ip_output.c revision 196019
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 196019 2009-08-01 19:26:27Z 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 55#include <net/if.h> 56#include <net/netisr.h> 57#include <net/pfil.h> 58#include <net/route.h> 59#include <net/flowtable.h> 60#ifdef RADIX_MPATH 61#include <net/radix_mpath.h> 62#endif 63#include <net/vnet.h> 64 65#include <netinet/in.h> 66#include <netinet/in_systm.h> 67#include <netinet/ip.h> 68#include <netinet/in_pcb.h> 69#include <netinet/in_var.h> 70#include <netinet/ip_var.h> 71#include <netinet/ip_options.h> 72#ifdef SCTP 73#include <netinet/sctp.h> 74#include <netinet/sctp_crc32.h> 75#endif 76 77#ifdef IPSEC 78#include <netinet/ip_ipsec.h> 79#include <netipsec/ipsec.h> 80#endif /* IPSEC*/ 81 82#include <machine/in_cksum.h> 83 84#include <security/mac/mac_framework.h> 85 86#define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 87 x, (ntohl(a.s_addr)>>24)&0xFF,\ 88 (ntohl(a.s_addr)>>16)&0xFF,\ 89 (ntohl(a.s_addr)>>8)&0xFF,\ 90 (ntohl(a.s_addr))&0xFF, y); 91 92VNET_DEFINE(u_short, ip_id); 93 94#ifdef MBUF_STRESS_TEST 95int mbuf_frag_size = 0; 96SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 97 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 98#endif 99 100static void ip_mloopback 101 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 102 103 104extern int in_mcast_loop; 105extern struct protosw inetsw[]; 106 107/* 108 * IP output. The packet in mbuf chain m contains a skeletal IP 109 * header (with len, off, ttl, proto, tos, src, dst). 110 * The mbuf chain containing the packet will be freed. 111 * The mbuf opt, if present, will not be freed. 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 len, error = 0; 125 int nortfree = 0; 126 struct sockaddr_in *dst = NULL; /* keep compiler happy */ 127 struct in_ifaddr *ia = NULL; 128 int isbroadcast, sw_csum; 129 struct route iproute; 130 struct in_addr odst; 131#ifdef IPFIREWALL_FORWARD 132 struct m_tag *fwd_tag = NULL; 133#endif 134#ifdef IPSEC 135 int no_route_but_check_spd = 0; 136#endif 137 M_ASSERTPKTHDR(m); 138 139 if (inp != NULL) { 140 INP_LOCK_ASSERT(inp); 141 M_SETFIB(m, inp->inp_inc.inc_fibnum); 142 if (inp->inp_flags & (INP_HW_FLOWID|INP_SW_FLOWID)) { 143 m->m_pkthdr.flowid = inp->inp_flowid; 144 m->m_flags |= M_FLOWID; 145 } 146 } 147 148 if (ro == NULL) { 149 ro = &iproute; 150 bzero(ro, sizeof (*ro)); 151 152#ifdef FLOWTABLE 153 /* 154 * The flow table returns route entries valid for up to 30 155 * seconds; we rely on the remainder of ip_output() taking no 156 * longer than that long for the stability of ro_rt. The 157 * flow ID assignment must have happened before this point. 158 */ 159 if (flowtable_lookup(V_ip_ft, m, ro) == 0) 160 nortfree = 1; 161#endif 162 } 163 164 if (opt) { 165 len = 0; 166 m = ip_insertoptions(m, opt, &len); 167 if (len != 0) 168 hlen = len; 169 } 170 ip = mtod(m, struct ip *); 171 172 /* 173 * Fill in IP header. If we are not allowing fragmentation, 174 * then the ip_id field is meaningless, but we don't set it 175 * to zero. Doing so causes various problems when devices along 176 * the path (routers, load balancers, firewalls, etc.) illegally 177 * disable DF on our packet. Note that a 16-bit counter 178 * will wrap around in less than 10 seconds at 100 Mbit/s on a 179 * medium with MTU 1500. See Steven M. Bellovin, "A Technique 180 * for Counting NATted Hosts", Proc. IMW'02, available at 181 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>. 182 */ 183 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 184 ip->ip_v = IPVERSION; 185 ip->ip_hl = hlen >> 2; 186 ip->ip_id = ip_newid(); 187 IPSTAT_INC(ips_localout); 188 } else { 189 hlen = ip->ip_hl << 2; 190 } 191 192 dst = (struct sockaddr_in *)&ro->ro_dst; 193again: 194 /* 195 * If there is a cached route, 196 * check that it is to the same destination 197 * and is still up. If not, free it and try again. 