ip_output.c revision 195699
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 195699 2009-07-14 22:48:30Z rwatson $"); 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#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 87#define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 88 x, (ntohl(a.s_addr)>>24)&0xFF,\ 89 (ntohl(a.s_addr)>>16)&0xFF,\ 90 (ntohl(a.s_addr)>>8)&0xFF,\ 91 (ntohl(a.s_addr))&0xFF, y); 92 93VNET_DEFINE(u_short, ip_id); 94 95#ifdef MBUF_STRESS_TEST 96int mbuf_frag_size = 0; 97SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 98 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 99#endif 100 101static void ip_mloopback 102 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 103 104 105extern int in_mcast_loop; 106extern struct protosw inetsw[]; 107 108/* 109 * IP output. The packet in mbuf chain m contains a skeletal IP 110 * header (with len, off, ttl, proto, tos, src, dst). 111 * The mbuf chain containing the packet will be freed. 112 * The mbuf opt, if present, will not be freed. 113 * In the IP forwarding case, the packet will arrive with options already 114 * inserted, so must have a NULL opt pointer. 115 */ 116int 117ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, 118 struct ip_moptions *imo, struct inpcb *inp) 119{ 120 struct ip *ip; 121 struct ifnet *ifp = NULL; /* keep compiler happy */ 122 struct mbuf *m0; 123 int hlen = sizeof (struct ip); 124 int mtu; 125 int len, error = 0; 126 int nortfree = 0; 127 struct sockaddr_in *dst = NULL; /* keep compiler happy */ 128 struct in_ifaddr *ia = NULL; 129 int isbroadcast, sw_csum; 130 struct route iproute; 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#ifdef FLOWTABLE 154 /* 155 * The flow table returns route entries valid for up to 30 156 * seconds; we rely on the remainder of ip_output() taking no 157 * longer than that long for the stability of ro_rt. The 158 * flow ID assignment must have happened before this point. 159 */ 160 if (flowtable_lookup(V_ip_ft, m, ro) == 0) 161 nortfree = 1; 162#endif 163 } 164 165 if (opt) { 166 len = 0; 167 m = ip_insertoptions(m, opt, &len); 168 if (len != 0) 169 hlen = len; 170 } 171 ip = mtod(m, struct ip *); 172 173 /* 174 * Fill in IP header. If we are not allowing fragmentation, 175 * then the ip_id field is meaningless, but we don't set it 176 * to zero. Doing so causes various problems when devices along 177 * the path (routers, load balancers, firewalls, etc.) illegally 178 * disable DF on our packet. Note that a 16-bit counter 179 * will wrap around in less than 10 seconds at 100 Mbit/s on a 180 * medium with MTU 1500. See Steven M. Bellovin, "A Technique 181 * for Counting NATted Hosts", Proc. IMW'02, available at 182 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>. 183 */ 184 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 185 ip->ip_v = IPVERSION; 186 ip->ip_hl = hlen >> 2; 187 ip->ip_id = ip_newid(); 188 IPSTAT_INC(ips_localout); 189 } else { 190 hlen = ip->ip_hl << 2; 191 } 192 193 dst = (struct sockaddr_in *)&ro->ro_dst; 194again: 195 /* 196 * If there is a cached route, 197 * check that it is to the same destination 198 * and is still up. If not, free it and try again. 199 * The address family should also be checked in case of sharing the 200 * cache with IPv6. 201 */ 202 if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 203 dst->sin_family != AF_INET || 204 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 205 if (!nortfree) 206 RTFREE(ro->ro_rt); 207 ro->ro_rt = (struct rtentry *)NULL; 208 } 209#ifdef IPFIREWALL_FORWARD 210 if (ro->ro_rt == NULL && fwd_tag == NULL) { 211#else 212 if (ro->ro_rt == NULL) { 213#endif 214 bzero(dst, sizeof(*dst)); 215 dst->sin_family = AF_INET; 216 dst->sin_len = sizeof(*dst); 217 dst->sin_addr = ip->ip_dst; 218 } 219 /* 220 * If routing to interface only, short circuit routing lookup. 221 * The use of an all-ones broadcast address implies this; an 222 * interface is specified by the broadcast address of an interface, 223 * or the destination address of a ptp interface. 