproute; 136 struct in_addr odst; 137#ifdef IPFIREWALL_FORWARD 138 struct m_tag *fwd_tag = NULL; 139#endif 140#ifdef IPSEC 141 int no_route_but_check_spd = 0; 142#endif 143 M_ASSERTPKTHDR(m); 144 145 if (inp != NULL) { 146 INP_LOCK_ASSERT(inp); 147 M_SETFIB(m, inp->inp_inc.inc_fibnum); 148 if (inp->inp_flags & (INP_HW_FLOWID|INP_SW_FLOWID)) { 149 m->m_pkthdr.flowid = inp->inp_flowid; 150 m->m_flags |= M_FLOWID; 151 } 152 } 153 154 if (ro == NULL) { 155 ro = &iproute; 156 bzero(ro, sizeof (*ro)); 157 158#ifdef FLOWTABLE 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#endif 168 } 169 170 if (opt) { 171 len = 0; 172 m = ip_insertoptions(m, opt, &len); 173 if (len != 0) 174 hlen = len; 175 } 176 ip = mtod(m, struct ip *); 177 178 /* 179 * Fill in IP header. If we are not allowing fragmentation, 180 * then the ip_id field is meaningless, but we don't set it 181 * to zero. Doing so causes various problems when devices along 182 * the path (routers, load balancers, firewalls, etc.) illegally 183 * disable DF on our packet. Note that a 16-bit counter 184 * will wrap around in less than 10 seconds at 100 Mbit/s on a 185 * medium with MTU 1500. See Steven M. Bellovin, "A Technique 186 * for Counting NATted Hosts", Proc. IMW'02, available at 187 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>. 188 */ 189 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 190 ip->ip_v = IPVERSION; 191 ip->ip_hl = hlen >> 2; 192 ip->ip_id = ip_newid(); 193 IPSTAT_INC(ips_localout); 194 } else { 195 hlen = ip->ip_hl << 2; 196 } 197 198 dst = (struct sockaddr_in *)&ro->ro_dst; 199again: 200 /* 201 * If there is a cached route, 202 * check that it is to the same destination 203 * and is still up. If not, free it and try again. 204 * The address family should also be checked in case of sharing the 205 * cache with IPv6. 206 */ 207 if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 208 dst->sin_family != AF_INET || 209 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 210 if (!nortfree) 211 RTFREE(ro->ro_rt); 212 ro->ro_rt = (struct rtentry *)NULL; 213 } 214#ifdef IPFIREWALL_FORWARD 215 if (ro->ro_rt == NULL && fwd_tag == NULL) { 216#else 217 if (ro->ro_rt == NULL) { 218#endif 219 bzero(dst, sizeof(*dst)); 220 dst->sin_family = AF_INET; 221 dst->sin_len = sizeof(*dst); 222 dst->sin_addr = ip->ip_dst; 223 } 224 /* 225 * If routing to interface only, short circuit routing lookup. 226 * The use of an all-ones broadcast address implies this; an 227 * interface is specified by the broadcast address of an interface, 228 * or the destination address of a ptp interface. 229 */ 230 if (flags & IP_SENDONES) { 231 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 232 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 233 IPSTAT_INC(ips_noroute); 234 error = ENETUNREACH; 235 goto bad; 236 } 237 ip->ip_dst.s_addr = INADDR_BROADCAST; 238 dst->sin_addr = ip->ip_dst; 239 ifp = ia->ia_ifp; 240 ip->ip_ttl = 1; 241 isbroadcast = 1; 242 } else if (flags & IP_ROUTETOIF) { 243 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 244 (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) { 245 IPSTAT_INC(ips_noroute); 246 error = ENETUNREACH; 247 goto bad; 248 } 249 ifp = ia->ia_ifp; 250 ip->ip_ttl = 1; 251 isbroadcast = in_broadcast(dst->sin_addr, ifp); 252 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 253 imo != NULL && imo->imo_multicast_ifp != NULL) { 254 /* 255 * Bypass the normal routing lookup for multicast 256 * packets if the interface is specified. 257 */ 258 ifp = imo->imo_multicast_ifp; 259 IFP_TO_IA(ifp, ia); 260 isbroadcast = 0; /* fool gcc */ 261 } else { 262 /* 263 * We want to do any cloning requested by the link layer, 264 * as this is probably required in all cases for correct 265 * operation (as it is for ARP). 