ip_output.c revision 238092
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 238092 2012-07-04 07:37:53Z glebius $"); 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 55#include <net/if.h> 56#include <net/if_llatbl.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#ifdef SCTP 74#include <netinet/sctp.h> 75#include <netinet/sctp_crc32.h> 76#endif 77 78#ifdef IPSEC 79#include <netinet/ip_ipsec.h> 80#include <netipsec/ipsec.h> 81#endif /* IPSEC*/ 82 83#include <machine/in_cksum.h> 84 85#include <security/mac/mac_framework.h> 86 87VNET_DEFINE(u_short, ip_id); 88 89#ifdef MBUF_STRESS_TEST 90static int mbuf_frag_size = 0; 91SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 92 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 93#endif 94 95static void ip_mloopback 96 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 97 98 99extern int in_mcast_loop; 100extern struct protosw inetsw[]; 101 102/* 103 * IP output. The packet in mbuf chain m contains a skeletal IP 104 * header (with len, off, ttl, proto, tos, src, dst). 105 * ip_len and ip_off are in host format. 106 * The mbuf chain containing the packet will be freed. 107 * The mbuf opt, if present, will not be freed. 108 * If route ro is present and has ro_rt initialized, route lookup would be 109 * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL, 110 * then result of route lookup is stored in ro->ro_rt. 111 * 112 * In the IP forwarding case, the packet will arrive with options already 113 * inserted, so must have a NULL opt pointer. 114 */ 115int 116ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, 117 struct ip_moptions *imo, struct inpcb *inp) 118{ 119 struct ip *ip; 120 struct ifnet *ifp = NULL; /* keep compiler happy */ 121 struct mbuf *m0; 122 int hlen = sizeof (struct ip); 123 int mtu; 124 int n; /* scratchpad */ 125 int error = 0; 126 struct sockaddr_in *dst; 127 struct in_ifaddr *ia = NULL; 128 int isbroadcast, sw_csum; 129 struct route iproute; 130 struct rtentry *rte; /* cache for ro->ro_rt */ 131 struct in_addr odst; 132#ifdef IPFIREWALL_FORWARD 133 struct m_tag *fwd_tag = NULL; 134#endif 135#ifdef IPSEC 136 int no_route_but_check_spd = 0; 137#endif 138 M_ASSERTPKTHDR(m); 139 140 if (inp != NULL) { 141 INP_LOCK_ASSERT(inp); 142 M_SETFIB(m, inp->inp_inc.inc_fibnum); 143 if (inp->inp_flags & (INP_HW_FLOWID|INP_SW_FLOWID)) { 144 m->m_pkthdr.flowid = inp->inp_flowid; 145 m->m_flags |= M_FLOWID; 146 } 147 } 148 149 if (ro == NULL) { 150 ro = &iproute; 151 bzero(ro, sizeof (*ro)); 152 } 153 154#ifdef FLOWTABLE 155 if (ro->ro_rt == NULL) { 156 struct flentry *fle; 157 158 /* 159 * The flow table returns route entries valid for up to 30 160 * seconds; we rely on the remainder of ip_output() taking no 161 * longer than that long for the stability of ro_rt. The 162 * flow ID assignment must have happened before this point. 163 */ 164 fle = flowtable_lookup_mbuf(V_ip_ft, m, AF_INET); 165 if (fle != NULL) 166 flow_to_route(fle, ro); 167 } 168#endif 169 170 if (opt) { 171 int len = 0; 172 m = ip_insertoptions(m, opt, &len); 173 if (len != 0) 174 hlen = len; /* ip->ip_hl is updated above */ 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 /* Header already set, fetch hlen from there */ 196 hlen = ip->ip_hl << 2; 197 } 198 199 dst = (struct sockaddr_in *)&ro->ro_dst; 200again: 201 /* 202 * If there is a cached route, 203 * check that it is to the same destination 204 * and is still up. If not, free it and try again. 205 * The address family should also be checked in case of sharing the 206 * cache with IPv6. 207 */ 208 rte = ro->ro_rt; 209 if (rte && ((rte->rt_flags & RTF_UP) == 0 || 210 rte->rt_ifp == NULL || 211 !RT_LINK_IS_UP(rte->rt_ifp) || 212 dst->sin_family != AF_INET || 213 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 214 RO_RTFREE(ro); 215 ro->ro_lle = NULL; 216 rte = NULL; 217 } 218#ifdef IPFIREWALL_FORWARD 219 if (rte == NULL && fwd_tag == NULL) { 220#else 221 if (rte == NULL) { 222#endif 223 bzero(dst, sizeof(*dst)); 224 dst->sin_family = AF_INET; 225 dst->sin_len = sizeof(*dst); 226 dst->sin_addr = ip->ip_dst; 227 } 228 /* 229 * If routing to interface only, short circuit routing lookup. 