198 * The address family should also be checked in case of sharing the 199 * cache with IPv6. 200 */ 201 if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 202 dst->sin_family != AF_INET || 203 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 204 if (!nortfree) 205 RTFREE(ro->ro_rt); 206 ro->ro_rt = (struct rtentry *)NULL; 207 } 208#ifdef IPFIREWALL_FORWARD 209 if (ro->ro_rt == NULL && fwd_tag == NULL) { 210#else 211 if (ro->ro_rt == NULL) { 212#endif 213 bzero(dst, sizeof(*dst)); 214 dst->sin_family = AF_INET; 215 dst->sin_len = sizeof(*dst); 216 dst->sin_addr = ip->ip_dst; 217 } 218 /* 219 * If routing to interface only, short circuit routing lookup. 220 * The use of an all-ones broadcast address implies this; an 221 * interface is specified by the broadcast address of an interface, 222 * or the destination address of a ptp interface. 223 */ 224 if (flags & IP_SENDONES) { 225 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 226 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 227 IPSTAT_INC(ips_noroute); 228 error = ENETUNREACH; 229 goto bad; 230 } 231 ip->ip_dst.s_addr = INADDR_BROADCAST; 232 dst->sin_addr = ip->ip_dst; 233 ifp = ia->ia_ifp; 234 ip->ip_ttl = 1; 235 isbroadcast = 1; 236 } else if (flags & IP_ROUTETOIF) { 237 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 238 (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) { 239 IPSTAT_INC(ips_noroute); 240 error = ENETUNREACH; 241 goto bad; 242 } 243 ifp = ia->ia_ifp; 244 ip->ip_ttl = 1; 245 isbroadcast = in_broadcast(dst->sin_addr, ifp); 246 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 247 imo != NULL && imo->imo_multicast_ifp != NULL) { 248 /* 249 * Bypass the normal routing lookup for multicast 250 * packets if the interface is specified. 251 */ 252 ifp = imo->imo_multicast_ifp; 253 IFP_TO_IA(ifp, ia); 254 isbroadcast = 0; /* fool gcc */ 255 } else { 256 /* 257 * We want to do any cloning requested by the link layer, 258 * as this is probably required in all cases for correct 259 * operation (as it is for ARP). 260 */ 261 if (ro->ro_rt == NULL) 262#ifdef RADIX_MPATH 263 rtalloc_mpath_fib(ro, 264 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 265 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 266#else 267 in_rtalloc_ign(ro, 0, 268 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 269#endif 270 if (ro->ro_rt == NULL) { 271#ifdef IPSEC 272 /* 273 * There is no route for this packet, but it is 274 * possible that a matching SPD entry exists. 275 */ 276 no_route_but_check_spd = 1; 277 mtu = 0; /* Silence GCC warning. */ 278 goto sendit; 279#endif 280 IPSTAT_INC(ips_noroute); 281 error = EHOSTUNREACH; 282 goto bad; 283 } 284 ia = ifatoia(ro->ro_rt->rt_ifa); 285 ifa_ref(&ia->ia_ifa); 286 ifp = ro->ro_rt->rt_ifp; 287 ro->ro_rt->rt_rmx.rmx_pksent++; 288 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 289 dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 290 if (ro->ro_rt->rt_flags & RTF_HOST) 291 isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 292 else 293 isbroadcast = in_broadcast(dst->sin_addr, ifp); 294 } 295 /* 296 * Calculate MTU. If we have a route that is up, use that, 297 * otherwise use the interface's MTU. 298 */ 299 if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) { 300 /* 301 * This case can happen if the user changed the MTU 302 * of an interface after enabling IP on it. Because 303 * most netifs don't keep track of routes pointing to 304 * them, there is no way for one to update all its 305 * routes when the MTU is changed. 