ip_output.c revision 189106
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 189106 2009-02-27 14:12:05Z bz $"); 34 35#include "opt_ipfw.h" 36#include "opt_inet.h" 37#include "opt_ipsec.h" 38#include "opt_route.h" 39#include "opt_mac.h" 40#include "opt_mbuf_stress_test.h" 41#include "opt_mpath.h" 42#include "opt_sctp.h" 43 44#include <sys/param.h> 45#include <sys/systm.h> 46#include <sys/kernel.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#include <sys/priv.h> 50#include <sys/proc.h> 51#include <sys/protosw.h> 52#include <sys/socket.h> 53#include <sys/socketvar.h> 54#include <sys/sysctl.h> 55#include <sys/ucred.h> 56#include <sys/vimage.h> 57 58#include <net/if.h> 59#include <net/netisr.h> 60#include <net/pfil.h> 61#include <net/route.h> 62#ifdef RADIX_MPATH 63#include <net/radix_mpath.h> 64#endif 65#include <net/vnet.h> 66 67#include <netinet/in.h> 68#include <netinet/in_systm.h> 69#include <netinet/ip.h> 70#include <netinet/in_pcb.h> 71#include <netinet/in_var.h> 72#include <netinet/ip_var.h> 73#include <netinet/ip_options.h> 74#include <netinet/vinet.h> 75#ifdef SCTP 76#include <netinet/sctp.h> 77#include <netinet/sctp_crc32.h> 78#endif 79 80#ifdef IPSEC 81#include <netinet/ip_ipsec.h> 82#include <netipsec/ipsec.h> 83#endif /* IPSEC*/ 84 85#include <machine/in_cksum.h> 86 87#include <security/mac/mac_framework.h> 88 89#define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 90 x, (ntohl(a.s_addr)>>24)&0xFF,\ 91 (ntohl(a.s_addr)>>16)&0xFF,\ 92 (ntohl(a.s_addr)>>8)&0xFF,\ 93 (ntohl(a.s_addr))&0xFF, y); 94 95#ifdef VIMAGE_GLOBALS 96u_short ip_id; 97#endif 98 99#ifdef MBUF_STRESS_TEST 100int mbuf_frag_size = 0; 101SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 102 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 103#endif 104 105#if defined(IP_NONLOCALBIND) 106static int ip_nonlocalok = 0; 107SYSCTL_INT(_net_inet_ip, OID_AUTO, nonlocalok, 108 CTLFLAG_RW|CTLFLAG_SECURE, &ip_nonlocalok, 0, ""); 109#endif 110 111static void ip_mloopback 112 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 113 114 115extern struct protosw inetsw[]; 116 117/* 118 * IP output. The packet in mbuf chain m contains a skeletal IP 119 * header (with len, off, ttl, proto, tos, src, dst). 120 * The mbuf chain containing the packet will be freed. 121 * The mbuf opt, if present, will not be freed. 122 * In the IP forwarding case, the packet will arrive with options already 123 * inserted, so must have a NULL opt pointer. 124 */ 125int 126ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, 127 struct ip_moptions *imo, struct inpcb *inp) 128{ 129 INIT_VNET_NET(curvnet); 130 INIT_VNET_INET(curvnet); 131 struct ip *ip; 132 struct ifnet *ifp = NULL; /* keep compiler happy */ 133 struct mbuf *m0; 134 int hlen = sizeof (struct ip); 135 int mtu; 136 int len, error = 0; 137 struct sockaddr_in *dst = NULL; /* keep compiler happy */ 138 struct in_ifaddr *ia = NULL; 139 int isbroadcast, sw_csum; 140 struct route iproute; 141 struct in_addr odst; 142#ifdef IPFIREWALL_FORWARD 143 struct m_tag *fwd_tag = NULL; 144#endif 145 M_ASSERTPKTHDR(m); 146 147 if (ro == NULL) { 148 ro = &iproute; 149 bzero(ro, sizeof (*ro)); 150 } 151 152 if (inp != NULL) { 153 M_SETFIB(m, inp->inp_inc.inc_fibnum); 154 INP_LOCK_ASSERT(inp); 155 } 156 157 if (opt) { 158 len = 0; 159 m = ip_insertoptions(m, opt, &len); 160 if (len != 0) 161 hlen = len; 162 } 163 ip = mtod(m, struct ip *); 164 165 /* 166 * Fill in IP header. If we are not allowing fragmentation, 167 * then the ip_id field is meaningless, but we don't set it 168 * to zero. Doing so causes various problems when devices along 169 * the path (routers, load balancers, firewalls, etc.) illegally 170 * disable DF on our packet. Note that a 16-bit counter 171 * will wrap around in less than 10 seconds at 100 Mbit/s on a 172 * medium with MTU 1500. See Steven M. Bellovin, "A Technique 173 * for Counting NATted Hosts", Proc. IMW'02, available at 174 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>. 