230 * The use of an all-ones broadcast address implies this; an 231 * interface is specified by the broadcast address of an interface, 232 * or the destination address of a ptp interface. 233 */ 234 if (flags & IP_SENDONES) { 235 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 236 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 237 IPSTAT_INC(ips_noroute); 238 error = ENETUNREACH; 239 goto bad; 240 } 241 ip->ip_dst.s_addr = INADDR_BROADCAST; 242 dst->sin_addr = ip->ip_dst; 243 ifp = ia->ia_ifp; 244 ip->ip_ttl = 1; 245 isbroadcast = 1; 246 } else if (flags & IP_ROUTETOIF) { 247 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 248 (ia = ifatoia(ifa_ifwithnet(sintosa(dst), 0))) == NULL) { 249 IPSTAT_INC(ips_noroute); 250 error = ENETUNREACH; 251 goto bad; 252 } 253 ifp = ia->ia_ifp; 254 ip->ip_ttl = 1; 255 isbroadcast = in_broadcast(dst->sin_addr, ifp); 256 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 257 imo != NULL && imo->imo_multicast_ifp != NULL) { 258 /* 259 * Bypass the normal routing lookup for multicast 260 * packets if the interface is specified. 261 */ 262 ifp = imo->imo_multicast_ifp; 263 IFP_TO_IA(ifp, ia); 264 isbroadcast = 0; /* fool gcc */ 265 } else { 266 /* 267 * We want to do any cloning requested by the link layer, 268 * as this is probably required in all cases for correct 269 * operation (as it is for ARP). 270 */ 271 if (rte == NULL) { 272#ifdef RADIX_MPATH 273 rtalloc_mpath_fib(ro, 274 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 275 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 276#else 277 in_rtalloc_ign(ro, 0, 278 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 279#endif 280 rte = ro->ro_rt; 281 } 282 if (rte == NULL || 283 rte->rt_ifp == NULL || 284 !RT_LINK_IS_UP(rte->rt_ifp)) { 285#ifdef IPSEC 286 /* 287 * There is no route for this packet, but it is 288 * possible that a matching SPD entry exists. 289 */ 290 no_route_but_check_spd = 1; 291 mtu = 0; /* Silence GCC warning. */ 292 goto sendit; 293#endif 294 IPSTAT_INC(ips_noroute); 295 error = EHOSTUNREACH; 296 goto bad; 297 } 298 ia = ifatoia(rte->rt_ifa); 299 ifa_ref(&ia->ia_ifa); 300 ifp = rte->rt_ifp; 301 rte->rt_rmx.rmx_pksent++; 302 if (rte->rt_flags & RTF_GATEWAY) 303 dst = (struct sockaddr_in *)rte->rt_gateway; 304 if (rte->rt_flags & RTF_HOST) 305 isbroadcast = (rte->rt_flags & RTF_BROADCAST); 306 else 307 isbroadcast = in_broadcast(dst->sin_addr, ifp); 308 } 309 /* 310 * Calculate MTU. If we have a route that is up, use that, 311 * otherwise use the interface's MTU. 312 */ 313 if (rte != NULL && (rte->rt_flags & (RTF_UP|RTF_HOST))) { 314 /* 315 * This case can happen if the user changed the MTU 316 * of an interface after enabling IP on it. Because 317 * most netifs don't keep track of routes pointing to 318 * them, there is no way for one to update all its 319 * routes when the MTU is changed. 320 */ 321 if (rte->rt_rmx.rmx_mtu > ifp->if_mtu) 322 rte->rt_rmx.rmx_mtu = ifp->if_mtu; 323 mtu = rte->rt_rmx.rmx_mtu; 324 } else { 325 mtu = ifp->if_mtu; 326 } 327 /* Catch a possible divide by zero later. */ 328 KASSERT(mtu > 0, ("%s: mtu %d <= 0, rte=%p (rt_flags=0x%08x) ifp=%p", 329 __func__, mtu, rte, (rte != NULL) ? rte->rt_flags : 0, ifp)); 330 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 331 m->m_flags |= M_MCAST; 332 /* 333 * IP destination address is multicast. Make sure "dst" 334 * still points to the address in "ro". (It may have been 335 * changed to point to a gateway address, above.) 336 */ 337 dst = (struct sockaddr_in *)&ro->ro_dst; 338 /* 339 * See if the caller provided any multicast options 340 */ 341 if (imo != NULL) { 342 ip->ip_ttl = imo->imo_multicast_ttl; 343 if (imo->imo_multicast_vif != -1) 344 ip->ip_src.s_addr = 345 ip_mcast_src ? 346 ip_mcast_src(imo->imo_multicast_vif) : 347 INADDR_ANY; 348 } else 349 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 350 /* 351 * Confirm that the outgoing interface supports multicast. 352 */ 353 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 354 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 355 IPSTAT_INC(ips_noroute); 356 error = ENETUNREACH; 357 goto bad; 358 } 359 } 360 /* 361 * If source address not specified yet, use address 362 * of outgoing interface. 