306 */ 307 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu) 308 ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 309 mtu = ro->ro_rt->rt_rmx.rmx_mtu; 310 } else { 311 mtu = ifp->if_mtu; 312 } 313 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 314 m->m_flags |= M_MCAST; 315 /* 316 * IP destination address is multicast. Make sure "dst" 317 * still points to the address in "ro". (It may have been 318 * changed to point to a gateway address, above.) 319 */ 320 dst = (struct sockaddr_in *)&ro->ro_dst; 321 /* 322 * See if the caller provided any multicast options 323 */ 324 if (imo != NULL) { 325 ip->ip_ttl = imo->imo_multicast_ttl; 326 if (imo->imo_multicast_vif != -1) 327 ip->ip_src.s_addr = 328 ip_mcast_src ? 329 ip_mcast_src(imo->imo_multicast_vif) : 330 INADDR_ANY; 331 } else 332 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 333 /* 334 * Confirm that the outgoing interface supports multicast. 335 */ 336 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 337 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 338 IPSTAT_INC(ips_noroute); 339 error = ENETUNREACH; 340 goto bad; 341 } 342 } 343 /* 344 * If source address not specified yet, use address 345 * of outgoing interface. 346 */ 347 if (ip->ip_src.s_addr == INADDR_ANY) { 348 /* Interface may have no addresses. */ 349 if (ia != NULL) 350 ip->ip_src = IA_SIN(ia)->sin_addr; 351 } 352 353 if ((imo == NULL && in_mcast_loop) || 354 (imo && imo->imo_multicast_loop)) { 355 /* 356 * Loop back multicast datagram if not expressly 357 * forbidden to do so, even if we are not a member 358 * of the group; ip_input() will filter it later, 359 * thus deferring a hash lookup and mutex acquisition 360 * at the expense of a cheap copy using m_copym(). 361 */ 362 ip_mloopback(ifp, m, dst, hlen); 363 } else { 364 /* 365 * If we are acting as a multicast router, perform 366 * multicast forwarding as if the packet had just 367 * arrived on the interface to which we are about 368 * to send. The multicast forwarding function 369 * recursively calls this function, using the 370 * IP_FORWARDING flag to prevent infinite recursion. 371 * 372 * Multicasts that are looped back by ip_mloopback(), 373 * above, will be forwarded by the ip_input() routine, 374 * if necessary. 375 */ 376 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 377 /* 378 * If rsvp daemon is not running, do not 379 * set ip_moptions. This ensures that the packet 380 * is multicast and not just sent down one link 381 * as prescribed by rsvpd. 382 */ 383 if (!V_rsvp_on) 384 imo = NULL; 385 if (ip_mforward && 386 ip_mforward(ip, ifp, m, imo) != 0) { 387 m_freem(m); 388 goto done; 389 } 390 } 391 } 392 393 /* 394 * Multicasts with a time-to-live of zero may be looped- 395 * back, above, but must not be transmitted on a network. 396 * Also, multicasts addressed to the loopback interface 397 * are not sent -- the above call to ip_mloopback() will 398 * loop back a copy. ip_input() will drop the copy if 399 * this host does not belong to the destination group on 400 * the loopback interface. 401 */ 402 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 403 m_freem(m); 404 goto done; 405 } 406 407 goto sendit; 408 } 409 410 /* 411 * If the source address is not specified yet, use the address 412 * of the outoing interface. 413 */ 414 if (ip->ip_src.s_addr == INADDR_ANY) { 415 /* Interface may have no addresses. */ 416 if (ia != NULL) { 417 ip->ip_src = IA_SIN(ia)->sin_addr; 418 } 419 } 420 421 /* 422 * Verify that we have any chance at all of being able to queue the 423 * packet or packet fragments, unless ALTQ is enabled on the given 424 * interface in which case packetdrop should be done by queueing. 