266 */ 267 if (ro->ro_rt == NULL) 268#ifdef RADIX_MPATH 269 rtalloc_mpath_fib(ro, 270 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 271 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 272#else 273 in_rtalloc_ign(ro, 0, 274 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 275#endif 276 if (ro->ro_rt == NULL) { 277#ifdef IPSEC 278 /* 279 * There is no route for this packet, but it is 280 * possible that a matching SPD entry exists. 281 */ 282 no_route_but_check_spd = 1; 283 mtu = 0; /* Silence GCC warning. */ 284 goto sendit; 285#endif 286 IPSTAT_INC(ips_noroute); 287 error = EHOSTUNREACH; 288 goto bad; 289 } 290 ia = ifatoia(ro->ro_rt->rt_ifa); 291 ifp = ro->ro_rt->rt_ifp; 292 ro->ro_rt->rt_rmx.rmx_pksent++; 293 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 294 dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 295 if (ro->ro_rt->rt_flags & RTF_HOST) 296 isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 297 else 298 isbroadcast = in_broadcast(dst->sin_addr, ifp); 299 } 300 /* 301 * Calculate MTU. If we have a route that is up, use that, 302 * otherwise use the interface's MTU. 303 */ 304 if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) { 305 /* 306 * This case can happen if the user changed the MTU 307 * of an interface after enabling IP on it. Because 308 * most netifs don't keep track of routes pointing to 309 * them, there is no way for one to update all its 310 * routes when the MTU is changed. 311 */ 312 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu) 313 ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 314 mtu = ro->ro_rt->rt_rmx.rmx_mtu; 315 } else { 316 mtu = ifp->if_mtu; 317 } 318 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 319 m->m_flags |= M_MCAST; 320 /* 321 * IP destination address is multicast. Make sure "dst" 322 * still points to the address in "ro". (It may have been 323 * changed to point to a gateway address, above.) 324 */ 325 dst = (struct sockaddr_in *)&ro->ro_dst; 326 /* 327 * See if the caller provided any multicast options 328 */ 329 if (imo != NULL) { 330 ip->ip_ttl = imo->imo_multicast_ttl; 331 if (imo->imo_multicast_vif != -1) 332 ip->ip_src.s_addr = 333 ip_mcast_src ? 334 ip_mcast_src(imo->imo_multicast_vif) : 335 INADDR_ANY; 336 } else 337 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 338 /* 339 * Confirm that the outgoing interface supports multicast. 340 */ 341 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 342 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 343 IPSTAT_INC(ips_noroute); 344 error = ENETUNREACH; 345 goto bad; 346 } 347 } 348 /* 349 * If source address not specified yet, use address 350 * of outgoing interface. 351 */ 352 if (ip->ip_src.s_addr == INADDR_ANY) { 353 /* Interface may have no addresses. */ 354 if (ia != NULL) 355 ip->ip_src = IA_SIN(ia)->sin_addr; 356 } 357 358 if ((imo == NULL && in_mcast_loop) || 359 (imo && imo->imo_multicast_loop)) { 360 /* 361 * Loop back multicast datagram if not expressly 362 * forbidden to do so, even if we are not a member 363 * of the group; ip_input() will filter it later, 364 * thus deferring a hash lookup and mutex acquisition 365 * at the expense of a cheap copy using m_copym(). 366 */ 367 ip_mloopback(ifp, m, dst, hlen); 368 } else { 369 /* 370 * If we are acting as a multicast router, perform 371 * multicast forwarding as if the packet had just 372 * arrived on the interface to which we are about 373 * to send. The multicast forwarding function 374 * recursively calls this function, using the 375 * IP_FORWARDING flag to prevent infinite recursion. 376 * 377 * Multicasts that are looped back by ip_mloopback(), 378 * above, will be forwarded by the ip_input() routine, 379 * if necessary. 380 */ 381 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 382 /* 383 * If rsvp daemon is not running, do not 384 * set ip_moptions. This ensures that the packet 385 * is multicast and not just sent down one link 386 * as prescribed by rsvpd. 