363 */ 364 if (ip->ip_src.s_addr == INADDR_ANY) { 365 /* Interface may have no addresses. */ 366 if (ia != NULL) 367 ip->ip_src = IA_SIN(ia)->sin_addr; 368 } 369 370 if ((imo == NULL && in_mcast_loop) || 371 (imo && imo->imo_multicast_loop)) { 372 /* 373 * Loop back multicast datagram if not expressly 374 * forbidden to do so, even if we are not a member 375 * of the group; ip_input() will filter it later, 376 * thus deferring a hash lookup and mutex acquisition 377 * at the expense of a cheap copy using m_copym(). 378 */ 379 ip_mloopback(ifp, m, dst, hlen); 380 } else { 381 /* 382 * If we are acting as a multicast router, perform 383 * multicast forwarding as if the packet had just 384 * arrived on the interface to which we are about 385 * to send. The multicast forwarding function 386 * recursively calls this function, using the 387 * IP_FORWARDING flag to prevent infinite recursion. 388 * 389 * Multicasts that are looped back by ip_mloopback(), 390 * above, will be forwarded by the ip_input() routine, 391 * if necessary. 392 */ 393 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 394 /* 395 * If rsvp daemon is not running, do not 396 * set ip_moptions. This ensures that the packet 397 * is multicast and not just sent down one link 398 * as prescribed by rsvpd. 399 */ 400 if (!V_rsvp_on) 401 imo = NULL; 402 if (ip_mforward && 403 ip_mforward(ip, ifp, m, imo) != 0) { 404 m_freem(m); 405 goto done; 406 } 407 } 408 } 409 410 /* 411 * Multicasts with a time-to-live of zero may be looped- 412 * back, above, but must not be transmitted on a network. 413 * Also, multicasts addressed to the loopback interface 414 * are not sent -- the above call to ip_mloopback() will 415 * loop back a copy. ip_input() will drop the copy if 416 * this host does not belong to the destination group on 417 * the loopback interface. 418 */ 419 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 420 m_freem(m); 421 goto done; 422 } 423 424 goto sendit; 425 } 426 427 /* 428 * If the source address is not specified yet, use the address 429 * of the outoing interface. 430 */ 431 if (ip->ip_src.s_addr == INADDR_ANY) { 432 /* Interface may have no addresses. */ 433 if (ia != NULL) { 434 ip->ip_src = IA_SIN(ia)->sin_addr; 435 } 436 } 437 438 /* 439 * Verify that we have any chance at all of being able to queue the 440 * packet or packet fragments, unless ALTQ is enabled on the given 441 * interface in which case packetdrop should be done by queueing. 442 */ 443 n = ip->ip_len / mtu + 1; /* how many fragments ? */ 444 if ( 445#ifdef ALTQ 446 (!ALTQ_IS_ENABLED(&ifp->if_snd)) && 447#endif /* ALTQ */ 448 (ifp->if_snd.ifq_len + n) >= ifp->if_snd.ifq_maxlen ) { 449 error = ENOBUFS; 450 IPSTAT_INC(ips_odropped); 451 ifp->if_snd.ifq_drops += n; 452 goto bad; 453 } 454 455 /* 456 * Look for broadcast address and 457 * verify user is allowed to send 458 * such a packet. 459 */ 460 if (isbroadcast) { 461 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 462 error = EADDRNOTAVAIL; 463 goto bad; 464 } 465 if ((flags & IP_ALLOWBROADCAST) == 0) { 466 error = EACCES; 467 goto bad; 468 } 469 /* don't allow broadcast messages to be fragmented */ 470 if (ip->ip_len > mtu) { 471 error = EMSGSIZE; 472 goto bad; 473 } 474 m->m_flags |= M_BCAST; 475 } else { 476 m->m_flags &= ~M_BCAST; 477 } 478 479sendit: 480#ifdef IPSEC 481 switch(ip_ipsec_output(&m, inp, &flags, &error)) { 482 case 1: 483 goto bad; 484 case -1: 485 goto done; 486 case 0: 487 default: 488 break; /* Continue with packet processing. */ 489 } 490 /* 491 * Check if there was a route for this packet; return error if not. 492 */ 493 if (no_route_but_check_spd) { 494 IPSTAT_INC(ips_noroute); 495 error = EHOSTUNREACH; 496 goto bad; 497 } 498 /* Update variables that are affected by ipsec4_output(). */ 499 ip = mtod(m, struct ip *); 500 hlen = ip->ip_hl << 2; 501#endif /* IPSEC */ 502 503 /* Jump over all PFIL processing if hooks are not active. */ 504 if (!PFIL_HOOKED(&V_inet_pfil_hook)) 505 goto passout; 506 507 /* Run through list of hooks for output packets. */ 508 odst.s_addr = ip->ip_dst.s_addr; 509 error = pfil_run_hooks(&V_inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 510 if (error != 0 || m == NULL) 511 goto done; 512 513 ip = mtod(m, struct ip *); 514 515 /* See if destination IP address was changed by packet filter. */ 516 if (odst.s_addr != ip->ip_dst.s_addr) { 517 m->m_flags |= M_SKIP_FIREWALL; 518 /* If destination is now ourself drop to ip_input(). */ 519 if (in_localip(ip->ip_dst)) { 520 m->m_flags |= M_FASTFWD_OURS; 521 if (m->m_pkthdr.rcvif == NULL) 522 m->m_pkthdr.rcvif = V_loif; 523 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 524 m->m_pkthdr.csum_flags |= 525 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 526 m->m_pkthdr.csum_data = 0xffff; 527 } 528 m->m_pkthdr.csum_flags |= 529 CSUM_IP_CHECKED | CSUM_IP_VALID; 530#ifdef SCTP 531 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 532 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 533#endif 534 error = netisr_queue(NETISR_IP, m); 535 goto done; 536 } else 537 goto again; /* Redo the routing table lookup. */ 538 } 539 540#ifdef IPFIREWALL_FORWARD 541 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 542 if (m->m_flags & M_FASTFWD_OURS) { 543 if (m->m_pkthdr.rcvif == NULL) 544 m->m_pkthdr.rcvif = V_loif; 545 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 546 m->m_pkthdr.csum_flags |= 547 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 548 m->m_pkthdr.csum_data = 0xffff; 549 } 550#ifdef SCTP 551 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 552 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 553#endif 554 m->m_pkthdr.csum_flags |= 555 CSUM_IP_CHECKED | CSUM_IP_VALID; 556 557 error = netisr_queue(NETISR_IP, m); 558 goto done; 559 } 560 /* Or forward to some other address? */ 561 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 562 if (fwd_tag) { 563 dst = (struct sockaddr_in *)&ro->ro_dst; 564 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 565 m->m_flags |= M_SKIP_FIREWALL; 566 m_tag_delete(m, fwd_tag); 567 goto again; 568 } 569#endif /* IPFIREWALL_FORWARD */ 570 571passout: 572 /* 127/8 must not appear on wire - RFC1122. */ 573 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 574 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 575 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 576 IPSTAT_INC(ips_badaddr); 577 error = EADDRNOTAVAIL; 578 goto bad; 579 } 580 } 581 582 m->m_pkthdr.csum_flags |= CSUM_IP; 583 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 584 if (sw_csum & CSUM_DELAY_DATA) { 585 in_delayed_cksum(m); 586 sw_csum &= ~CSUM_DELAY_DATA; 587 } 588#ifdef SCTP 589 if (sw_csum & CSUM_SCTP) { 590 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 591 sw_csum &= ~CSUM_SCTP; 592 } 593#endif 594 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 595 596 /* 597 * If small enough for interface, or the interface will take 598 * care of the fragmentation for us, we can just send directly. 599 */ 600 if (ip->ip_len <= mtu || 601 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 602 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 603 ip->ip_len = htons(ip->ip_len); 604 ip->ip_off = htons(ip->ip_off); 605 ip->ip_sum = 0; 606 if (sw_csum & CSUM_DELAY_IP) 607 ip->ip_sum = in_cksum(m, hlen); 608 609 /* 610 * Record statistics for this interface address. 611 * With CSUM_TSO the byte/packet count will be slightly 612 * incorrect because we count the IP+TCP headers only 613 * once instead of for every generated packet. 614 */ 615 if (!(flags & IP_FORWARDING) && ia) { 616 if (m->m_pkthdr.csum_flags & CSUM_TSO) 617 ia->ia_ifa.if_opackets += 618 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 619 else 620 ia->ia_ifa.if_opackets++; 621 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 622 } 623#ifdef MBUF_STRESS_TEST 624 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 625 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 626#endif 627 /* 628 * Reset layer specific mbuf flags 629 * to avoid confusing lower layers. 