425 */ 426#ifdef ALTQ 427 if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) && 428 ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 429 ifp->if_snd.ifq_maxlen)) 430#else 431 if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 432 ifp->if_snd.ifq_maxlen) 433#endif /* ALTQ */ 434 { 435 error = ENOBUFS; 436 IPSTAT_INC(ips_odropped); 437 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1); 438 goto bad; 439 } 440 441 /* 442 * Look for broadcast address and 443 * verify user is allowed to send 444 * such a packet. 445 */ 446 if (isbroadcast) { 447 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 448 error = EADDRNOTAVAIL; 449 goto bad; 450 } 451 if ((flags & IP_ALLOWBROADCAST) == 0) { 452 error = EACCES; 453 goto bad; 454 } 455 /* don't allow broadcast messages to be fragmented */ 456 if (ip->ip_len > mtu) { 457 error = EMSGSIZE; 458 goto bad; 459 } 460 m->m_flags |= M_BCAST; 461 } else { 462 m->m_flags &= ~M_BCAST; 463 } 464 465sendit: 466#ifdef IPSEC 467 switch(ip_ipsec_output(&m, inp, &flags, &error, &ifp)) { 468 case 1: 469 goto bad; 470 case -1: 471 goto done; 472 case 0: 473 default: 474 break; /* Continue with packet processing. */ 475 } 476 /* 477 * Check if there was a route for this packet; return error if not. 478 */ 479 if (no_route_but_check_spd) { 480 IPSTAT_INC(ips_noroute); 481 error = EHOSTUNREACH; 482 goto bad; 483 } 484 /* Update variables that are affected by ipsec4_output(). */ 485 ip = mtod(m, struct ip *); 486 hlen = ip->ip_hl << 2; 487#endif /* IPSEC */ 488 489 /* Jump over all PFIL processing if hooks are not active. */ 490 if (!PFIL_HOOKED(&inet_pfil_hook)) 491 goto passout; 492 493 /* Run through list of hooks for output packets. */ 494 odst.s_addr = ip->ip_dst.s_addr; 495 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 496 if (error != 0 || m == NULL) 497 goto done; 498 499 ip = mtod(m, struct ip *); 500 501 /* See if destination IP address was changed by packet filter. */ 502 if (odst.s_addr != ip->ip_dst.s_addr) { 503 m->m_flags |= M_SKIP_FIREWALL; 504 /* If destination is now ourself drop to ip_input(). */ 505 if (in_localip(ip->ip_dst)) { 506 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 m->m_pkthdr.csum_flags |= 515 CSUM_IP_CHECKED | CSUM_IP_VALID; 516#ifdef SCTP 517 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 518 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 519#endif 520 error = netisr_queue(NETISR_IP, m); 521 goto done; 522 } else 523 goto again; /* Redo the routing table lookup. */ 524 } 525 526#ifdef IPFIREWALL_FORWARD 527 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 528 if (m->m_flags & M_FASTFWD_OURS) { 529 if (m->m_pkthdr.rcvif == NULL) 530 m->m_pkthdr.rcvif = V_loif; 531 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 532 m->m_pkthdr.csum_flags |= 533 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 534 m->m_pkthdr.csum_data = 0xffff; 535 } 536#ifdef SCTP 537 if (m->m_pkthdr.csum_flags & CSUM_SCTP) 538 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 539#endif 540 m->m_pkthdr.csum_flags |= 541 CSUM_IP_CHECKED | CSUM_IP_VALID; 542 543 error = netisr_queue(NETISR_IP, m); 544 goto done; 545 } 546 /* Or forward to some other address? */ 547 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 548 if (fwd_tag) { 549 dst = (struct sockaddr_in *)&ro->ro_dst; 550 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 551 m->m_flags |= M_SKIP_FIREWALL; 552 m_tag_delete(m, fwd_tag); 553 goto again; 554 } 555#endif /* IPFIREWALL_FORWARD */ 556 557passout: 558 /* 127/8 must not appear on wire - RFC1122. */ 559 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 560 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 561 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 562 IPSTAT_INC(ips_badaddr); 563 error = EADDRNOTAVAIL; 564 goto bad; 565 } 566 } 567 568 m->m_pkthdr.csum_flags |= CSUM_IP; 569 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 570 if (sw_csum & CSUM_DELAY_DATA) { 571 in_delayed_cksum(m); 572 sw_csum &= ~CSUM_DELAY_DATA; 573 } 574#ifdef SCTP 575 if (sw_csum & CSUM_SCTP) { 576 sctp_delayed_cksum(m); 577 sw_csum &= ~CSUM_SCTP; 578 } 579#endif 580 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 581 582 /* 583 * If small enough for interface, or the interface will take 584 * care of the fragmentation for us, we can just send directly. 