175 */ 176 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 177 ip->ip_v = IPVERSION; 178 ip->ip_hl = hlen >> 2; 179 ip->ip_id = ip_newid(); 180 V_ipstat.ips_localout++; 181 } else { 182 hlen = ip->ip_hl << 2; 183 } 184 185 dst = (struct sockaddr_in *)&ro->ro_dst; 186again: 187 /* 188 * If there is a cached route, 189 * check that it is to the same destination 190 * and is still up. If not, free it and try again. 191 * The address family should also be checked in case of sharing the 192 * cache with IPv6. 193 */ 194 if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 195 dst->sin_family != AF_INET || 196 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 197 RTFREE(ro->ro_rt); 198 ro->ro_rt = (struct rtentry *)NULL; 199 } 200#ifdef IPFIREWALL_FORWARD 201 if (ro->ro_rt == NULL && fwd_tag == NULL) { 202#else 203 if (ro->ro_rt == NULL) { 204#endif 205 bzero(dst, sizeof(*dst)); 206 dst->sin_family = AF_INET; 207 dst->sin_len = sizeof(*dst); 208 dst->sin_addr = ip->ip_dst; 209 } 210 /* 211 * If routing to interface only, short circuit routing lookup. 212 * The use of an all-ones broadcast address implies this; an 213 * interface is specified by the broadcast address of an interface, 214 * or the destination address of a ptp interface. 215 */ 216 if (flags & IP_SENDONES) { 217 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 218 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 219 V_ipstat.ips_noroute++; 220 error = ENETUNREACH; 221 goto bad; 222 } 223 ip->ip_dst.s_addr = INADDR_BROADCAST; 224 dst->sin_addr = ip->ip_dst; 225 ifp = ia->ia_ifp; 226 ip->ip_ttl = 1; 227 isbroadcast = 1; 228 } else if (flags & IP_ROUTETOIF) { 229 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 230 (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) { 231 V_ipstat.ips_noroute++; 232 error = ENETUNREACH; 233 goto bad; 234 } 235 ifp = ia->ia_ifp; 236 ip->ip_ttl = 1; 237 isbroadcast = in_broadcast(dst->sin_addr, ifp); 238 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 239 imo != NULL && imo->imo_multicast_ifp != NULL) { 240 /* 241 * Bypass the normal routing lookup for multicast 242 * packets if the interface is specified. 243 */ 244 ifp = imo->imo_multicast_ifp; 245 IFP_TO_IA(ifp, ia); 246 isbroadcast = 0; /* fool gcc */ 247 } else { 248 /* 249 * We want to do any cloning requested by the link layer, 250 * as this is probably required in all cases for correct 251 * operation (as it is for ARP). 252 */ 253 if (ro->ro_rt == NULL) 254#ifdef RADIX_MPATH 255 rtalloc_mpath_fib(ro, 256 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 257 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 258#else 259 in_rtalloc_ign(ro, 0, 260 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 261#endif 262 if (ro->ro_rt == NULL) { 263 V_ipstat.ips_noroute++; 264 error = EHOSTUNREACH; 265 goto bad; 266 } 267 ia = ifatoia(ro->ro_rt->rt_ifa); 268 ifp = ro->ro_rt->rt_ifp; 269 ro->ro_rt->rt_rmx.rmx_pksent++; 270 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 271 dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 272 if (ro->ro_rt->rt_flags & RTF_HOST) 273 isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 274 else 275 isbroadcast = in_broadcast(dst->sin_addr, ifp); 276 } 277 /* 278 * Calculate MTU. If we have a route that is up, use that, 279 * otherwise use the interface's MTU. 280 */ 281 if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) { 282 /* 283 * This case can happen if the user changed the MTU 284 * of an interface after enabling IP on it. Because 285 * most netifs don't keep track of routes pointing to 286 * them, there is no way for one to update all its 287 * routes when the MTU is changed. 288 */ 289 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu) 290 ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 291 mtu = ro->ro_rt->rt_rmx.rmx_mtu; 292 } else { 293 mtu = ifp->if_mtu; 294 } 295 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 296 struct in_multi *inm; 297 298 m->m_flags |= M_MCAST; 299 /* 300 * IP destination address is multicast. Make sure "dst" 301 * still points to the address in "ro". (It may have been 302 * changed to point to a gateway address, above.) 