630 */ 631 m->m_flags &= ~(M_PROTOFLAGS); 632 error = (*ifp->if_output)(ifp, m, 633 (struct sockaddr *)dst, ro); 634 goto done; 635 } 636 637 /* Balk when DF bit is set or the interface didn't support TSO. */ 638 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 639 error = EMSGSIZE; 640 IPSTAT_INC(ips_cantfrag); 641 goto bad; 642 } 643 644 /* 645 * Too large for interface; fragment if possible. If successful, 646 * on return, m will point to a list of packets to be sent. 647 */ 648 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 649 if (error) 650 goto bad; 651 for (; m; m = m0) { 652 m0 = m->m_nextpkt; 653 m->m_nextpkt = 0; 654 if (error == 0) { 655 /* Record statistics for this interface address. */ 656 if (ia != NULL) { 657 ia->ia_ifa.if_opackets++; 658 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 659 } 660 /* 661 * Reset layer specific mbuf flags 662 * to avoid confusing upper layers. 663 */ 664 m->m_flags &= ~(M_PROTOFLAGS); 665 666 error = (*ifp->if_output)(ifp, m, 667 (struct sockaddr *)dst, ro); 668 } else 669 m_freem(m); 670 } 671 672 if (error == 0) 673 IPSTAT_INC(ips_fragmented); 674 675done: 676 if (ro == &iproute) 677 RO_RTFREE(ro); 678 if (ia != NULL) 679 ifa_free(&ia->ia_ifa); 680 return (error); 681bad: 682 m_freem(m); 683 goto done; 684} 685 686/* 687 * Create a chain of fragments which fit the given mtu. m_frag points to the 688 * mbuf to be fragmented; on return it points to the chain with the fragments. 689 * Return 0 if no error. If error, m_frag may contain a partially built 690 * chain of fragments that should be freed by the caller. 691 * 692 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 693 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 694 */ 695int 696ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 697 u_long if_hwassist_flags, int sw_csum) 698{ 699 int error = 0; 700 int hlen = ip->ip_hl << 2; 701 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 702 int off; 703 struct mbuf *m0 = *m_frag; /* the original packet */ 704 int firstlen; 705 struct mbuf **mnext; 706 int nfrags; 707 708 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 709 IPSTAT_INC(ips_cantfrag); 710 return EMSGSIZE; 711 } 712 713 /* 714 * Must be able to put at least 8 bytes per fragment. 715 */ 716 if (len < 8) 717 return EMSGSIZE; 718 719 /* 720 * If the interface will not calculate checksums on 721 * fragmented packets, then do it here. 722 */ 723 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 724 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 725 in_delayed_cksum(m0); 726 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 727 } 728#ifdef SCTP 729 if (m0->m_pkthdr.csum_flags & CSUM_SCTP && 730 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 731 sctp_delayed_cksum(m0, hlen); 732 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 733 } 734#endif 735 if (len > PAGE_SIZE) { 736 /* 737 * Fragment large datagrams such that each segment 738 * contains a multiple of PAGE_SIZE amount of data, 739 * plus headers. This enables a receiver to perform 740 * page-flipping zero-copy optimizations. 741 * 742 * XXX When does this help given that sender and receiver 743 * could have different page sizes, and also mtu could 744 * be less than the receiver's page size ? 745 */ 746 int newlen; 747 struct mbuf *m; 748 749 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 750 off += m->m_len; 751 752 /* 753 * firstlen (off - hlen) must be aligned on an 754 * 8-byte boundary 755 */ 756 if (off < hlen) 757 goto smart_frag_failure; 758 off = ((off - hlen) & ~7) + hlen; 759 newlen = (~PAGE_MASK) & mtu; 760 if ((newlen + sizeof (struct ip)) > mtu) { 761 /* we failed, go back the default */ 762smart_frag_failure: 763 newlen = len; 764 off = hlen + len; 765 } 766 len = newlen; 767 768 } else { 769 off = hlen + len; 770 } 771 772 firstlen = off - hlen; 773 mnext = &m0->m_nextpkt; /* pointer to next packet */ 774 775 /* 776 * Loop through length of segment after first fragment, 777 * make new header and copy data of each part and link onto chain. 778 * Here, m0 is the original packet, m is the fragment being created. 779 * The fragments are linked off the m_nextpkt of the original 780 * packet, which after processing serves as the first fragment. 