224 */ 225 if (flags & IP_SENDONES) { 226 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 227 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 228 IPSTAT_INC(ips_noroute); 229 error = ENETUNREACH; 230 goto bad; 231 } 232 ip->ip_dst.s_addr = INADDR_BROADCAST; 233 dst->sin_addr = ip->ip_dst; 234 ifp = ia->ia_ifp; 235 ip->ip_ttl = 1; 236 isbroadcast = 1; 237 } else if (flags & IP_ROUTETOIF) { 238 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 239 (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) { 240 IPSTAT_INC(ips_noroute); 241 error = ENETUNREACH; 242 goto bad; 243 } 244 ifp = ia->ia_ifp; 245 ip->ip_ttl = 1; 246 isbroadcast = in_broadcast(dst->sin_addr, ifp); 247 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 248 imo != NULL && imo->imo_multicast_ifp != NULL) { 249 /* 250 * Bypass the normal routing lookup for multicast 251 * packets if the interface is specified. 252 */ 253 ifp = imo->imo_multicast_ifp; 254 IFP_TO_IA(ifp, ia); 255 isbroadcast = 0; /* fool gcc */ 256 } else { 257 /* 258 * We want to do any cloning requested by the link layer, 259 * as this is probably required in all cases for correct 260 * operation (as it is for ARP). 261 */ 262 if (ro->ro_rt == NULL) 263#ifdef RADIX_MPATH 264 rtalloc_mpath_fib(ro, 265 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 266 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 267#else 268 in_rtalloc_ign(ro, 0, 269 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 270#endif 271 if (ro->ro_rt == NULL) { 272#ifdef IPSEC 273 /* 274 * There is no route for this packet, but it is 275 * possible that a matching SPD entry exists. 276 */ 277 no_route_but_check_spd = 1; 278 mtu = 0; /* Silence GCC warning. */ 279 goto sendit; 280#endif 281 IPSTAT_INC(ips_noroute); 282 error = EHOSTUNREACH; 283 goto bad; 284 } 285 ia = ifatoia(ro->ro_rt->rt_ifa); 286 ifa_ref(&ia->ia_ifa); 287 ifp = ro->ro_rt->rt_ifp; 288 ro->ro_rt->rt_rmx.rmx_pksent++; 289 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 290 dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 291 if (ro->ro_rt->rt_flags & RTF_HOST) 292 isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 293 else 294 isbroadcast = in_broadcast(dst->sin_addr, ifp); 295 } 296 /* 297 * Calculate MTU. If we have a route that is up, use that, 298 * otherwise use the interface's MTU. 299 */ 300 if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) { 301 /* 302 * This case can happen if the user changed the MTU 303 * of an interface after enabling IP on it. Because 304 * most netifs don't keep track of routes pointing to 305 * them, there is no way for one to update all its 306 * routes when the MTU is changed. 307 */ 308 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu) 309 ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 310 mtu = ro->ro_rt->rt_rmx.rmx_mtu; 311 } else { 312 mtu = ifp->if_mtu; 313 } 314 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 315 m->m_flags |= M_MCAST; 316 /* 317 * IP destination address is multicast. Make sure "dst" 318 * still points to the address in "ro". (It may have been 319 * changed to point to a gateway address, above.) 320 */ 321 dst = (struct sockaddr_in *)&ro->ro_dst; 322 /* 323 * See if the caller provided any multicast options 324 */ 325 if (imo != NULL) { 326 ip->ip_ttl = imo->imo_multicast_ttl; 327 if (imo->imo_multicast_vif != -1) 328 ip->ip_src.s_addr = 329 ip_mcast_src ? 330 ip_mcast_src(imo->imo_multicast_vif) : 331 INADDR_ANY; 332 } else 333 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 334 /* 335 * Confirm that the outgoing interface supports multicast. 336 */ 337 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 338 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 339 IPSTAT_INC(ips_noroute); 340 error = ENETUNREACH; 341 goto bad; 342 } 343 } 344 /* 345 * If source address not specified yet, use address 346 * of outgoing interface. 