387 */ 388 if (!V_rsvp_on) 389 imo = NULL; 390 if (ip_mforward && 391 ip_mforward(ip, ifp, m, imo) != 0) { 392 m_freem(m); 393 goto done; 394 } 395 } 396 } 397 398 /* 399 * Multicasts with a time-to-live of zero may be looped- 400 * back, above, but must not be transmitted on a network. 401 * Also, multicasts addressed to the loopback interface 402 * are not sent -- the above call to ip_mloopback() will 403 * loop back a copy. ip_input() will drop the copy if 404 * this host does not belong to the destination group on 405 * the loopback interface. 406 */ 407 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 408 m_freem(m); 409 goto done; 410 } 411 412 goto sendit; 413 } 414 415 /* 416 * If the source address is not specified yet, use the address 417 * of the outoing interface. 418 */ 419 if (ip->ip_src.s_addr == INADDR_ANY) { 420 /* Interface may have no addresses. */ 421 if (ia != NULL) { 422 ip->ip_src = IA_SIN(ia)->sin_addr; 423 } 424 } 425 426 /* 427 * Verify that we have any chance at all of being able to queue the 428 * packet or packet fragments, unless ALTQ is enabled on the given 429 * interface in which case packetdrop should be done by queueing. 430 */ 431#ifdef ALTQ 432 if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) && 433 ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 434 ifp->if_snd.ifq_maxlen)) 435#else 436 if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 437 ifp->if_snd.ifq_maxlen) 438#endif /* ALTQ */ 439 { 440 error = ENOBUFS; 441 IPSTAT_INC(ips_odropped); 442 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1); 443 goto bad; 444 } 445 446 /* 447 * Look for broadcast address and 448 * verify user is allowed to send 449 * such a packet. 450 */ 451 if (isbroadcast) { 452 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 453 error = EADDRNOTAVAIL; 454 goto bad; 455 } 456 if ((flags & IP_ALLOWBROADCAST) == 0) { 457 error = EACCES; 458 goto bad; 459 } 460 /* don't allow broadcast messages to be fragmented */ 461 if (ip->ip_len > mtu) { 462 error = EMSGSIZE; 463 goto bad; 464 } 465 m->m_flags |= M_BCAST; 466 } else { 467 m->m_flags &= ~M_BCAST; 468 } 469 470sendit: 471#ifdef IPSEC 472 switch(ip_ipsec_output(&m, inp, &flags, &error, &ifp)) { 473 case 1: 474 goto bad; 475 case -1: 476 goto done; 477 case 0: 478 default: 479 break; /* Continue with packet processing. */ 480 } 481 /* 482 * Check if there was a route for this packet; return error if not. 483 */ 484 if (no_route_but_check_spd) { 485 IPSTAT_INC(ips_noroute); 486 error = EHOSTUNREACH; 487 goto bad; 488 } 489 /* Update variables that are affected by ipsec4_output(). */ 490 ip = mtod(m, struct ip *); 491 hlen = ip->ip_hl << 2; 492#endif /* IPSEC */ 493 494 /* Jump over all PFIL processing if hooks are not active. */ 495 if (!PFIL_HOOKED(&inet_pfil_hook)) 496 goto passout; 497 498 /* Run through list of hooks for output packets. */ 499 odst.s_addr = ip->ip_dst.s_addr; 500 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 501 if (error != 0 || m == NULL) 502 goto done; 503 504 ip = mtod(m, struct ip *); 505 506 /* See if destination IP address was changed by packet filter. */ 507 if (odst.s_addr != ip->ip_dst.s_addr) { 508 m->m_flags |= M_SKIP_FIREWALL; 509 /* If destination is now ourself drop to ip_input(). */ 510 if (in_localip(ip->ip_dst)) { 511 m->m_flags |= M_FASTFWD_OURS; 512 if (m->m_pkthdr.rcvif == NULL) 513 m->m_pkthdr.rcvif = V_loif; 514 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 515 m->m_pkthdr.csum_flags |= 516 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 517 m->m_pkthdr.csum_data = 0xffff; 518 } 519 m->m_pkthdr.csum_flags |= 520 CSUM_IP_CHECKED | CSUM_IP_VALID; 521#ifdef SCTP 522 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 523 