781 */ 782 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 783 struct ip *mhip; /* ip header on the fragment */ 784 struct mbuf *m; 785 int mhlen = sizeof (struct ip); 786 787 MGETHDR(m, M_DONTWAIT, MT_DATA); 788 if (m == NULL) { 789 error = ENOBUFS; 790 IPSTAT_INC(ips_odropped); 791 goto done; 792 } 793 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 794 /* 795 * In the first mbuf, leave room for the link header, then 796 * copy the original IP header including options. The payload 797 * goes into an additional mbuf chain returned by m_copym(). 798 */ 799 m->m_data += max_linkhdr; 800 mhip = mtod(m, struct ip *); 801 *mhip = *ip; 802 if (hlen > sizeof (struct ip)) { 803 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 804 mhip->ip_v = IPVERSION; 805 mhip->ip_hl = mhlen >> 2; 806 } 807 m->m_len = mhlen; 808 /* XXX do we need to add ip->ip_off below ? */ 809 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 810 if (off + len >= ip->ip_len) { /* last fragment */ 811 len = ip->ip_len - off; 812 m->m_flags |= M_LASTFRAG; 813 } else 814 mhip->ip_off |= IP_MF; 815 mhip->ip_len = htons((u_short)(len + mhlen)); 816 m->m_next = m_copym(m0, off, len, M_DONTWAIT); 817 if (m->m_next == NULL) { /* copy failed */ 818 m_free(m); 819 error = ENOBUFS; /* ??? */ 820 IPSTAT_INC(ips_odropped); 821 goto done; 822 } 823 m->m_pkthdr.len = mhlen + len; 824 m->m_pkthdr.rcvif = NULL; 825#ifdef MAC 826 mac_netinet_fragment(m0, m); 827#endif 828 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 829 mhip->ip_off = htons(mhip->ip_off); 830 mhip->ip_sum = 0; 831 if (sw_csum & CSUM_DELAY_IP) 832 mhip->ip_sum = in_cksum(m, mhlen); 833 *mnext = m; 834 mnext = &m->m_nextpkt; 835 } 836 IPSTAT_ADD(ips_ofragments, nfrags); 837 838 /* set first marker for fragment chain */ 839 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 840 m0->m_pkthdr.csum_data = nfrags; 841 842 /* 843 * Update first fragment by trimming what's been copied out 844 * and updating header. 845 */ 846 m_adj(m0, hlen + firstlen - ip->ip_len); 847 m0->m_pkthdr.len = hlen + firstlen; 848 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 849 ip->ip_off |= IP_MF; 850 ip->ip_off = htons(ip->ip_off); 851 ip->ip_sum = 0; 852 if (sw_csum & CSUM_DELAY_IP) 853 ip->ip_sum = in_cksum(m0, hlen); 854 855done: 856 *m_frag = m0; 857 return error; 858} 859 860void 861in_delayed_cksum(struct mbuf *m) 862{ 863 struct ip *ip; 864 u_short csum, offset; 865 866 ip = mtod(m, struct ip *); 867 offset = ip->ip_hl << 2 ; 868 csum = in_cksum_skip(m, ip->ip_len, offset); 869 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 870 csum = 0xffff; 871 offset += m->m_pkthdr.csum_data; /* checksum offset */ 872 873 if (offset + sizeof(u_short) > m->m_len) { 874 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 875 m->m_len, offset, ip->ip_p); 876 /* 877 * XXX 878 * this shouldn't happen, but if it does, the 879 * correct behavior may be to insert the checksum 880 * in the appropriate next mbuf in the chain. 881 */ 882 return; 883 } 884 *(u_short *)(m->m_data + offset) = csum; 885} 886 887/* 888 * IP socket option processing. 889 */ 890int 891ip_ctloutput(struct socket *so, struct sockopt *sopt) 892{ 893 struct inpcb *inp = sotoinpcb(so); 894 int error, optval; 895 896 error = optval = 0; 897 if (sopt->sopt_level != IPPROTO_IP) { 898 error = EINVAL; 899 900 if (sopt->sopt_level == SOL_SOCKET && 901 sopt->sopt_dir == SOPT_SET) { 902 switch (sopt->sopt_name) { 903 case SO_REUSEADDR: 904 INP_WLOCK(inp); 905 if (IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) { 906 if ((so->so_options & 907 (SO_REUSEADDR | SO_REUSEPORT)) != 0) 908 inp->inp_flags2 |= INP_REUSEPORT; 909 else 910 inp->inp_flags2 &= ~INP_REUSEPORT; 911 } 912 INP_WUNLOCK(inp); 913 error = 0; 914 break; 915 case SO_REUSEPORT: 916 INP_WLOCK(inp); 917 if ((so->so_options & SO_REUSEPORT) != 0) 918 inp->inp_flags2 |= INP_REUSEPORT; 919 else 920 inp->inp_flags2 &= ~INP_REUSEPORT; 921 INP_WUNLOCK(inp); 922 error = 0; 923 break; 924 case SO_SETFIB: 925 INP_WLOCK(inp); 926 inp->inp_inc.inc_fibnum = so->so_fibnum; 927 INP_WUNLOCK(inp); 928 error = 0; 929 break; 930 default: 931 break; 932 } 933 } 934 return (error); 935 } 936 937 switch (sopt->sopt_dir) { 938 case SOPT_SET: 939 switch (sopt->sopt_name) { 940 case IP_OPTIONS: 941#ifdef notyet 942 case IP_RETOPTS: 943#endif 944 { 945 struct mbuf *m; 946 if (sopt->sopt_valsize > MLEN) { 947 error = EMSGSIZE; 948 break; 949 } 950 