347 */ 348 if (ip->ip_src.s_addr == INADDR_ANY) { 349 /* Interface may have no addresses. */ 350 if (ia != NULL) 351 ip->ip_src = IA_SIN(ia)->sin_addr; 352 } 353 354 if ((imo == NULL && in_mcast_loop) || 355 (imo && imo->imo_multicast_loop)) { 356 /* 357 * Loop back multicast datagram if not expressly 358 * forbidden to do so, even if we are not a member 359 * of the group; ip_input() will filter it later, 360 * thus deferring a hash lookup and mutex acquisition 361 * at the expense of a cheap copy using m_copym(). 362 */ 363 ip_mloopback(ifp, m, dst, hlen); 364 } else { 365 /* 366 * If we are acting as a multicast router, perform 367 * multicast forwarding as if the packet had just 368 * arrived on the interface to which we are about 369 * to send. The multicast forwarding function 370 * recursively calls this function, using the 371 * IP_FORWARDING flag to prevent infinite recursion. 372 * 373 * Multicasts that are looped back by ip_mloopback(), 374 * above, will be forwarded by the ip_input() routine, 375 * if necessary. 376 */ 377 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 378 /* 379 * If rsvp daemon is not running, do not 380 * set ip_moptions. This ensures that the packet 381 * is multicast and not just sent down one link 382 * as prescribed by rsvpd. 383 */ 384 if (!V_rsvp_on) 385 imo = NULL; 386 if (ip_mforward && 387 ip_mforward(ip, ifp, m, imo) != 0) { 388 m_freem(m); 389 goto done; 390 } 391 } 392 } 393 394 /* 395 * Multicasts with a time-to-live of zero may be looped- 396 * back, above, but must not be transmitted on a network. 397 * Also, multicasts addressed to the loopback interface 398 * are not sent -- the above call to ip_mloopback() will 399 * loop back a copy. ip_input() will drop the copy if 400 * this host does not belong to the destination group on 401 * the loopback interface. 402 */ 403 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 404 m_freem(m); 405 goto done; 406 } 407 408 goto sendit; 409 } 410 411 /* 412 * If the source address is not specified yet, use the address 413 * of the outoing interface. 414 */ 415 if (ip->ip_src.s_addr == INADDR_ANY) { 416 /* Interface may have no addresses. */ 417 if (ia != NULL) { 418 ip->ip_src = IA_SIN(ia)->sin_addr; 419 } 420 } 421 422 /* 423 * Verify that we have any chance at all of being able to queue the 424 * packet or packet fragments, unless ALTQ is enabled on the given 425 * interface in which case packetdrop should be done by queueing. 426 */ 427#ifdef ALTQ 428 if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) && 429 ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 430 ifp->if_snd.ifq_maxlen)) 431#else 432 if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 433 ifp->if_snd.ifq_maxlen) 434#endif /* ALTQ */ 435 { 436 error = ENOBUFS; 437 IPSTAT_INC(ips_odropped); 438 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1); 439 goto bad; 440 } 441 442 /* 443 * Look for broadcast address and 444 * verify user is allowed to send 445 * such a packet. 446 */ 447 if (isbroadcast) { 448 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 449 error = EADDRNOTAVAIL; 450 goto bad; 451 } 452 if ((flags & IP_ALLOWBROADCAST) == 0) { 453 error = EACCES; 454 goto bad; 455 } 456 /* don't allow broadcast messages to be fragmented */ 457 if (ip->ip_len > mtu) { 458 error = EMSGSIZE; 459 goto bad; 460 } 461 m->m_flags |= M_BCAST; 462 } else { 463 m->m_flags &= ~M_BCAST; 464 } 465 466sendit: 467#ifdef IPSEC 468 switch(ip_ipsec_output(&m, inp, &flags, &error, &ifp)) { 469 case 1: 470 goto bad; 471 case -1: 472 goto done; 473 case 0: 474 default: 475 break; /* Continue with packet processing. */ 476 } 477 /* 478 * Check if there was a route for this packet; return error if not. 479 */ 480 if (no_route_but_check_spd) { 481 IPSTAT_INC(ips_noroute); 482 error = EHOSTUNREACH; 483 goto bad; 484 } 485 /* Update variables that are affected by ipsec4_output(). */ 486 ip = mtod(m, struct ip *); 487 hlen = ip->ip_hl << 2; 488#endif /* IPSEC */ 489 490 /* Jump over all PFIL processing if hooks are not active. */ 491 if (!PFIL_HOOKED(&inet_pfil_hook)) 492 goto passout; 493 494 /* Run through