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 524#endif 525 error = netisr_queue(NETISR_IP, m); 526 goto done; 527 } else 528 goto again; /* Redo the routing table lookup. */ 529 } 530 531#ifdef IPFIREWALL_FORWARD 532 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 533 if (m->m_flags & M_FASTFWD_OURS) { 534 if (m->m_pkthdr.rcvif == NULL) 535 m->m_pkthdr.rcvif = V_loif; 536 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 537 m->m_pkthdr.csum_flags |= 538 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 539 m->m_pkthdr.csum_data = 0xffff; 540 } 541#ifdef SCTP 542 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 543 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 544#endif 545 m->m_pkthdr.csum_flags |= 546 CSUM_IP_CHECKED | CSUM_IP_VALID; 547 548 error = netisr_queue(NETISR_IP, m); 549 goto done; 550 } 551 /* Or forward to some other address? */ 552 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 553 if (fwd_tag) { 554 dst = (struct sockaddr_in *)&ro->ro_dst; 555 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 556 m->m_flags |= M_SKIP_FIREWALL; 557 m_tag_delete(m, fwd_tag); 558 goto again; 559 } 560#endif /* IPFIREWALL_FORWARD */ 561 562passout: 563 /* 127/8 must not appear on wire - RFC1122. */ 564 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 565 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 566 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 567 IPSTAT_INC(ips_badaddr); 568 error = EADDRNOTAVAIL; 569 goto bad; 570 } 571 } 572 573 m->m_pkthdr.csum_flags |= CSUM_IP; 574 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 575 if (sw_csum & CSUM_DELAY_DATA) { 576 in_delayed_cksum(m); 577 sw_csum &= ~CSUM_DELAY_DATA; 578 } 579#ifdef SCTP 580 if (sw_csum & CSUM_SCTP) { 581 sctp_delayed_cksum(m); 582 sw_csum &= ~CSUM_SCTP; 583 } 584#endif 585 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 586 587 /* 588 * If small enough for interface, or the interface will take 589 * care of the fragmentation for us, we can just send directly. 590 */ 591 if (ip->ip_len <= mtu || 592 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 593 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 594 ip->ip_len = htons(ip->ip_len); 595 ip->ip_off = htons(ip->ip_off); 596 ip->ip_sum = 0; 597 if (sw_csum & CSUM_DELAY_IP) 598 ip->ip_sum = in_cksum(m, hlen); 599 600 /* 601 * Record statistics for this interface address. 602 * With CSUM_TSO the byte/packet count will be slightly 603 * incorrect because we count the IP+TCP headers only 604 * once instead of for every generated packet. 605 */ 606 if (!(flags & IP_FORWARDING) && ia) { 607 if (m->m_pkthdr.csum_flags & CSUM_TSO) 608 ia->ia_ifa.if_opackets += 609 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 610 else 611 ia->ia_ifa.if_opackets++; 612 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 613 } 614#ifdef MBUF_STRESS_TEST 615 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 616 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 617#endif 618 /* 619 * Reset layer specific mbuf flags 620 * to avoid confusing lower layers. 621 */ 622 m->m_flags &= ~(M_PROTOFLAGS); 623 error = (*ifp->if_output)(ifp, m, 624 (struct sockaddr *)dst, ro); 625 goto done; 626 } 627 628 /* Balk when DF bit is set or the interface didn't support TSO. */ 629 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 630 error = EMSGSIZE; 631 IPSTAT_INC(ips_cantfrag); 632 goto bad; 633 } 634 635 /* 636 * Too large for interface; fragment if possible. If successful, 637 * on return, m will point to a list of packets to be sent. 638 */ 639 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 640 if (error) 641 goto bad; 642 for (; m; m = m0) { 643 m0 = m->m_nextpkt; 644 m->m_nextpkt = 0; 645 if (error == 0) { 646 /* Record statistics for this interface address. */ 647 if (ia != NULL) { 648 ia->ia_ifa.if_opackets++; 649 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 650 } 651 /* 652 * Reset layer specific mbuf flags 653 * to avoid confusing upper layers. 