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 951 if (m == NULL) { 952 error = ENOBUFS; 953 break; 954 } 955 m->m_len = sopt->sopt_valsize; 956 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 957 m->m_len); 958 if (error) { 959 m_free(m); 960 break; 961 } 962 INP_WLOCK(inp); 963 error = ip_pcbopts(inp, sopt->sopt_name, m); 964 INP_WUNLOCK(inp); 965 return (error); 966 } 967 968 case IP_BINDANY: 969 if (sopt->sopt_td != NULL) { 970 error = priv_check(sopt->sopt_td, 971 PRIV_NETINET_BINDANY); 972 if (error) 973 break; 974 } 975 /* FALLTHROUGH */ 976 case IP_TOS: 977 case IP_TTL: 978 case IP_MINTTL: 979 case IP_RECVOPTS: 980 case IP_RECVRETOPTS: 981 case IP_RECVDSTADDR: 982 case IP_RECVTTL: 983 case IP_RECVIF: 984 case IP_FAITH: 985 case IP_ONESBCAST: 986 case IP_DONTFRAG: 987 case IP_RECVTOS: 988 error = sooptcopyin(sopt, &optval, sizeof optval, 989 sizeof optval); 990 if (error) 991 break; 992 993 switch (sopt->sopt_name) { 994 case IP_TOS: 995 inp->inp_ip_tos = optval; 996 break; 997 998 case IP_TTL: 999 inp->inp_ip_ttl = optval; 1000 break; 1001 1002 case IP_MINTTL: 1003 if (optval >= 0 && optval <= MAXTTL) 1004 inp->inp_ip_minttl = optval; 1005 else 1006 error = EINVAL; 1007 break; 1008 1009#define OPTSET(bit) do { \ 1010 INP_WLOCK(inp); \ 1011 if (optval) \ 1012 inp->inp_flags |= bit; \ 1013 else \ 1014 inp->inp_flags &= ~bit; \ 1015 INP_WUNLOCK(inp); \ 1016} while (0) 1017 1018 case IP_RECVOPTS: 1019 OPTSET(INP_RECVOPTS); 1020 break; 1021 1022 case IP_RECVRETOPTS: 1023 OPTSET(INP_RECVRETOPTS); 1024 break; 1025 1026 case IP_RECVDSTADDR: 1027 OPTSET(INP_RECVDSTADDR); 1028 break; 1029 1030 case IP_RECVTTL: 1031 OPTSET(INP_RECVTTL); 1032 break; 1033 1034 case IP_RECVIF: 1035 OPTSET(INP_RECVIF); 1036 break; 1037 1038 case IP_FAITH: 1039 OPTSET(INP_FAITH); 1040 break; 1041 1042 case IP_ONESBCAST: 1043 OPTSET(INP_ONESBCAST); 1044 break; 1045 case IP_DONTFRAG: 1046 OPTSET(INP_DONTFRAG); 1047 break; 1048 case IP_BINDANY: 1049 OPTSET(INP_BINDANY); 1050 break; 1051 case IP_RECVTOS: 1052 OPTSET(INP_RECVTOS); 1053 break; 1054 } 1055 break; 1056#undef OPTSET 1057 1058 /* 1059 * Multicast socket options are processed by the in_mcast 1060 * module. 1061 */ 1062 case IP_MULTICAST_IF: 1063 case IP_MULTICAST_VIF: 1064 case IP_MULTICAST_TTL: 1065 case IP_MULTICAST_LOOP: 1066 case IP_ADD_MEMBERSHIP: 1067 case IP_DROP_MEMBERSHIP: 1068 case IP_ADD_SOURCE_MEMBERSHIP: 1069 case IP_DROP_SOURCE_MEMBERSHIP: 1070 case IP_BLOCK_SOURCE: 1071 case IP_UNBLOCK_SOURCE: 1072 case IP_MSFILTER: 1073 case MCAST_JOIN_GROUP: 1074 case MCAST_LEAVE_GROUP: 1075 case MCAST_JOIN_SOURCE_GROUP: 1076 case MCAST_LEAVE_SOURCE_GROUP: 1077 case MCAST_BLOCK_SOURCE: 1078 case MCAST_UNBLOCK_SOURCE: 1079 error = inp_setmoptions(inp, sopt); 1080 break; 1081 1082 case IP_PORTRANGE: 1083 error = sooptcopyin(sopt, &optval, sizeof optval, 1084 sizeof optval); 1085 if (error) 1086 break; 1087 1088 INP_WLOCK(inp); 1089 switch (optval) { 1090 case IP_PORTRANGE_DEFAULT: 1091 inp->inp_flags &= ~(INP_LOWPORT); 1092 inp->inp_flags &= ~(INP_HIGHPORT); 1093 break; 1094 1095 case IP_PORTRANGE_HIGH: 1096 inp->inp_flags &= ~(INP_LOWPORT); 1097 inp->inp_flags |= INP_HIGHPORT; 1098 break; 1099 1100 case IP_PORTRANGE_LOW: 1101 inp->inp_flags &= ~(INP_HIGHPORT); 1102 inp->inp_flags |= INP_LOWPORT; 1103 break; 1104 1105 default: 1106 error = EINVAL; 1107 break; 1108 } 1109 INP_WUNLOCK(inp); 1110 break; 1111 1112#ifdef IPSEC 1113 case IP_IPSEC_POLICY: 1114 { 1115 caddr_t req; 1116 struct mbuf *m; 1117 1118 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1119 break; 1120 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1121 break; 1122 req = mtod(m, caddr_t); 1123 error = ipsec_set_policy(inp, sopt->sopt_name, req, 1124 m->m_len, (sopt->sopt_td != NULL) ? 