list of hooks for output packets. */ 495 odst.s_addr = ip->ip_dst.s_addr; 496 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 497 if (error != 0 || m == NULL) 498 goto done; 499 500 ip = mtod(m, struct ip *); 501 502 /* See if destination IP address was changed by packet filter. */ 503 if (odst.s_addr != ip->ip_dst.s_addr) { 504 m->m_flags |= M_SKIP_FIREWALL; 505 /* If destination is now ourself drop to ip_input(). */ 506 if (in_localip(ip->ip_dst)) { 507 m->m_flags |= M_FASTFWD_OURS; 508 if (m->m_pkthdr.rcvif == NULL) 509 m->m_pkthdr.rcvif = V_loif; 510 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 511 m->m_pkthdr.csum_flags |= 512 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 513 m->m_pkthdr.csum_data = 0xffff; 514 } 515 m->m_pkthdr.csum_flags |= 516 CSUM_IP_CHECKED | CSUM_IP_VALID; 517#ifdef SCTP 518 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 519 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 520#endif 521 error = netisr_queue(NETISR_IP, m); 522 goto done; 523 } else 524 goto again; /* Redo the routing table lookup. */ 525 } 526 527#ifdef IPFIREWALL_FORWARD 528 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 529 if (m->m_flags & M_FASTFWD_OURS) { 530 if (m->m_pkthdr.rcvif == NULL) 531 m->m_pkthdr.rcvif = V_loif; 532 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 533 m->m_pkthdr.csum_flags |= 534 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 535 m->m_pkthdr.csum_data = 0xffff; 536 } 537#ifdef SCTP 538 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 539 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 540#endif 541 m->m_pkthdr.csum_flags |= 542 CSUM_IP_CHECKED | CSUM_IP_VALID; 543 544 error = netisr_queue(NETISR_IP, m); 545 goto done; 546 } 547 /* Or forward to some other address? */ 548 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 549 if (fwd_tag) { 550 dst = (struct sockaddr_in *)&ro->ro_dst; 551 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 552 m->m_flags |= M_SKIP_FIREWALL; 553 m_tag_delete(m, fwd_tag); 554 goto again; 555 } 556#endif /* IPFIREWALL_FORWARD */ 557 558passout: 559 /* 127/8 must not appear on wire - RFC1122. */ 560 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 561 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 562 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 563 IPSTAT_INC(ips_badaddr); 564 error = EADDRNOTAVAIL; 565 goto bad; 566 } 567 } 568 569 m->m_pkthdr.csum_flags |= CSUM_IP; 570 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 571 if (sw_csum & CSUM_DELAY_DATA) { 572 in_delayed_cksum(m); 573 sw_csum &= ~CSUM_DELAY_DATA; 574 } 575#ifdef SCTP 576 if (sw_csum & CSUM_SCTP) { 577 sctp_delayed_cksum(m); 578 sw_csum &= ~CSUM_SCTP; 579 } 580#endif 581 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 582 583 /* 584 * If small enough for interface, or the interface will take 585 * care of the fragmentation for us, we can just send directly. 586 */ 587 if (ip->ip_len <= mtu || 588 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 589 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 590 ip->ip_len = htons(ip->ip_len); 591 ip->ip_off = htons(ip->ip_off); 592 ip->ip_sum = 0; 593 if (sw_csum & CSUM_DELAY_IP) 594 ip->ip_sum = in_cksum(m, hlen); 595 596 /* 597 * Record statistics for this interface address. 598 * With CSUM_TSO the byte/packet count will be slightly 599 * incorrect because we count the IP+TCP headers only 600 * once instead of for every generated packet. 601 */ 602 if (!(flags & IP_FORWARDING) && ia) { 603 if (m->m_pkthdr.csum_flags & CSUM_TSO) 604 ia->ia_ifa.if_opackets += 605 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 606 else 607 ia->ia_ifa.if_opackets++; 608 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 609 } 610#ifdef MBUF_STRESS_TEST 611 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 612 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 613#endif 614 /* 615 * Reset layer specific mbuf flags 616 * to avoid confusing lower layers. 