585 */ 586 if (ip->ip_len <= mtu || 587 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 588 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 589 ip->ip_len = htons(ip->ip_len); 590 ip->ip_off = htons(ip->ip_off); 591 ip->ip_sum = 0; 592 if (sw_csum & CSUM_DELAY_IP) 593 ip->ip_sum = in_cksum(m, hlen); 594 595 /* 596 * Record statistics for this interface address. 597 * With CSUM_TSO the byte/packet count will be slightly 598 * incorrect because we count the IP+TCP headers only 599 * once instead of for every generated packet. 600 */ 601 if (!(flags & IP_FORWARDING) && ia) { 602 if (m->m_pkthdr.csum_flags & CSUM_TSO) 603 ia->ia_ifa.if_opackets += 604 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 605 else 606 ia->ia_ifa.if_opackets++; 607 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 608 } 609#ifdef MBUF_STRESS_TEST 610 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 611 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 612#endif 613 /* 614 * Reset layer specific mbuf flags 615 * to avoid confusing lower layers. 616 */ 617 m->m_flags &= ~(M_PROTOFLAGS); 618 error = (*ifp->if_output)(ifp, m, 619 (struct sockaddr *)dst, ro); 620 goto done; 621 } 622 623 /* Balk when DF bit is set or the interface didn't support TSO. */ 624 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 625 error = EMSGSIZE; 626 IPSTAT_INC(ips_cantfrag); 627 goto bad; 628 } 629 630 /* 631 * Too large for interface; fragment if possible. If successful, 632 * on return, m will point to a list of packets to be sent. 633 */ 634 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 635 if (error) 636 goto bad; 637 for (; m; m = m0) { 638 m0 = m->m_nextpkt; 639 m->m_nextpkt = 0; 640 if (error == 0) { 641 /* Record statistics for this interface address. */ 642 if (ia != NULL) { 643 ia->ia_ifa.if_opackets++; 644 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 645 } 646 /* 647 * Reset layer specific mbuf flags 648 * to avoid confusing upper layers. 649 */ 650 m->m_flags &= ~(M_PROTOFLAGS); 651 652 error = (*ifp->if_output)(ifp, m, 653 (struct sockaddr *)dst, ro); 654 } else 655 m_freem(m); 656 } 657 658 if (error == 0) 659 IPSTAT_INC(ips_fragmented); 660 661done: 662 if (ro == &iproute && ro->ro_rt && !nortfree) { 663 RTFREE(ro->ro_rt); 664 } 665 if (ia != NULL) 666 ifa_free(&ia->ia_ifa); 667 return (error); 668bad: 669 m_freem(m); 670 goto done; 671} 672 673/* 674 * Create a chain of fragments which fit the given mtu. m_frag points to the 675 * mbuf to be fragmented; on return it points to the chain with the fragments. 676 * Return 0 if no error. If error, m_frag may contain a partially built 677 * chain of fragments that should be freed by the caller. 678 * 679 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 680 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 681 */ 682int 683ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 684 u_long if_hwassist_flags, int sw_csum) 685{ 686 int error = 0; 687 int hlen = ip->ip_hl << 2; 688 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 689 int off; 690 struct mbuf *m0 = *m_frag; /* the original packet */ 691 int firstlen; 692 struct mbuf **mnext; 693 int nfrags; 694 695 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 696 IPSTAT_INC(ips_cantfrag); 697 return EMSGSIZE; 698 } 699 700 /* 701 * Must be able to put at least 8 bytes per fragment. 