1125 sopt->sopt_td->td_ucred : NULL); 1126 m_freem(m); 1127 break; 1128 } 1129#endif /* IPSEC */ 1130 1131 default: 1132 error = ENOPROTOOPT; 1133 break; 1134 } 1135 break; 1136 1137 case SOPT_GET: 1138 switch (sopt->sopt_name) { 1139 case IP_OPTIONS: 1140 case IP_RETOPTS: 1141 if (inp->inp_options) 1142 error = sooptcopyout(sopt, 1143 mtod(inp->inp_options, 1144 char *), 1145 inp->inp_options->m_len); 1146 else 1147 sopt->sopt_valsize = 0; 1148 break; 1149 1150 case IP_TOS: 1151 case IP_TTL: 1152 case IP_MINTTL: 1153 case IP_RECVOPTS: 1154 case IP_RECVRETOPTS: 1155 case IP_RECVDSTADDR: 1156 case IP_RECVTTL: 1157 case IP_RECVIF: 1158 case IP_PORTRANGE: 1159 case IP_FAITH: 1160 case IP_ONESBCAST: 1161 case IP_DONTFRAG: 1162 case IP_BINDANY: 1163 case IP_RECVTOS: 1164 switch (sopt->sopt_name) { 1165 1166 case IP_TOS: 1167 optval = inp->inp_ip_tos; 1168 break; 1169 1170 case IP_TTL: 1171 optval = inp->inp_ip_ttl; 1172 break; 1173 1174 case IP_MINTTL: 1175 optval = inp->inp_ip_minttl; 1176 break; 1177 1178#define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1179 1180 case IP_RECVOPTS: 1181 optval = OPTBIT(INP_RECVOPTS); 1182 break; 1183 1184 case IP_RECVRETOPTS: 1185 optval = OPTBIT(INP_RECVRETOPTS); 1186 break; 1187 1188 case IP_RECVDSTADDR: 1189 optval = OPTBIT(INP_RECVDSTADDR); 1190 break; 1191 1192 case IP_RECVTTL: 1193 optval = OPTBIT(INP_RECVTTL); 1194 break; 1195 1196 case IP_RECVIF: 1197 optval = OPTBIT(INP_RECVIF); 1198 break; 1199 1200 case IP_PORTRANGE: 1201 if (inp->inp_flags & INP_HIGHPORT) 1202 optval = IP_PORTRANGE_HIGH; 1203 else if (inp->inp_flags & INP_LOWPORT) 1204 optval = IP_PORTRANGE_LOW; 1205 else 1206 optval = 0; 1207 break; 1208 1209 case IP_FAITH: 1210 optval = OPTBIT(INP_FAITH); 1211 break; 1212 1213 case IP_ONESBCAST: 1214 optval = OPTBIT(INP_ONESBCAST); 1215 break; 1216 case IP_DONTFRAG: 1217 optval = OPTBIT(INP_DONTFRAG); 1218 break; 1219 case IP_BINDANY: 1220 optval = OPTBIT(INP_BINDANY); 1221 break; 1222 case IP_RECVTOS: 1223 optval = OPTBIT(INP_RECVTOS); 1224 break; 1225 } 1226 error = sooptcopyout(sopt, &optval, sizeof optval); 1227 break; 1228 1229 /* 1230 * Multicast socket options are processed by the in_mcast 1231 * module. 1232 */ 1233 case IP_MULTICAST_IF: 1234 case IP_MULTICAST_VIF: 1235 case IP_MULTICAST_TTL: 1236 case IP_MULTICAST_LOOP: 1237 case IP_MSFILTER: 1238 error = inp_getmoptions(inp, sopt); 1239 break; 1240 1241#ifdef IPSEC 1242 case IP_IPSEC_POLICY: 1243 { 1244 struct mbuf *m = NULL; 1245 caddr_t req = NULL; 1246 size_t len = 0; 1247 1248 if (m != 0) { 1249 req = mtod(m, caddr_t); 1250 len = m->m_len; 1251 } 1252 error = ipsec_get_policy(sotoinpcb(so), req, len, &m); 1253 if (error == 0) 1254 error = soopt_mcopyout(sopt, m); /* XXX */ 1255 if (error == 0) 1256 m_freem(m); 1257 break; 1258 } 1259#endif /* IPSEC */ 1260 1261 default: 1262 error = ENOPROTOOPT; 1263 break; 1264 } 1265 break; 1266 } 1267 return (error); 1268} 1269 1270/* 1271 * Routine called from ip_output() to loop back a copy of an IP multicast 1272 * packet to the input queue of a specified interface. Note that this 1273 * calls the output routine of the loopback "driver", but with an interface 1274 * pointer that might NOT be a loopback interface -- evil, but easier than 1275 * replicating that code here. 1276 */ 1277static void 1278ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1279 int hlen) 1280{ 1281 register struct ip *ip; 1282 struct mbuf *copym; 1283 1284 /* 1285 * Make a deep copy of the packet because we're going to 1286 * modify the pack in order to generate checksums. 1287 */ 1288 copym = m_dup(m, M_DONTWAIT); 1289 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1290 copym = m_pullup(copym, hlen); 1291 if (copym != NULL) { 1292 /* If needed, compute the checksum and mark it as valid. */ 1293 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1294 in_delayed_cksum(copym); 1295 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1296 copym->m_pkthdr.csum_flags |= 1297 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1298 copym->m_pkthdr.csum_data = 0xffff; 1299 } 1300 /* 1301 * We don't bother to fragment if the IP length is greater 1302 * than the interface's MTU. Can this possibly matter? 1303 */ 1304 ip = mtod(copym, struct ip *); 1305 ip->ip_len = htons(ip->ip_len); 1306 ip->ip_off = htons(ip->ip_off); 1307 ip->ip_sum = 0; 1308 ip->ip_sum = in_cksum(copym, hlen); 1309#if 1 /* XXX */ 1310 if (dst->sin_family != AF_INET) { 1311 printf("ip_mloopback: bad address family %d\n", 1312 dst->sin_family); 1313 dst->sin_family = AF_INET; 1314 } 1315#endif 1316 if_simloop(ifp, copym, dst->sin_family, 0); 1317 } 1318} 1319