617 */ 618 m->m_flags &= ~(M_PROTOFLAGS); 619 error = (*ifp->if_output)(ifp, m, 620 (struct sockaddr *)dst, ro); 621 goto done; 622 } 623 624 /* Balk when DF bit is set or the interface didn't support TSO. */ 625 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 626 error = EMSGSIZE; 627 IPSTAT_INC(ips_cantfrag); 628 goto bad; 629 } 630 631 /* 632 * Too large for interface; fragment if possible. If successful, 633 * on return, m will point to a list of packets to be sent. 634 */ 635 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 636 if (error) 637 goto bad; 638 for (; m; m = m0) { 639 m0 = m->m_nextpkt; 640 m->m_nextpkt = 0; 641 if (error == 0) { 642 /* Record statistics for this interface address. */ 643 if (ia != NULL) { 644 ia->ia_ifa.if_opackets++; 645 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 646 } 647 /* 648 * Reset layer specific mbuf flags 649 * to avoid confusing upper layers. 650 */ 651 m->m_flags &= ~(M_PROTOFLAGS); 652 653 error = (*ifp->if_output)(ifp, m, 654 (struct sockaddr *)dst, ro); 655 } else 656 m_freem(m); 657 } 658 659 if (error == 0) 660 IPSTAT_INC(ips_fragmented); 661 662done: 663 if (ro == &iproute && ro->ro_rt && !nortfree) { 664 RTFREE(ro->ro_rt); 665 } 666 if (ia != NULL) 667 ifa_free(&ia->ia_ifa); 668 return (error); 669bad: 670 m_freem(m); 671 goto done; 672} 673 674/* 675 * Create a chain of fragments which fit the given mtu. m_frag points to the 676 * mbuf to be fragmented; on return it points to the chain with the fragments. 677 * Return 0 if no error. If error, m_frag may contain a partially built 678 * chain of fragments that should be freed by the caller. 679 * 680 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 681 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 682 */ 683int 684ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 685 u_long if_hwassist_flags, int sw_csum) 686{ 687 int error = 0; 688 int hlen = ip->ip_hl << 2; 689 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 690 int off; 691 struct mbuf *m0 = *m_frag; /* the original packet */ 692 int firstlen; 693 struct mbuf **mnext; 694 int nfrags; 695 696 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 697 IPSTAT_INC(ips_cantfrag); 698 return EMSGSIZE; 699 } 700 701 /* 702 * Must be able to put at least 8 bytes per fragment. 703 */ 704 if (len < 8) 705 return EMSGSIZE; 706 707 /* 708 * If the interface will not calculate checksums on 709 * fragmented packets, then do it here. 710 */ 711 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 712 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 713 in_delayed_cksum(m0); 714 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 715 } 716#ifdef SCTP 717 if (m0->m_pkthdr.csum_flags & CSUM_SCTP && 718 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 719 sctp_delayed_cksum(m0); 720 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 721 } 722#endif 723 if (len > PAGE_SIZE) { 724 /* 725 * Fragment large datagrams such that each segment 726 * contains a multiple of PAGE_SIZE amount of data, 727 * plus headers. This enables a receiver to perform 728 * page-flipping zero-copy optimizations. 729 * 730 * XXX When does this help given that sender and receiver 731 * could have different page sizes, and also mtu could 732 * be less than the receiver's page size ? 733 */ 734 int newlen; 735 struct mbuf *m; 736 737 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 738 off += m->m_len; 739 740 /* 741 * firstlen (off - hlen) must be aligned on an 742 * 8-byte boundary 743 */ 744 if (off < hlen) 745 goto smart_frag_failure; 746 off = ((off - hlen) & ~7) + hlen; 747 newlen = (~PAGE_MASK) & mtu; 748 if ((newlen + sizeof (struct ip)) > mtu) { 749 /* we failed, go back the default */ 750smart_frag_failure: 751 newlen = len; 752 off = hlen + len; 753 } 754 len = newlen; 755 756 } else { 757 off = hlen + len; 758 } 759 760 firstlen = off - hlen; 761 mnext = &m0->m_nextpkt; /* pointer to next packet */ 762 763 /* 764 * Loop through length of segment after first fragment, 765 * make new header and copy data of each part and link onto chain. 