702 */ 703 if (len < 8) 704 return EMSGSIZE; 705 706 /* 707 * If the interface will not calculate checksums on 708 * fragmented packets, then do it here. 709 */ 710 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 711 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 712 in_delayed_cksum(m0); 713 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 714 } 715#ifdef SCTP 716 if (m0->m_pkthdr.csum_flags & CSUM_SCTP && 717 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 718 sctp_delayed_cksum(m0); 719 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 720 } 721#endif 722 if (len > PAGE_SIZE) { 723 /* 724 * Fragment large datagrams such that each segment 725 * contains a multiple of PAGE_SIZE amount of data, 726 * plus headers. This enables a receiver to perform 727 * page-flipping zero-copy optimizations. 728 * 729 * XXX When does this help given that sender and receiver 730 * could have different page sizes, and also mtu could 731 * be less than the receiver's page size ? 732 */ 733 int newlen; 734 struct mbuf *m; 735 736 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 737 off += m->m_len; 738 739 /* 740 * firstlen (off - hlen) must be aligned on an 741 * 8-byte boundary 742 */ 743 if (off < hlen) 744 goto smart_frag_failure; 745 off = ((off - hlen) & ~7) + hlen; 746 newlen = (~PAGE_MASK) & mtu; 747 if ((newlen + sizeof (struct ip)) > mtu) { 748 /* we failed, go back the default */ 749smart_frag_failure: 750 newlen = len; 751 off = hlen + len; 752 } 753 len = newlen; 754 755 } else { 756 off = hlen + len; 757 } 758 759 firstlen = off - hlen; 760 mnext = &m0->m_nextpkt; /* pointer to next packet */ 761 762 /* 763 * Loop through length of segment after first fragment, 764 * make new header and copy data of each part and link onto chain. 765 * Here, m0 is the original packet, m is the fragment being created. 766 * The fragments are linked off the m_nextpkt of the original 767 * packet, which after processing serves as the first fragment. 768 */ 769 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 770 struct ip *mhip; /* ip header on the fragment */ 771 struct mbuf *m; 772 int mhlen = sizeof (struct ip); 773 774 MGETHDR(m, M_DONTWAIT, MT_DATA); 775 if (m == NULL) { 776 error = ENOBUFS; 777 IPSTAT_INC(ips_odropped); 778 goto done; 779 } 780 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 781 /* 782 * In the first mbuf, leave room for the link header, then 783 * copy the original IP header including options. The payload 784 * goes into an additional mbuf chain returned by m_copym(). 785 */ 786 m->m_data += max_linkhdr; 787 mhip = mtod(m, struct ip *); 788 *mhip = *ip; 789 if (hlen > sizeof (struct ip)) { 790 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 791 mhip->ip_v = IPVERSION; 792 mhip->ip_hl = mhlen >> 2; 793 } 794 m->m_len = mhlen; 795 /* XXX do we need to add ip->ip_off below ? */ 796 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 797 if (off + len >= ip->ip_len) { /* last fragment */ 798 len = ip->ip_len - off; 799 m->m_flags |= M_LASTFRAG; 800 } else 801 mhip->ip_off |= IP_MF; 802 mhip->ip_len = htons((u_short)(len + mhlen)); 803 m->m_next = m_copym(m0, off, len, M_DONTWAIT); 804 if (m->m_next == NULL) { /* copy failed */ 805 m_free(m); 806 error = ENOBUFS; /* ??? */ 807 IPSTAT_INC(ips_odropped); 808 goto done; 809 } 810 m->m_pkthdr.len = mhlen + len; 811 m->m_pkthdr.rcvif = NULL; 812#ifdef MAC 813 mac_netinet_fragment(m0, m); 814#endif 815 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 816 mhip->ip_off = htons(mhip->ip_off); 817 mhip->ip_sum = 0; 818 if (sw_csum & CSUM_DELAY_IP) 819 mhip->ip_sum = in_cksum(m, mhlen); 820 *mnext = m; 821 mnext = &m->m_nextpkt; 822 } 823 IPSTAT_ADD(ips_ofragments, nfrags); 824 825 /* set first marker for fragment chain */ 826 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 827 m0->m_pkthdr.csum_data = nfrags; 828 829 /* 830 * Update first fragment by trimming what's been copied out 831 * and updating header. 