654 */ 655 m->m_flags &= ~(M_PROTOFLAGS); 656 657 error = (*ifp->if_output)(ifp, m, 658 (struct sockaddr *)dst, ro); 659 } else 660 m_freem(m); 661 } 662 663 if (error == 0) 664 IPSTAT_INC(ips_fragmented); 665 666done: 667 if (ro == &iproute && ro->ro_rt && !nortfree) { 668 RTFREE(ro->ro_rt); 669 } 670 return (error); 671bad: 672 m_freem(m); 673 goto done; 674} 675 676/* 677 * Create a chain of fragments which fit the given mtu. m_frag points to the 678 * mbuf to be fragmented; on return it points to the chain with the fragments. 679 * Return 0 if no error. If error, m_frag may contain a partially built 680 * chain of fragments that should be freed by the caller. 681 * 682 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 683 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 684 */ 685int 686ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 687 u_long if_hwassist_flags, int sw_csum) 688{ 689 INIT_VNET_INET(curvnet); 690 int error = 0; 691 int hlen = ip->ip_hl << 2; 692 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 693 int off; 694 struct mbuf *m0 = *m_frag; /* the original packet */ 695 int firstlen; 696 struct mbuf **mnext; 697 int nfrags; 698 699 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 700 IPSTAT_INC(ips_cantfrag); 701 return EMSGSIZE; 702 } 703 704 /* 705 * Must be able to put at least 8 bytes per fragment. 706 */ 707 if (len < 8) 708 return EMSGSIZE; 709 710 /* 711 * If the interface will not calculate checksums on 712 * fragmented packets, then do it here. 713 */ 714 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 715 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 716 in_delayed_cksum(m0); 717 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 718 } 719#ifdef SCTP 720 if (m0->m_pkthdr.csum_flags & CSUM_SCTP && 721 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 722 sctp_delayed_cksum(m0); 723 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 724 } 725#endif 726 if (len > PAGE_SIZE) { 727 /* 728 * Fragment large datagrams such that each segment 729 * contains a multiple of PAGE_SIZE amount of data, 730 * plus headers. This enables a receiver to perform 731 * page-flipping zero-copy optimizations. 732 * 733 * XXX When does this help given that sender and receiver 734 * could have different page sizes, and also mtu could 735 * be less than the receiver's page size ? 736 */ 737 int newlen; 738 struct mbuf *m; 739 740 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 741 off += m->m_len; 742 743 /* 744 * firstlen (off - hlen) must be aligned on an 745 * 8-byte boundary 746 */ 747 if (off < hlen) 748 goto smart_frag_failure; 749 off = ((off - hlen) & ~7) + hlen; 750 newlen = (~PAGE_MASK) & mtu; 751 if ((newlen + sizeof (struct ip)) > mtu) { 752 /* we failed, go back the default */ 753smart_frag_failure: 754 newlen = len; 755 off = hlen + len; 756 } 757 len = newlen; 758 759 } else { 760 off = hlen + len; 761 } 762 763 firstlen = off - hlen; 764 mnext = &m0->m_nextpkt; /* pointer to next packet */ 765 766 /* 767 * Loop through length of segment after first fragment, 768 * make new header and copy data of each part and link onto chain. 769 * Here, m0 is the original packet, m is the fragment being created. 770 * The fragments are linked off the m_nextpkt of the original 771 * packet, which after processing serves as the first fragment. 772 */ 773 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 774 struct ip *mhip; /* ip header on the fragment */ 775 struct mbuf *m; 776 int mhlen = sizeof (struct ip); 777 778 MGETHDR(m, M_DONTWAIT, MT_DATA); 779 if (m == NULL) { 780 error = ENOBUFS; 781 IPSTAT_INC(ips_odropped); 782 goto done; 783 } 784 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 785 /* 786 * In the first mbuf, leave room for the link header, then 787 * copy the original IP header including options. The payload 788 * goes into an additional mbuf chain returned by m_copym(). 