766 * Here, m0 is the original packet, m is the fragment being created. 767 * The fragments are linked off the m_nextpkt of the original 768 * packet, which after processing serves as the first fragment. 769 */ 770 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 771 struct ip *mhip; /* ip header on the fragment */ 772 struct mbuf *m; 773 int mhlen = sizeof (struct ip); 774 775 MGETHDR(m, M_DONTWAIT, MT_DATA); 776 if (m == NULL) { 777 error = ENOBUFS; 778 IPSTAT_INC(ips_odropped); 779 goto done; 780 } 781 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 782 /* 783 * In the first mbuf, leave room for the link header, then 784 * copy the original IP header including options. The payload 785 * goes into an additional mbuf chain returned by m_copym(). 786 */ 787 m->m_data += max_linkhdr; 788 mhip = mtod(m, struct ip *); 789 *mhip = *ip; 790 if (hlen > sizeof (struct ip)) { 791 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 792 mhip->ip_v = IPVERSION; 793 mhip->ip_hl = mhlen >> 2; 794 } 795 m->m_len = mhlen; 796 /* XXX do we need to add ip->ip_off below ? */ 797 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 798 if (off + len >= ip->ip_len) { /* last fragment */ 799 len = ip->ip_len - off; 800 m->m_flags |= M_LASTFRAG; 801 } else 802 mhip->ip_off |= IP_MF; 803 mhip->ip_len = htons((u_short)(len + mhlen)); 804 m->m_next = m_copym(m0, off, len, M_DONTWAIT); 805 if (m->m_next == NULL) { /* copy failed */ 806 m_free(m); 807 error = ENOBUFS; /* ??? */ 808 IPSTAT_INC(ips_odropped); 809 goto done; 810 } 811 m->m_pkthdr.len = mhlen + len; 812 m->m_pkthdr.rcvif = NULL; 813#ifdef MAC 814 mac_netinet_fragment(m0, m); 815#endif 816 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 817 mhip->ip_off = htons(mhip->ip_off); 818 mhip->ip_sum = 0; 819 if (sw_csum & CSUM_DELAY_IP) 820 mhip->ip_sum = in_cksum(m, mhlen); 821 *mnext = m; 822 mnext = &m->m_nextpkt; 823 } 824 IPSTAT_ADD(ips_ofragments, nfrags); 825 826 /* set first marker for fragment chain */ 827 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 828 m0->m_pkthdr.csum_data = nfrags; 829 830 /* 831 * Update first fragment by trimming what's been copied out 832 * and updating header. 833 */ 834 m_adj(m0, hlen + firstlen - ip->ip_len); 835 m0->m_pkthdr.len = hlen + firstlen; 836 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 837 ip->ip_off |= IP_MF; 838 ip->ip_off = htons(ip->ip_off); 839 ip->ip_sum = 0; 840 if (sw_csum & CSUM_DELAY_IP) 841 ip->ip_sum = in_cksum(m0, hlen); 842 843done: 844 *m_frag = m0; 845 return error; 846} 847 848void 849in_delayed_cksum(struct mbuf *m) 850{ 851 struct ip *ip; 852 u_short csum, offset; 853 854 ip = mtod(m, struct ip *); 855 offset = ip->ip_hl << 2 ; 856 csum = in_cksum_skip(m, ip->ip_len, offset); 857 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 858 csum = 0xffff; 859 offset += m->m_pkthdr.csum_data; /* checksum offset */ 860 861 if (offset + sizeof(u_short) > m->m_len) { 862 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 863 m->m_len, offset, ip->ip_p); 864 /* 865 * XXX 866 * this shouldn't happen, but if it does, the 867 * correct behavior may be to insert the checksum 868 * in the appropriate next mbuf in the chain. 869 */ 870 return; 871 } 872 *(u_short *)(m->m_data + offset) = csum; 873} 874 875/* 876 * IP socket option processing. 877 */ 878int 879ip_ctloutput(struct socket *so, struct sockopt *sopt) 880{ 881 struct inpcb *inp = sotoinpcb(so); 882 int error, optval; 883 884 error = optval = 0; 885 if (sopt->sopt_level != IPPROTO_IP) { 886 if ((sopt->sopt_level == SOL_SOCKET) && 887 (sopt->sopt_name == SO_SETFIB)) { 888 inp->inp_inc.inc_fibnum = so->so_fibnum; 889 return (0); 890 } 891 return (EINVAL); 892 } 893 894 switch (sopt->sopt_dir) { 895 case SOPT_SET: 896 switch (sopt->sopt_name) { 897 case IP_OPTIONS: 898#ifdef notyet 899 case IP_RETOPTS: 900#endif 901 { 902 struct mbuf *m; 903 if (sopt->sopt_valsize > MLEN) { 904 error = EMSGSIZE; 905 break; 906 } 907 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 