303 */ 304 dst = (struct sockaddr_in *)&ro->ro_dst; 305 /* 306 * See if the caller provided any multicast options 307 */ 308 if (imo != NULL) { 309 ip->ip_ttl = imo->imo_multicast_ttl; 310 if (imo->imo_multicast_vif != -1) 311 ip->ip_src.s_addr = 312 ip_mcast_src ? 313 ip_mcast_src(imo->imo_multicast_vif) : 314 INADDR_ANY; 315 } else 316 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 317 /* 318 * Confirm that the outgoing interface supports multicast. 319 */ 320 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 321 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 322 V_ipstat.ips_noroute++; 323 error = ENETUNREACH; 324 goto bad; 325 } 326 } 327 /* 328 * If source address not specified yet, use address 329 * of outgoing interface. 330 */ 331 if (ip->ip_src.s_addr == INADDR_ANY) { 332 /* Interface may have no addresses. */ 333 if (ia != NULL) 334 ip->ip_src = IA_SIN(ia)->sin_addr; 335 } 336 337 IN_MULTI_LOCK(); 338 IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm); 339 if (inm != NULL && 340 (imo == NULL || imo->imo_multicast_loop)) { 341 IN_MULTI_UNLOCK(); 342 /* 343 * If we belong to the destination multicast group 344 * on the outgoing interface, and the caller did not 345 * forbid loopback, loop back a copy. 346 */ 347 ip_mloopback(ifp, m, dst, hlen); 348 } 349 else { 350 IN_MULTI_UNLOCK(); 351 /* 352 * If we are acting as a multicast router, perform 353 * multicast forwarding as if the packet had just 354 * arrived on the interface to which we are about 355 * to send. The multicast forwarding function 356 * recursively calls this function, using the 357 * IP_FORWARDING flag to prevent infinite recursion. 358 * 359 * Multicasts that are looped back by ip_mloopback(), 360 * above, will be forwarded by the ip_input() routine, 361 * if necessary. 362 */ 363 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 364 /* 365 * If rsvp daemon is not running, do not 366 * set ip_moptions. This ensures that the packet 367 * is multicast and not just sent down one link 368 * as prescribed by rsvpd. 369 */ 370 if (!V_rsvp_on) 371 imo = NULL; 372 if (ip_mforward && 373 ip_mforward(ip, ifp, m, imo) != 0) { 374 m_freem(m); 375 goto done; 376 } 377 } 378 } 379 380 /* 381 * Multicasts with a time-to-live of zero may be looped- 382 * back, above, but must not be transmitted on a network. 383 * Also, multicasts addressed to the loopback interface 384 * are not sent -- the above call to ip_mloopback() will 385 * loop back a copy if this host actually belongs to the 386 * destination group on the loopback interface. 387 */ 388 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 389 m_freem(m); 390 goto done; 391 } 392 393 goto sendit; 394 } 395 396 /* 397 * If the source address is not specified yet, use the address 398 * of the outoing interface. 399 */ 400 if (ip->ip_src.s_addr == INADDR_ANY) { 401 /* Interface may have no addresses. */ 402 if (ia != NULL) { 403 ip->ip_src = IA_SIN(ia)->sin_addr; 404 } 405 } 406 407 /* 408 * Verify that we have any chance at all of being able to queue the 409 * packet or packet fragments, unless ALTQ is enabled on the given 410 * interface in which case packetdrop should be done by queueing. 411 */ 412#ifdef ALTQ 413 if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) && 414 ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 415 ifp->if_snd.ifq_maxlen)) 416#else 417 if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 418 ifp->if_snd.ifq_maxlen) 419#endif /* ALTQ */ 420 { 421 error = ENOBUFS; 422 V_ipstat.ips_odropped++; 423 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1); 424 goto bad; 425 } 426 427 /* 428 * Look for broadcast address and 429 * verify user is allowed to send 430 * such a packet. 