789 */ 790 m->m_data += max_linkhdr; 791 mhip = mtod(m, struct ip *); 792 *mhip = *ip; 793 if (hlen > sizeof (struct ip)) { 794 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 795 mhip->ip_v = IPVERSION; 796 mhip->ip_hl = mhlen >> 2; 797 } 798 m->m_len = mhlen; 799 /* XXX do we need to add ip->ip_off below ? */ 800 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 801 if (off + len >= ip->ip_len) { /* last fragment */ 802 len = ip->ip_len - off; 803 m->m_flags |= M_LASTFRAG; 804 } else 805 mhip->ip_off |= IP_MF; 806 mhip->ip_len = htons((u_short)(len + mhlen)); 807 m->m_next = m_copym(m0, off, len, M_DONTWAIT); 808 if (m->m_next == NULL) { /* copy failed */ 809 m_free(m); 810 error = ENOBUFS; /* ??? */ 811 IPSTAT_INC(ips_odropped); 812 goto done; 813 } 814 m->m_pkthdr.len = mhlen + len; 815 m->m_pkthdr.rcvif = NULL; 816#ifdef MAC 817 mac_netinet_fragment(m0, m); 818#endif 819 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 820 mhip->ip_off = htons(mhip->ip_off); 821 mhip->ip_sum = 0; 822 if (sw_csum & CSUM_DELAY_IP) 823 mhip->ip_sum = in_cksum(m, mhlen); 824 *mnext = m; 825 mnext = &m->m_nextpkt; 826 } 827 IPSTAT_ADD(ips_ofragments, nfrags); 828 829 /* set first marker for fragment chain */ 830 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 831 m0->m_pkthdr.csum_data = nfrags; 832 833 /* 834 * Update first fragment by trimming what's been copied out 835 * and updating header. 836 */ 837 m_adj(m0, hlen + firstlen - ip->ip_len); 838 m0->m_pkthdr.len = hlen + firstlen; 839 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 840 ip->ip_off |= IP_MF; 841 ip->ip_off = htons(ip->ip_off); 842 ip->ip_sum = 0; 843 if (sw_csum & CSUM_DELAY_IP) 844 ip->ip_sum = in_cksum(m0, hlen); 845 846done: 847 *m_frag = m0; 848 return error; 849} 850 851void 852in_delayed_cksum(struct mbuf *m) 853{ 854 struct ip *ip; 855 u_short csum, offset; 856 857 ip = mtod(m, struct ip *); 858 offset = ip->ip_hl << 2 ; 859 csum = in_cksum_skip(m, ip->ip_len, offset); 860 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 861 csum = 0xffff; 862 offset += m->m_pkthdr.csum_data; /* checksum offset */ 863 864 if (offset + sizeof(u_short) > m->m_len) { 865 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 866 m->m_len, offset, ip->ip_p); 867 /* 868 * XXX 869 * this shouldn't happen, but if it does, the 870 * correct behavior may be to insert the checksum 871 * in the appropriate next mbuf in the chain. 872 */ 873 return; 874 } 875 *(u_short *)(m->m_data + offset) = csum; 876} 877 878/* 879 * IP socket option processing. 880 */ 881int 882ip_ctloutput(struct socket *so, struct sockopt *sopt) 883{ 884 struct inpcb *inp = sotoinpcb(so); 885 int error, optval; 886 887 error = optval = 0; 888 if (sopt->sopt_level != IPPROTO_IP) { 889 if ((sopt->sopt_level == SOL_SOCKET) && 890 (sopt->sopt_name == SO_SETFIB)) { 891 inp->inp_inc.inc_fibnum = so->so_fibnum; 892 return (0); 893 } 894 return (EINVAL); 895 } 896 897 switch (sopt->sopt_dir) { 898 case SOPT_SET: 899 switch (sopt->sopt_name) { 900 case IP_OPTIONS: 901#ifdef notyet 902 case IP_RETOPTS: 903#endif 904 { 905 struct mbuf *m; 906 if (sopt->sopt_valsize > MLEN) { 907 error = EMSGSIZE; 908 break; 909 } 910 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 911 if (m == NULL) { 912 error = ENOBUFS; 913 break; 914 } 915 m->m_len = sopt->sopt_valsize; 916 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 917 m->m_len); 918 if (error) { 919 m_free(m); 920 break; 921 } 922 INP_WLOCK(inp); 923 error = ip_pcbopts(inp, sopt->sopt_name, m); 924 INP_WUNLOCK(inp); 925 return (error); 926 } 927 928 case IP_BINDANY: 929 if (sopt->sopt_td != NULL) { 930 error = priv_check(sopt->sopt_td, 931 PRIV_NETINET_BINDANY); 932 if (error) 933 break; 934 } 935 /* FALLTHROUGH */ 936 case IP_TOS: 937 case IP_TTL: 938 case IP_MINTTL: 939 case IP_RECVOPTS: 940 case IP_RECVRETOPTS: 941 case IP_RECVDSTADDR: 942 case IP_RECVTTL: 943 case IP_RECVIF: 944 case IP_FAITH: 945 case IP_ONESBCAST: 946 case IP_DONTFRAG: 947 error = sooptcopyin(sopt, &optval, sizeof optval, 948 sizeof optval); 949 if (error) 950 break; 951 952 switch (sopt->sopt_name) { 953 case IP_TOS: 954 inp->inp_ip_tos = optval; 955 break; 956 957 case IP_TTL: 958 inp->inp_ip_ttl = optval; 959 break; 960 961 case IP_MINTTL: 962 if (optval >= 0 && optval <= MAXTTL) 963 inp->inp_ip_minttl = optval; 964 else 965 error = EINVAL; 966 break; 967 968#define OPTSET(bit) do { \ 969 INP_WLOCK(inp); \ 970 if (optval) \ 971 inp->inp_flags |= bit; \ 972 else \ 973 inp->inp_flags &= ~bit; \ 974 INP_WUNLOCK(inp); \ 975} while (0) 976 977 case IP_RECVOPTS: 978 OPTSET(INP_RECVOPTS); 979 break; 980 981 case IP_RECVRETOPTS: 982 OPTSET(INP_RECVRETOPTS); 983 break; 984 985 case IP_RECVDSTADDR: 986 OPTSET(INP_RECVDSTADDR); 987 break; 988 989 case IP_RECVTTL: 990 OPTSET(INP_RECVTTL); 991 break; 992 993 case IP_RECVIF: 994 OPTSET(INP_RECVIF); 995 break; 996 997 case IP_FAITH: 998 OPTSET(INP_FAITH); 999 break; 1000 1001 case IP_ONESBCAST: 1002 OPTSET(INP_ONESBCAST); 1003 break; 1004 case IP_DONTFRAG: 1005 OPTSET(INP_DONTFRAG); 1006 break; 1007 case IP_BINDANY: 1008 OPTSET(INP_BINDANY); 1009 break; 1010 } 1011 break; 1012#undef OPTSET 1013 1014 /* 1015 * Multicast socket options are processed by the in_mcast 1016 * module. 1017 */ 1018 case IP_MULTICAST_IF: 1019 case IP_MULTICAST_VIF: 1020 case IP_MULTICAST_TTL: 1021 case IP_MULTICAST_LOOP: 1022 case IP_ADD_MEMBERSHIP: 1023 case IP_DROP_MEMBERSHIP: 1024 case IP_ADD_SOURCE_MEMBERSHIP: 1025 case IP_DROP_SOURCE_MEMBERSHIP: 1026 case IP_BLOCK_SOURCE: 1027 case IP_UNBLOCK_SOURCE: 1028 case IP_MSFILTER: 1029 case MCAST_JOIN_GROUP: 1030 case MCAST_LEAVE_GROUP: 1031 case MCAST_JOIN_SOURCE_GROUP: 1032 case MCAST_LEAVE_SOURCE_GROUP: 1033 case MCAST_BLOCK_SOURCE: 1034 case MCAST_UNBLOCK_SOURCE: 1035 error = inp_setmoptions(inp, sopt); 1036 break; 1037 1038 case IP_PORTRANGE: 1039 error = sooptcopyin(sopt, &optval, sizeof optval, 1040 sizeof optval); 1041 if (error) 1042 break; 1043 1044 INP_WLOCK(inp); 1045 switch (optval) { 1046 case IP_PORTRANGE_DEFAULT: 1047 inp->inp_flags &= ~(INP_LOWPORT); 1048 inp->inp_flags &= ~(INP_HIGHPORT); 1049 break; 1050 1051 case IP_PORTRANGE_HIGH: 1052 inp->inp_flags &= ~(INP_LOWPORT); 1053 inp->inp_flags |= INP_HIGHPORT; 1054 break; 1055 1056 case IP_PORTRANGE_LOW: 1057 inp->inp_flags &= ~(INP_HIGHPORT); 1058 inp->inp_flags |= INP_LOWPORT; 1059 break; 1060 1061 default: 1062 error = EINVAL; 1063 break; 1064 } 1065 INP_WUNLOCK(inp); 1066 break; 1067 1068#ifdef IPSEC 1069 case IP_IPSEC_POLICY: 1070 { 1071 caddr_t req; 1072 struct mbuf *m; 1073 1074 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1075 break; 1076 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1077 break; 1078 req = mtod(m, caddr_t); 1079 error = ipsec_set_policy(inp, sopt->sopt_name, req, 1080 m->m_len, (sopt->sopt_td != NULL) ? 1081 sopt->sopt_td->td_ucred : NULL); 1082 m_freem(m); 1083 break; 1084 } 1085#endif /* IPSEC */ 1086 1087 default: 1088 error = ENOPROTOOPT; 1089 break; 1090 } 1091 break; 1092 1093 case SOPT_GET: 1094 switch (sopt->sopt_name) { 1095 case IP_OPTIONS: 1096 case IP_RETOPTS: 1097 if (inp->inp_options) 1098 error = sooptcopyout(sopt, 1099 mtod(inp->inp_options, 1100 char *), 1101 inp->inp_options->m_len); 1102 else 1103 sopt->sopt_valsize = 0; 1104 break; 1105 1106 case IP_TOS: 1107 case IP_TTL: 1108 case IP_MINTTL: 1109 case IP_RECVOPTS: 1110 case IP_RECVRETOPTS: 1111 case IP_RECVDSTADDR: 1112 case IP_RECVTTL: 1113 