832 */ 833 m_adj(m0, hlen + firstlen - ip->ip_len); 834 m0->m_pkthdr.len = hlen + firstlen; 835 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 836 ip->ip_off |= IP_MF; 837 ip->ip_off = htons(ip->ip_off); 838 ip->ip_sum = 0; 839 if (sw_csum & CSUM_DELAY_IP) 840 ip->ip_sum = in_cksum(m0, hlen); 841 842done: 843 *m_frag = m0; 844 return error; 845} 846 847void 848in_delayed_cksum(struct mbuf *m) 849{ 850 struct ip *ip; 851 u_short csum, offset; 852 853 ip = mtod(m, struct ip *); 854 offset = ip->ip_hl << 2 ; 855 csum = in_cksum_skip(m, ip->ip_len, offset); 856 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 857 csum = 0xffff; 858 offset += m->m_pkthdr.csum_data; /* checksum offset */ 859 860 if (offset + sizeof(u_short) > m->m_len) { 861 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 862 m->m_len, offset, ip->ip_p); 863 /* 864 * XXX 865 * this shouldn't happen, but if it does, the 866 * correct behavior may be to insert the checksum 867 * in the appropriate next mbuf in the chain. 868 */ 869 return; 870 } 871 *(u_short *)(m->m_data + offset) = csum; 872} 873 874/* 875 * IP socket option processing. 876 */ 877int 878ip_ctloutput(struct socket *so, struct sockopt *sopt) 879{ 880 struct inpcb *inp = sotoinpcb(so); 881 int error, optval; 882 883 error = optval = 0; 884 if (sopt->sopt_level != IPPROTO_IP) { 885 if ((sopt->sopt_level == SOL_SOCKET) && 886 (sopt->sopt_name == SO_SETFIB)) { 887 inp->inp_inc.inc_fibnum = so->so_fibnum; 888 return (0); 889 } 890 return (EINVAL); 891 } 892 893 switch (sopt->sopt_dir) { 894 case SOPT_SET: 895 switch (sopt->sopt_name) { 896 case IP_OPTIONS: 897#ifdef notyet 898 case IP_RETOPTS: 899#endif 900 { 901 struct mbuf *m; 902 if (sopt->sopt_valsize > MLEN) { 903 error = EMSGSIZE; 904 break; 905 } 906 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 907 if (m == NULL) { 908 error = ENOBUFS; 909 break; 910 } 911 m->m_len = sopt->sopt_valsize; 912 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 913 m->m_len); 914 if (error) { 915 m_free(m); 916 break; 917 } 918 INP_WLOCK(inp); 919 error = ip_pcbopts(inp, sopt->sopt_name, m); 920 INP_WUNLOCK(inp); 921 return (error); 922 } 923 924 case IP_BINDANY: 925 if (sopt->sopt_td != NULL) { 926 error = priv_check(sopt->sopt_td, 927 PRIV_NETINET_BINDANY); 928 if (error) 929 break; 930 } 931 /* FALLTHROUGH */ 932 case IP_TOS: 933 case IP_TTL: 934 case IP_MINTTL: 935 case IP_RECVOPTS: 936 case IP_RECVRETOPTS: 937 case IP_RECVDSTADDR: 938 case IP_RECVTTL: 939 case IP_RECVIF: 940 case IP_FAITH: 941 case IP_ONESBCAST: 942 case IP_DONTFRAG: 943 error = sooptcopyin(sopt, &optval, sizeof optval, 944 sizeof optval); 945 if (error) 946 break; 947 948 switch (sopt->sopt_name) { 949 case IP_TOS: 950 inp->inp_ip_tos = optval; 951 break; 952 953 case IP_TTL: 954 inp->inp_ip_ttl = optval; 955 break; 956 957 case IP_MINTTL: 958 if (optval >= 0 && optval <= MAXTTL) 959 inp->inp_ip_minttl = optval; 960 else 961 error = EINVAL; 962 break; 963 964#define OPTSET(bit) do { \ 965 INP_WLOCK(inp); \ 966 if (optval) \ 967 inp->inp_flags |= bit; \ 968 else \ 969 inp->inp_flags &= ~bit; \ 970 INP_WUNLOCK(inp); \ 971} while (0) 972 973 case IP_RECVOPTS: 974 OPTSET(INP_RECVOPTS); 975 break; 976 977 case IP_RECVRETOPTS: 978 OPTSET(INP_RECVRETOPTS); 979 break; 980 981 case IP_RECVDSTADDR: 982 OPTSET(INP_RECVDSTADDR); 983 break; 984 985 case IP_RECVTTL: 986 OPTSET(INP_RECVTTL); 987 break; 988 989 case IP_RECVIF: 990 OPTSET(INP_RECVIF); 991 break; 992 993 case IP_FAITH: 994 OPTSET(INP_FAITH); 995 break; 996 997 case IP_ONESBCAST: 998 OPTSET(INP_ONESBCAST); 999 break; 1000 case IP_DONTFRAG: 1001 OPTSET(INP_DONTFRAG); 1002 break; 1003 case IP_BINDANY: 1004 OPTSET(INP_BINDANY); 1005 break; 1006 } 1007 break; 1008#undef OPTSET 1009 1010 /* 1011 * Multicast socket options are processed by the in_mcast 1012 * module. 