431 */ 432 if (isbroadcast) { 433 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 434 error = EADDRNOTAVAIL; 435 goto bad; 436 } 437 if ((flags & IP_ALLOWBROADCAST) == 0) { 438 error = EACCES; 439 goto bad; 440 } 441 /* don't allow broadcast messages to be fragmented */ 442 if (ip->ip_len > mtu) { 443 error = EMSGSIZE; 444 goto bad; 445 } 446 m->m_flags |= M_BCAST; 447 } else { 448 m->m_flags &= ~M_BCAST; 449 } 450 451sendit: 452#ifdef IPSEC 453 switch(ip_ipsec_output(&m, inp, &flags, &error, &ro, &iproute, &dst, &ia, &ifp)) { 454 case 1: 455 goto bad; 456 case -1: 457 goto done; 458 case 0: 459 default: 460 break; /* Continue with packet processing. */ 461 } 462 /* Update variables that are affected by ipsec4_output(). */ 463 ip = mtod(m, struct ip *); 464 hlen = ip->ip_hl << 2; 465#endif /* IPSEC */ 466 467 /* Jump over all PFIL processing if hooks are not active. */ 468 if (!PFIL_HOOKED(&inet_pfil_hook)) 469 goto passout; 470 471 /* Run through list of hooks for output packets. */ 472 odst.s_addr = ip->ip_dst.s_addr; 473 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 474 if (error != 0 || m == NULL) 475 goto done; 476 477 ip = mtod(m, struct ip *); 478 479 /* See if destination IP address was changed by packet filter. */ 480 if (odst.s_addr != ip->ip_dst.s_addr) { 481 m->m_flags |= M_SKIP_FIREWALL; 482 /* If destination is now ourself drop to ip_input(). */ 483 if (in_localip(ip->ip_dst)) { 484 m->m_flags |= M_FASTFWD_OURS; 485 if (m->m_pkthdr.rcvif == NULL) 486 m->m_pkthdr.rcvif = V_loif; 487 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 488 m->m_pkthdr.csum_flags |= 489 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 490 m->m_pkthdr.csum_data = 0xffff; 491 } 492 m->m_pkthdr.csum_flags |= 493 CSUM_IP_CHECKED | CSUM_IP_VALID; 494#ifdef SCTP 495 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 496 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 497#endif 498 error = netisr_queue(NETISR_IP, m); 499 goto done; 500 } else 501 goto again; /* Redo the routing table lookup. */ 502 } 503 504#ifdef IPFIREWALL_FORWARD 505 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 506 if (m->m_flags & M_FASTFWD_OURS) { 507 if (m->m_pkthdr.rcvif == NULL) 508 m->m_pkthdr.rcvif = V_loif; 509 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 510 m->m_pkthdr.csum_flags |= 511 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 512 m->m_pkthdr.csum_data = 0xffff; 513 } 514#ifdef SCTP 515 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 516 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 517#endif 518 m->m_pkthdr.csum_flags |= 519 CSUM_IP_CHECKED | CSUM_IP_VALID; 520 521 error = netisr_queue(NETISR_IP, m); 522 goto done; 523 } 524 /* Or forward to some other address? */ 525 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 526 if (fwd_tag) { 527 dst = (struct sockaddr_in *)&ro->ro_dst; 528 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 529 m->m_flags |= M_SKIP_FIREWALL; 530 m_tag_delete(m, fwd_tag); 531 goto again; 532 } 533#endif /* IPFIREWALL_FORWARD */ 534 535passout: 536 /* 127/8 must not appear on wire - RFC1122. */ 537 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 538 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 539 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 540 V_ipstat.ips_badaddr++; 541 error = EADDRNOTAVAIL; 542 goto bad; 543 } 544 } 545 546 m->m_pkthdr.csum_flags |= CSUM_IP; 547 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 548 if (sw_csum & CSUM_DELAY_DATA) { 549 in_delayed_cksum(m); 550 sw_csum &= ~CSUM_DELAY_DATA; 551 } 552#ifdef SCTP 553 if (sw_csum & CSUM_SCTP) { 554 sctp_delayed_cksum(m); 555 sw_csum &= ~CSUM_SCTP; 556 } 557#endif 558 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 559 560 /* 561 * If small enough for interface, or the interface will take 562 * care of the fragmentation for us, we can just send directly. 563 */ 564 if (ip->ip_len <= mtu || 565 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 566 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 567 ip->ip_len = htons(ip->ip_len); 568 ip->ip_off = htons(ip->ip_off); 569 ip->ip_sum = 0; 570 if (sw_csum & CSUM_DELAY_IP) 571 ip->ip_sum = in_cksum(m, hlen); 572 573 /* 574 * Record statistics for this interface address. 575 * With CSUM_TSO the byte/packet count will be slightly 576 * incorrect because we count the IP+TCP headers only 577 * once instead of for every generated packet. 578 */ 579 if (!