908 if (m == NULL) { 909 error = ENOBUFS; 910 break; 911 } 912 m->m_len = sopt->sopt_valsize; 913 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 914 m->m_len); 915 if (error) { 916 m_free(m); 917 break; 918 } 919 INP_WLOCK(inp); 920 error = ip_pcbopts(inp, sopt->sopt_name, m); 921 INP_WUNLOCK(inp); 922 return (error); 923 } 924 925 case IP_BINDANY: 926 if (sopt->sopt_td != NULL) { 927 error = priv_check(sopt->sopt_td, 928 PRIV_NETINET_BINDANY); 929 if (error) 930 break; 931 } 932 /* FALLTHROUGH */ 933 case IP_TOS: 934 case IP_TTL: 935 case IP_MINTTL: 936 case IP_RECVOPTS: 937 case IP_RECVRETOPTS: 938 case IP_RECVDSTADDR: 939 case IP_RECVTTL: 940 case IP_RECVIF: 941 case IP_FAITH: 942 case IP_ONESBCAST: 943 case IP_DONTFRAG: 944 error = sooptcopyin(sopt, &optval, sizeof optval, 945 sizeof optval); 946 if (error) 947 break; 948 949 switch (sopt->sopt_name) { 950 case IP_TOS: 951 inp->inp_ip_tos = optval; 952 break; 953 954 case IP_TTL: 955 inp->inp_ip_ttl = optval; 956 break; 957 958 case IP_MINTTL: 959 if (optval >= 0 && optval <= MAXTTL) 960 inp->inp_ip_minttl = optval; 961 else 962 error = EINVAL; 963 break; 964 965#define OPTSET(bit) do { \ 966 INP_WLOCK(inp); \ 967 if (optval) \ 968 inp->inp_flags |= bit; \ 969 else \ 970 inp->inp_flags &= ~bit; \ 971 INP_WUNLOCK(inp); \ 972} while (0) 973 974 case IP_RECVOPTS: 975 OPTSET(INP_RECVOPTS); 976 break; 977 978 case IP_RECVRETOPTS: 979 OPTSET(INP_RECVRETOPTS); 980 break; 981 982 case IP_RECVDSTADDR: 983 OPTSET(INP_RECVDSTADDR); 984 break; 985 986 case IP_RECVTTL: 987 OPTSET(INP_RECVTTL); 988 break; 989 990 case IP_RECVIF: 991 OPTSET(INP_RECVIF); 992 break; 993 994 case IP_FAITH: 995 OPTSET(INP_FAITH); 996 break; 997 998 case IP_ONESBCAST: 999 OPTSET(INP_ONESBCAST); 1000 break; 1001 case IP_DONTFRAG: 1002 OPTSET(INP_DONTFRAG); 1003 break; 1004 case IP_BINDANY: 1005 OPTSET(INP_BINDANY); 1006 break; 1007 } 1008 break; 1009#undef OPTSET 1010 1011 /* 1012 * Multicast socket options are processed by the in_mcast 1013 * module. 1014 */ 1015 case IP_MULTICAST_IF: 1016 case IP_MULTICAST_VIF: 1017 case IP_MULTICAST_TTL: 1018 case IP_MULTICAST_LOOP: 1019 case IP_ADD_MEMBERSHIP: 1020 case IP_DROP_MEMBERSHIP: 1021 case IP_ADD_SOURCE_MEMBERSHIP: 1022 case IP_DROP_SOURCE_MEMBERSHIP: 1023 case IP_BLOCK_SOURCE: 1024 case IP_UNBLOCK_SOURCE: 1025 case IP_MSFILTER: 1026 case MCAST_JOIN_GROUP: 1027 case MCAST_LEAVE_GROUP: 1028 case MCAST_JOIN_SOURCE_GROUP: 1029 case MCAST_LEAVE_SOURCE_GROUP: 1030 case MCAST_BLOCK_SOURCE: 1031 case MCAST_UNBLOCK_SOURCE: 1032 error = inp_setmoptions(inp, sopt); 1033 break; 1034 1035 case IP_PORTRANGE: 1036 error = sooptcopyin(sopt, &optval, sizeof optval, 1037 sizeof optval); 1038 if (error) 1039 break; 1040 1041 INP_WLOCK(inp); 1042 switch (optval) { 1043 case IP_PORTRANGE_DEFAULT: 1044 inp->inp_flags &= ~(INP_LOWPORT); 1045 inp->inp_flags &= ~(INP_HIGHPORT); 1046 break; 1047 1048 case IP_PORTRANGE_HIGH: 1049 inp->inp_flags &= ~(INP_LOWPORT); 1050 inp->inp_flags |= INP_HIGHPORT; 1051 break; 1052 1053 case IP_PORTRANGE_LOW: 1054 inp->inp_flags &= ~(INP_HIGHPORT); 1055 inp->inp_flags |= INP_LOWPORT; 1056 break; 1057 1058 default: 1059 error = EINVAL; 1060 break; 1061 } 1062 INP_WUNLOCK(inp); 1063 break; 1064 1065#ifdef IPSEC 1066 case IP_IPSEC_POLICY: 1067 { 1068 caddr_t req; 1069 struct mbuf *m; 1070 1071 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1072 break; 1073 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1074 break; 1075 req = mtod(m, caddr_t); 1076 error = ipsec_set_policy(inp, sopt->sopt_name, req, 1077 m->m_len, (sopt->sopt_td != NULL) ? 