case IP_RECVIF: 1114 case IP_PORTRANGE: 1115 case IP_FAITH: 1116 case IP_ONESBCAST: 1117 case IP_DONTFRAG: 1118 switch (sopt->sopt_name) { 1119 1120 case IP_TOS: 1121 optval = inp->inp_ip_tos; 1122 break; 1123 1124 case IP_TTL: 1125 optval = inp->inp_ip_ttl; 1126 break; 1127 1128 case IP_MINTTL: 1129 optval = inp->inp_ip_minttl; 1130 break; 1131 1132#define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1133 1134 case IP_RECVOPTS: 1135 optval = OPTBIT(INP_RECVOPTS); 1136 break; 1137 1138 case IP_RECVRETOPTS: 1139 optval = OPTBIT(INP_RECVRETOPTS); 1140 break; 1141 1142 case IP_RECVDSTADDR: 1143 optval = OPTBIT(INP_RECVDSTADDR); 1144 break; 1145 1146 case IP_RECVTTL: 1147 optval = OPTBIT(INP_RECVTTL); 1148 break; 1149 1150 case IP_RECVIF: 1151 optval = OPTBIT(INP_RECVIF); 1152 break; 1153 1154 case IP_PORTRANGE: 1155 if (inp->inp_flags & INP_HIGHPORT) 1156 optval = IP_PORTRANGE_HIGH; 1157 else if (inp->inp_flags & INP_LOWPORT) 1158 optval = IP_PORTRANGE_LOW; 1159 else 1160 optval = 0; 1161 break; 1162 1163 case IP_FAITH: 1164 optval = OPTBIT(INP_FAITH); 1165 break; 1166 1167 case IP_ONESBCAST: 1168 optval = OPTBIT(INP_ONESBCAST); 1169 break; 1170 case IP_DONTFRAG: 1171 optval = OPTBIT(INP_DONTFRAG); 1172 break; 1173 } 1174 error = sooptcopyout(sopt, &optval, sizeof optval); 1175 break; 1176 1177 /* 1178 * Multicast socket options are processed by the in_mcast 1179 * module. 1180 */ 1181 case IP_MULTICAST_IF: 1182 case IP_MULTICAST_VIF: 1183 case IP_MULTICAST_TTL: 1184 case IP_MULTICAST_LOOP: 1185 case IP_MSFILTER: 1186 error = inp_getmoptions(inp, sopt); 1187 break; 1188 1189#ifdef IPSEC 1190 case IP_IPSEC_POLICY: 1191 { 1192 struct mbuf *m = NULL; 1193 caddr_t req = NULL; 1194 size_t len = 0; 1195 1196 if (m != 0) { 1197 req = mtod(m, caddr_t); 1198 len = m->m_len; 1199 } 1200 error = ipsec_get_policy(sotoinpcb(so), req, len, &m); 1201 if (error == 0) 1202 error = soopt_mcopyout(sopt, m); /* XXX */ 1203 if (error == 0) 1204 m_freem(m); 1205 break; 1206 } 1207#endif /* IPSEC */ 1208 1209 default: 1210 error = ENOPROTOOPT; 1211 break; 1212 } 1213 break; 1214 } 1215 return (error); 1216} 1217 1218/* 1219 * Routine called from ip_output() to loop back a copy of an IP multicast 1220 * packet to the input queue of a specified interface. Note that this 1221 * calls the output routine of the loopback "driver", but with an interface 1222 * pointer that might NOT be a loopback interface -- evil, but easier than 1223 * replicating that code here. 1224 */ 1225static void 1226ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1227 int hlen) 1228{ 1229 register struct ip *ip; 1230 struct mbuf *copym; 1231 1232 /* 1233 * Make a deep copy of the packet because we're going to 1234 * modify the pack in order to generate checksums. 1235 */ 1236 copym = m_dup(m, M_DONTWAIT); 1237 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1238 copym = m_pullup(copym, hlen); 1239 if (copym != NULL) { 1240 /* If needed, compute the checksum and mark it as valid. */ 1241 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1242 in_delayed_cksum(copym); 1243 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1244 copym->m_pkthdr.csum_flags |= 1245 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1246 copym->m_pkthdr.csum_data = 0xffff; 1247 } 1248 /* 1249 * We don't bother to fragment if the IP length is greater 1250 * than the interface's MTU. Can this possibly matter? 1251 */ 1252 ip = mtod(copym, struct ip *); 1253 ip->ip_len = htons(ip->ip_len); 1254 ip->ip_off = htons(ip->ip_off); 1255 ip->ip_sum = 0; 1256 ip->ip_sum = in_cksum(copym, hlen); 1257#if 1 /* XXX */ 1258 if (dst->sin_family != AF_INET) { 1259 printf("ip_mloopback: bad address family %d\n", 1260 dst->sin_family); 1261 dst->sin_family = AF_INET; 1262 } 1263#endif 1264 if_simloop(ifp, copym, dst->sin_family, 0); 1265 } 1266} 1267