(flags & IP_FORWARDING) && ia) { 580 if (m->m_pkthdr.csum_flags & CSUM_TSO) 581 ia->ia_ifa.if_opackets += 582 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 583 else 584 ia->ia_ifa.if_opackets++; 585 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 586 } 587#ifdef MBUF_STRESS_TEST 588 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 589 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 590#endif 591 /* 592 * Reset layer specific mbuf flags 593 * to avoid confusing lower layers. 594 */ 595 m->m_flags &= ~(M_PROTOFLAGS); 596 error = (*ifp->if_output)(ifp, m, 597 (struct sockaddr *)dst, ro->ro_rt); 598 goto done; 599 } 600 601 /* Balk when DF bit is set or the interface didn't support TSO. */ 602 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 603 error = EMSGSIZE; 604 V_ipstat.ips_cantfrag++; 605 goto bad; 606 } 607 608 /* 609 * Too large for interface; fragment if possible. If successful, 610 * on return, m will point to a list of packets to be sent. 611 */ 612 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 613 if (error) 614 goto bad; 615 for (; m; m = m0) { 616 m0 = m->m_nextpkt; 617 m->m_nextpkt = 0; 618 if (error == 0) { 619 /* Record statistics for this interface address. */ 620 if (ia != NULL) { 621 ia->ia_ifa.if_opackets++; 622 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 623 } 624 /* 625 * Reset layer specific mbuf flags 626 * to avoid confusing upper layers. 627 */ 628 m->m_flags &= ~(M_PROTOFLAGS); 629 630 error = (*ifp->if_output)(ifp, m, 631 (struct sockaddr *)dst, ro->ro_rt); 632 } else 633 m_freem(m); 634 } 635 636 if (error == 0) 637 V_ipstat.ips_fragmented++; 638 639done: 640 if (ro == &iproute && ro->ro_rt) { 641 RTFREE(ro->ro_rt); 642 } 643 return (error); 644bad: 645 m_freem(m); 646 goto done; 647} 648 649/* 650 * Create a chain of fragments which fit the given mtu. m_frag points to the 651 * mbuf to be fragmented; on return it points to the chain with the fragments. 652 * Return 0 if no error. If error, m_frag may contain a partially built 653 * chain of fragments that should be freed by the caller. 654 * 655 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 656 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 657 */ 658int 659ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 660 u_long if_hwassist_flags, int sw_csum) 661{ 662 INIT_VNET_INET(curvnet); 663 int error = 0; 664 int hlen = ip->ip_hl << 2; 665 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 666 int off; 667 struct mbuf *m0 = *m_frag; /* the original packet */ 668 int firstlen; 669 struct mbuf **mnext; 670 int nfrags; 671 672 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 673 V_ipstat.ips_cantfrag++; 674 return EMSGSIZE; 675 } 676 677 /* 678 * Must be able to put at least 8 bytes per fragment. 679 */ 680 if (len < 8) 681 return EMSGSIZE; 682 683 /* 684 * If the interface will not calculate checksums on 685 * fragmented packets, then do it here. 686 */ 687 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 688 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 689 in_delayed_cksum(m0); 690 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 691 } 692#ifdef SCTP 693 if (m0->m_pkthdr.csum_flags & CSUM_SCTP && 694 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 695 sctp_delayed_cksum(m0); 696 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 697 } 698#endif 699 if (len > PAGE_SIZE) { 700 /* 701 * Fragment large datagrams such that each segment 702 * contains a multiple of PAGE_SIZE amount of data, 703 * plus headers. This enables a receiver to perform 704 * page-flipping zero-copy optimizations. 705 * 706 * XXX When does this help given that sender and receiver 707 * could have different page sizes, and also mtu could 708 * be less than the receiver's page size ? 709 */ 710 int newlen; 711 struct mbuf *m; 712 713 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 714 off += m->m_len; 715 716 /* 717 * firstlen (off - hlen) must be aligned on an 718 * 8-byte boundary 719 */ 720 if (off < hlen) 721 goto smart_frag_failure; 722 off = ((off - hlen) & ~7) + hlen; 723 newlen = (~PAGE_MASK) & mtu; 724 if ((newlen + sizeof (struct ip)) > mtu) { 725 /* we failed, go back the default */ 726smart_frag_failure: 727 newlen = len; 728 off = hlen + len; 729 } 730 len = newlen; 731 732 } else { 733 off = hlen + len; 734 } 735 736 firstlen = off - hlen; 737 mnext = &m0->m_nextpkt; /* pointer to next packet */ 738 739 /* 740 * Loop through length of segment after first fragment, 741 * make new header and copy data of each part and link onto chain. 