1078 sopt->sopt_td->td_ucred : NULL); 1079 m_freem(m); 1080 break; 1081 } 1082#endif /* IPSEC */ 1083 1084 default: 1085 error = ENOPROTOOPT; 1086 break; 1087 } 1088 break; 1089 1090 case SOPT_GET: 1091 switch (sopt->sopt_name) { 1092 case IP_OPTIONS: 1093 case IP_RETOPTS: 1094 if (inp->inp_options) 1095 error = sooptcopyout(sopt, 1096 mtod(inp->inp_options, 1097 char *), 1098 inp->inp_options->m_len); 1099 else 1100 sopt->sopt_valsize = 0; 1101 break; 1102 1103 case IP_TOS: 1104 case IP_TTL: 1105 case IP_MINTTL: 1106 case IP_RECVOPTS: 1107 case IP_RECVRETOPTS: 1108 case IP_RECVDSTADDR: 1109 case IP_RECVTTL: 1110 case IP_RECVIF: 1111 case IP_PORTRANGE: 1112 case IP_FAITH: 1113 case IP_ONESBCAST: 1114 case IP_DONTFRAG: 1115 switch (sopt->sopt_name) { 1116 1117 case IP_TOS: 1118 optval = inp->inp_ip_tos; 1119 break; 1120 1121 case IP_TTL: 1122 optval = inp->inp_ip_ttl; 1123 break; 1124 1125 case IP_MINTTL: 1126 optval = inp->inp_ip_minttl; 1127 break; 1128 1129#define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1130 1131 case IP_RECVOPTS: 1132 optval = OPTBIT(INP_RECVOPTS); 1133 break; 1134 1135 case IP_RECVRETOPTS: 1136 optval = OPTBIT(INP_RECVRETOPTS); 1137 break; 1138 1139 case IP_RECVDSTADDR: 1140 optval = OPTBIT(INP_RECVDSTADDR); 1141 break; 1142 1143 case IP_RECVTTL: 1144 optval = OPTBIT(INP_RECVTTL); 1145 break; 1146 1147 case IP_RECVIF: 1148 optval = OPTBIT(INP_RECVIF); 1149 break; 1150 1151 case IP_PORTRANGE: 1152 if (inp->inp_flags & INP_HIGHPORT) 1153 optval = IP_PORTRANGE_HIGH; 1154 else if (inp->inp_flags & INP_LOWPORT) 1155 optval = IP_PORTRANGE_LOW; 1156 else 1157 optval = 0; 1158 break; 1159 1160 case IP_FAITH: 1161 optval = OPTBIT(INP_FAITH); 1162 break; 1163 1164 case IP_ONESBCAST: 1165 optval = OPTBIT(INP_ONESBCAST); 1166 break; 1167 case IP_DONTFRAG: 1168 optval = OPTBIT(INP_DONTFRAG); 1169 break; 1170 } 1171 error = sooptcopyout(sopt, &optval, sizeof optval); 1172 break; 1173 1174 /* 1175 * Multicast socket options are processed by the in_mcast 1176 * module. 1177 */ 1178 case IP_MULTICAST_IF: 1179 case IP_MULTICAST_VIF: 1180 case IP_MULTICAST_TTL: 1181 case IP_MULTICAST_LOOP: 1182 case IP_MSFILTER: 1183 error = inp_getmoptions(inp, sopt); 1184 break; 1185 1186#ifdef IPSEC 1187 case IP_IPSEC_POLICY: 1188 { 1189 struct mbuf *m = NULL; 1190 caddr_t req = NULL; 1191 size_t len = 0; 1192 1193 if (m != 0) { 1194 req = mtod(m, caddr_t); 1195 len = m->m_len; 1196 } 1197 error = ipsec_get_policy(sotoinpcb(so), req, len, &m); 1198 if (error == 0) 1199 error = soopt_mcopyout(sopt, m); /* XXX */ 1200 if (error == 0) 1201 m_freem(m); 1202 break; 1203 } 1204#endif /* IPSEC */ 1205 1206 default: 1207 error = ENOPROTOOPT; 1208 break; 1209 } 1210 break; 1211 } 1212 return (error); 1213} 1214 1215/* 1216 * Routine called from ip_output() to loop back a copy of an IP multicast 1217 * packet to the input queue of a specified interface. Note that this 1218 * calls the output routine of the loopback "driver", but with an interface 1219 * pointer that might NOT be a loopback interface -- evil, but easier than 1220 * replicating that code here. 1221 */ 1222static void 1223ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1224 int hlen) 1225{ 1226 register struct ip *ip; 1227 struct mbuf *copym; 1228 1229 /* 1230 * Make a deep copy of the packet because we're going to 1231 * modify the pack in order to generate checksums. 1232 */ 1233 copym = m_dup(m, M_DONTWAIT); 1234 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1235 copym = m_pullup(copym, hlen); 1236 if (copym != NULL) { 1237 /* If needed, compute the checksum and mark it as valid. */ 1238 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1239 in_delayed_cksum(copym); 1240 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1241 copym->m_pkthdr.csum_flags |= 1242 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1243 copym->m_pkthdr.csum_data = 0xffff; 1244 } 1245 /* 1246 * We don't bother to fragment if the IP length is greater 1247 * than the interface's MTU. Can this possibly matter? 1248 */ 1249 ip = mtod(copym, struct ip *); 1250 ip->ip_len = htons(ip->ip_len); 1251 ip->ip_off = htons(ip->ip_off); 1252 ip->ip_sum = 0; 1253 ip->ip_sum = in_cksum(copym, hlen); 1254#if 1 /* XXX */ 1255 if (dst->sin_family != AF_INET) { 1256 printf("ip_mloopback: bad address family %d\n", 1257 dst->sin_family); 1258 dst->sin_family = AF_INET; 1259 } 1260#endif 1261 if_simloop(ifp, copym, dst->sin_family, 0); 1262 } 1263} 1264