742 * Here, m0 is the original packet, m is the fragment being created. 743 * The fragments are linked off the m_nextpkt of the original 744 * packet, which after processing serves as the first fragment. 745 */ 746 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 747 struct ip *mhip; /* ip header on the fragment */ 748 struct mbuf *m; 749 int mhlen = sizeof (struct ip); 750 751 MGETHDR(m, M_DONTWAIT, MT_DATA); 752 if (m == NULL) { 753 error = ENOBUFS; 754 V_ipstat.ips_odropped++; 755 goto done; 756 } 757 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 758 /* 759 * In the first mbuf, leave room for the link header, then 760 * copy the original IP header including options. The payload 761 * goes into an additional mbuf chain returned by m_copy(). 762 */ 763 m->m_data += max_linkhdr; 764 mhip = mtod(m, struct ip *); 765 *mhip = *ip; 766 if (hlen > sizeof (struct ip)) { 767 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 768 mhip->ip_v = IPVERSION; 769 mhip->ip_hl = mhlen >> 2; 770 } 771 m->m_len = mhlen; 772 /* XXX do we need to add ip->ip_off below ? */ 773 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 774 if (off + len >= ip->ip_len) { /* last fragment */ 775 len = ip->ip_len - off; 776 m->m_flags |= M_LASTFRAG; 777 } else 778 mhip->ip_off |= IP_MF; 779 mhip->ip_len = htons((u_short)(len + mhlen)); 780 m->m_next = m_copy(m0, off, len); 781 if (m->m_next == NULL) { /* copy failed */ 782 m_free(m); 783 error = ENOBUFS; /* ??? */ 784 V_ipstat.ips_odropped++; 785 goto done; 786 } 787 m->m_pkthdr.len = mhlen + len; 788 m->m_pkthdr.rcvif = NULL; 789#ifdef MAC 790 mac_netinet_fragment(m0, m); 791#endif 792 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 793 mhip->ip_off = htons(mhip->ip_off); 794 mhip->ip_sum = 0; 795 if (sw_csum & CSUM_DELAY_IP) 796 mhip->ip_sum = in_cksum(m, mhlen); 797 *mnext = m; 798 mnext = &m->m_nextpkt; 799 } 800 V_ipstat.ips_ofragments += nfrags; 801 802 /* set first marker for fragment chain */ 803 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 804 m0->m_pkthdr.csum_data = nfrags; 805 806 /* 807 * Update first fragment by trimming what's been copied out 808 * and updating header. 809 */ 810 m_adj(m0, hlen + firstlen - ip->ip_len); 811 m0->m_pkthdr.len = hlen + firstlen; 812 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 813 ip->ip_off |= IP_MF; 814 ip->ip_off = htons(ip->ip_off); 815 ip->ip_sum = 0; 816 if (sw_csum & CSUM_DELAY_IP) 817 ip->ip_sum = in_cksum(m0, hlen); 818 819done: 820 *m_frag = m0; 821 return error; 822} 823 824void 825in_delayed_cksum(struct mbuf *m) 826{ 827 struct ip *ip; 828 u_short csum, offset; 829 830 ip = mtod(m, struct ip *); 831 offset = ip->ip_hl << 2 ; 832 csum = in_cksum_skip(m, ip->ip_len, offset); 833 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 834 csum = 0xffff; 835 offset += m->m_pkthdr.csum_data; /* checksum offset */ 836 837 if (offset + sizeof(u_short) > m->m_len) { 838 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 839 m->m_len, offset, ip->ip_p); 840 /* 841 * XXX 842 * this shouldn't happen, but if it does, the 843 * correct behavior may be to insert the checksum 844 * in the appropriate next mbuf in the chain. 845 */ 846 return; 847 } 848 *(u_short *)(m->m_data + offset) = csum; 849} 850 851/* 852 * IP socket option processing. 853 */ 854int 855ip_ctloutput(struct socket *so, struct sockopt *sopt) 856{ 857 struct inpcb *inp = sotoinpcb(so); 858 int error, optval; 859 860 error = optval = 0; 861 if (sopt->sopt_level != IPPROTO_IP) { 862 if ((sopt->sopt_level == SOL_SOCKET) && 863 (sopt->sopt_name == SO_SETFIB)) { 864 inp->inp_inc.inc_fibnum = so->so_fibnum; 865 return (0); 866 } 867 return (EINVAL); 868 } 869 870 switch (sopt->sopt_dir) { 871 case SOPT_SET: 872 switch (sopt->sopt_name) { 873 case IP_OPTIONS: 874#ifdef notyet 875 case IP_RETOPTS: 876#endif 877 { 878 struct mbuf *m; 879 if (sopt->sopt_valsize > MLEN) { 880 error = EMSGSIZE; 881 break; 882 } 883 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 884 if (m == NULL) { 885 error = ENOBUFS; 886 break; 887 } 888 m->m_len = sopt->sopt_valsize; 889 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 890 m->m_len); 891 if (error) { 892 m_free(m); 893 break; 894 } 895 INP_WLOCK(inp); 896 error = ip_pcbopts(inp, sopt->sopt_name, m); 897 INP_WUNLOCK(inp); 898 return (error); 899 } 900 901#if defined(IP_NONLOCALBIND) 902 case IP_NONLOCALOK: 903 if (! ip_nonlocalok) { 904 error = ENOPROTOOPT; 905 break; 906 } 907 /* FALLTHROUGH */ 908#endif 909 case IP_TOS: 910 case IP_TTL: 911 case IP_MINTTL: 912 case IP_RECVOPTS: 913 case IP_RECVRETOPTS: 914 case IP_RECVDSTADDR: 915 case IP_RECVTTL: 916 case IP_RECVIF: 917 case IP_FAITH: 918 case IP_ONESBCAST: 919 case IP_DONTFRAG: 920 error = sooptcopyin(sopt, &optval, sizeof optval, 921 sizeof optval); 922 if (error) 923 break; 924 925 switch (sopt->sopt_name) { 926 case IP_TOS: 927 inp->inp_ip_tos = optval; 928 break; 929 930 case IP_TTL: 931 inp->inp_ip_ttl = optval; 932 break; 933 934 case IP_MINTTL: 935 if (optval >= 0 && optval <= MAXTTL) 936 inp->inp_ip_minttl = optval; 937 else 938 error = EINVAL; 939 break; 940 941#define OPTSET(bit) do { \ 942 INP_WLOCK(inp); \ 943 if (optval) \ 944 inp->inp_flags |= bit; \ 945 else \ 946 inp->inp_flags &= ~bit; \ 947 INP_WUNLOCK(inp); \ 948} while (0) 949 950 case IP_RECVOPTS: 951 OPTSET(INP_RECVOPTS); 952 break; 953 954 case IP_RECVRETOPTS: 955 OPTSET(INP_RECVRETOPTS); 956 break; 957 958 case IP_RECVDSTADDR: 959 OPTSET(INP_RECVDSTADDR); 960 break; 961 962 case IP_RECVTTL: 963 OPTSET(INP_RECVTTL); 964 break; 965 966 case IP_RECVIF: 967 OPTSET(INP_RECVIF); 968 break; 969 970 case IP_FAITH: 971 OPTSET(INP_FAITH); 972 break; 973 974 case IP_ONESBCAST: 975 OPTSET(INP_ONESBCAST); 976 break; 977 case IP_DONTFRAG: 978 OPTSET(INP_DONTFRAG); 979 break; 980#if defined(IP_NONLOCALBIND) 981 case IP_NONLOCALOK: 982 OPTSET(INP_NONLOCALOK); 983 break; 984#endif 985 } 986 break; 987#undef OPTSET 988 989 /* 990 * Multicast socket options are processed by the in_mcast 991 * module. 992 */ 993 case IP_MULTICAST_IF: 994 case IP_MULTICAST_VIF: 995 case IP_MULTICAST_TTL: 996 case IP_MULTICAST_LOOP: 997 case IP_ADD_MEMBERSHIP: 998 case IP_DROP_MEMBERSHIP: 999 case IP_ADD_SOURCE_MEMBERSHIP: 1000 case IP_DROP_SOURCE_MEMBERSHIP: 1001 case IP_BLOCK_SOURCE: 1002 case IP_UNBLOCK_SOURCE: 1003 case IP_MSFILTER: 1004 case MCAST_JOIN_GROUP: 1005 case MCAST_LEAVE_GROUP: 1006 case MCAST_JOIN_SOURCE_GROUP: 1007 case MCAST_LEAVE_SOURCE_GROUP: 1008 case MCAST_BLOCK_SOURCE: 1009 case MCAST_UNBLOCK_SOURCE: 1010 error = inp_setmoptions(inp, sopt); 1011 break; 1012 1013 case IP_PORTRANGE: 1014 error = sooptcopyin(sopt, &optval, sizeof optval, 1015 sizeof optval); 1016 if (error) 1017 break; 1018 1019 INP_WLOCK(inp); 1020 switch (optval) { 1021 case IP_PORTRANGE_DEFAULT: 1022 inp->inp_flags &= ~(INP_LOWPORT); 1023 inp->inp_flags &= ~(INP_HIGHPORT); 1024 break; 1025 1026 case IP_PORTRANGE_HIGH: 1027 inp->inp_flags &= ~(INP_LOWPORT); 1028 inp->inp_flags |= INP_HIGHPORT; 1029 break; 1030 1031 case IP_PORTRANGE_LOW: 1032 inp->inp_flags &= ~(INP_HIGHPORT); 1033 inp->inp_flags |= INP_LOWPORT; 1034 break; 1035 1036 default: 1037 error = EINVAL; 1038 break; 1039 } 1040 INP_WUNLOCK(inp); 1041 break; 1042 1043#ifdef IPSEC 1044 case IP_IPSEC_POLICY: 1045 { 1046 caddr_t req; 1047 struct mbuf *m; 1048 1049 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1050 break; 1051 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1052 break; 1053 req = mtod(m, caddr_t); 1054 error = ipsec_set_policy(inp, sopt->sopt_name, req, 1055 m->m_len, (sopt->sopt_td != NULL) ? 1056 sopt->sopt_td->td_ucred : NULL); 1057 m_freem(m); 1058 break; 1059 } 1060#endif /* IPSEC */ 1061 1062 default: 1063 error = ENOPROTOOPT; 1064 break; 1065 } 1066 break; 1067 1068 case SOPT_GET: 1069 switch (sopt->sopt_name) { 1070 case IP_OPTIONS: 1071 case IP_RETOPTS: 1072 if (inp->inp_options) 1073 error = sooptcopyout(sopt, 1074 mtod(inp->inp_options, 1075 char *), 1076 inp->inp_options->m_len); 1077 else 1078 sopt->sopt_valsize = 0; 1079 break; 1080 1081 case IP_TOS: 1082 case IP_TTL: 1083 case IP_MINTTL: 1084 case IP_RECVOPTS: 1085 case IP_RECVRETOPTS: 1086 case IP_RECVDSTADDR: 1087 case IP_RECVTTL: 1088 case IP_RECVIF: 1089 case IP_PORTRANGE: 1090 case IP_FAITH: 1091 case IP_ONESBCAST: 1092 case IP_DONTFRAG: 1093 switch (sopt->sopt_name) { 1094 1095 case IP_TOS: 1096 optval = inp->inp_ip_tos; 1097 break; 1098 1099 case IP_TTL: 1100 optval = inp->inp_ip_ttl; 1101 break; 1102 1103 case IP_MINTTL: 1104 optval = inp->inp_ip_minttl; 1105 break; 1106 1107#define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1108 1109 case IP_RECVOPTS: 1110 optval = OPTBIT(INP_RECVOPTS); 1111 break; 1112 1113 case IP_RECVRETOPTS: 1114 optval = OPTBIT(INP_RECVRETOPTS); 1115 break; 1116 1117 case IP_RECVDSTADDR: 1118 optval = OPTBIT(INP_RECVDSTADDR); 1119 break; 1120 1121 case IP_RECVTTL: 1122 optval = OPTBIT(INP_RECVTTL); 1123 break; 1124 1125 case IP_RECVIF: 1126 optval = OPTBIT(INP_RECVIF); 1127 break; 1128 1129 case IP_PORTRANGE: 1130 if (inp->inp_flags & INP_HIGHPORT) 1131 optval = IP_PORTRANGE_HIGH; 1132 else if (inp->inp_flags & INP_LOWPORT) 1133 optval = IP_PORTRANGE_LOW; 1134 else 1135 optval = 0; 1136 break; 1137 1138 case IP_FAITH: 1139 optval = OPTBIT(INP_FAITH); 1140 break; 1141 1142 case IP_ONESBCAST: 1143 optval = OPTBIT(INP_ONESBCAST); 1144 break; 1145 case IP_DONTFRAG: 1146 optval = OPTBIT(INP_DONTFRAG); 1147 break; 1148 } 1149 error = sooptcopyout(sopt, &optval, sizeof optval); 1150 break; 1151 1152 /* 1153 * Multicast socket options are processed by the in_mcast 1154 * module. 1155 */ 1156 case IP_MULTICAST_IF: 1157 case IP_MULTICAST_VIF: 1158 case IP_MULTICAST_TTL: 1159 case IP_MULTICAST_LOOP: 1160 case IP_MSFILTER: 1161 error = inp_getmoptions(inp, sopt); 1162 break; 1163 1164#ifdef IPSEC 1165 case IP_IPSEC_POLICY: 1166 { 1167 struct mbuf *m = NULL; 1168 caddr_t req = NULL; 1169 size_t len = 0; 1170 1171 if (m != 0) { 1172 req = mtod(m, caddr_t); 1173 len = m->m_len; 1174 } 1175 error = ipsec_get_policy(sotoinpcb(so), req, len, &m); 1176 if (error == 0) 1177 error = soopt_mcopyout(sopt, m); /* XXX */ 1178 if (error == 0) 1179 m_freem(m); 1180 break; 1181 } 1182#endif /* IPSEC */ 1183 1184 default: 1185 error = ENOPROTOOPT; 1186 break; 1187 } 1188 break; 1189 } 1190 return (error); 1191} 1192 1193/* 1194 * Routine called from ip_output() to loop back a copy of an IP multicast 1195 * packet to the input queue of a specified interface. Note that this 1196 * calls the output routine of the loopback "driver", but with an interface 1197 * pointer that might NOT be a loopback interface -- evil, but easier than 1198 * replicating that code here. 1199 */ 1200static void 1201ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1202 int hlen) 1203{ 1204 register struct ip *ip; 1205 struct mbuf *copym; 1206 1207 /* 1208 * Make a deep copy of the packet because we're going to 1209 * modify the pack in order to generate checksums. 1210 */ 1211 copym = m_dup(m, M_DONTWAIT); 1212 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1213 copym = m_pullup(copym, hlen); 1214 if (copym != NULL) { 1215 /* If needed, compute the checksum and mark it as valid. */ 1216 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1217 in_delayed_cksum(copym); 1218 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1219 copym->m_pkthdr.csum_flags |= 1220 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1221 copym->m_pkthdr.csum_data = 0xffff; 1222 } 1223 /* 1224 * We don't bother to fragment if the IP length is greater 1225 * than the interface's MTU. Can this possibly matter? 1226 */ 1227 ip = mtod(copym, struct ip *); 1228 ip->ip_len = htons(ip->ip_len); 1229 ip->ip_off = htons(ip->ip_off); 1230 ip->ip_sum = 0; 1231 ip->ip_sum = in_cksum(copym, hlen); 1232#if 1 /* XXX */ 1233 if (dst->sin_family != AF_INET) { 1234 printf("ip_mloopback: bad address family %d\n", 1235 dst->sin_family); 1236 dst->sin_family = AF_INET; 1237 } 1238#endif 1239 if_simloop(ifp, copym, dst->sin_family, 0); 1240 } 1241} 1242