1013 */ 1014 case IP_MULTICAST_IF: 1015 case IP_MULTICAST_VIF: 1016 case IP_MULTICAST_TTL: 1017 case IP_MULTICAST_LOOP: 1018 case IP_ADD_MEMBERSHIP: 1019 case IP_DROP_MEMBERSHIP: 1020 case IP_ADD_SOURCE_MEMBERSHIP: 1021 case IP_DROP_SOURCE_MEMBERSHIP: 1022 case IP_BLOCK_SOURCE: 1023 case IP_UNBLOCK_SOURCE: 1024 case IP_MSFILTER: 1025 case MCAST_JOIN_GROUP: 1026 case MCAST_LEAVE_GROUP: 1027 case MCAST_JOIN_SOURCE_GROUP: 1028 case MCAST_LEAVE_SOURCE_GROUP: 1029 case MCAST_BLOCK_SOURCE: 1030 case MCAST_UNBLOCK_SOURCE: 1031 error = inp_setmoptions(inp, sopt); 1032 break; 1033 1034 case IP_PORTRANGE: 1035 error = sooptcopyin(sopt, &optval, sizeof optval, 1036 sizeof optval); 1037 if (error) 1038 break; 1039 1040 INP_WLOCK(inp); 1041 switch (optval) { 1042 case IP_PORTRANGE_DEFAULT: 1043 inp->inp_flags &= ~(INP_LOWPORT); 1044 inp->inp_flags &= ~(INP_HIGHPORT); 1045 break; 1046 1047 case IP_PORTRANGE_HIGH: 1048 inp->inp_flags &= ~(INP_LOWPORT); 1049 inp->inp_flags |= INP_HIGHPORT; 1050 break; 1051 1052 case IP_PORTRANGE_LOW: 1053 inp->inp_flags &= ~(INP_HIGHPORT); 1054 inp->inp_flags |= INP_LOWPORT; 1055 break; 1056 1057 default: 1058 error = EINVAL; 1059 break; 1060 } 1061 INP_WUNLOCK(inp); 1062 break; 1063 1064#ifdef IPSEC 1065 case IP_IPSEC_POLICY: 1066 { 1067 caddr_t req; 1068 struct mbuf *m; 1069 1070 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1071 break; 1072 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1073 break; 1074 req = mtod(m, caddr_t); 1075 error = ipsec_set_policy(inp, sopt->sopt_name, req, 1076 m->m_len, (sopt->sopt_td != NULL) ? 1077 sopt->sopt_td->td_ucred : NULL); 1078 m_freem(m); 1079 break; 1080 } 1081#endif /* IPSEC */ 1082 1083 default: 1084 error = ENOPROTOOPT; 1085 break; 1086 } 1087 break; 1088 1089 case SOPT_GET: 1090 switch (sopt->sopt_name) { 1091 case IP_OPTIONS: 1092 case IP_RETOPTS: 1093 if (inp->inp_options) 1094 error = sooptcopyout(sopt, 1095 mtod(inp->inp_options, 1096 char *), 1097 inp->inp_options->m_len); 1098 else 1099 sopt->sopt_valsize = 0; 1100 break; 1101 1102 case IP_TOS: 1103 case IP_TTL: 1104 case IP_MINTTL: 1105 case IP_RECVOPTS: 1106 case IP_RECVRETOPTS: 1107 case IP_RECVDSTADDR: 1108 case IP_RECVTTL: 1109 case IP_RECVIF: 1110 case IP_PORTRANGE: 1111 case IP_FAITH: 1112 case IP_ONESBCAST: 1113 case IP_DONTFRAG: 1114 switch (sopt->sopt_name) { 1115 1116 case IP_TOS: 1117 optval = inp->inp_ip_tos; 1118 break; 1119 1120 case IP_TTL: 1121 optval = inp->inp_ip_ttl; 1122 break; 1123 1124 case IP_MINTTL: 1125 optval = inp->inp_ip_minttl; 1126 break; 1127 1128#define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1129 1130 case IP_RECVOPTS: 1131 optval = OPTBIT(INP_RECVOPTS); 1132 break; 1133 1134 case IP_RECVRETOPTS: 1135 optval = OPTBIT(INP_RECVRETOPTS); 1136 break; 1137 1138 case IP_RECVDSTADDR: 1139 optval = OPTBIT(INP_RECVDSTADDR); 1140 break; 1141 1142 case IP_RECVTTL: 1143 optval = OPTBIT(INP_RECVTTL); 1144 break; 1145 1146 case IP_RECVIF: 1147 optval = OPTBIT(INP_RECVIF); 1148 break; 1149 1150 case IP_PORTRANGE: 1151 if (inp->inp_flags & INP_HIGHPORT) 1152 optval = IP_PORTRANGE_HIGH; 1153 else if (inp->inp_flags & INP_LOWPORT) 1154 optval = IP_PORTRANGE_LOW; 1155 else 1156 optval = 0; 1157 break; 1158 1159 case IP_FAITH: 1160 optval = OPTBIT(INP_FAITH); 1161 break; 1162 1163 case IP_ONESBCAST: 1164 optval = OPTBIT(INP_ONESBCAST); 1165 break; 1166 case IP_DONTFRAG: 1167 optval = OPTBIT(INP_DONTFRAG); 1168 break; 1169 } 1170 error = sooptcopyout(sopt, &optval, sizeof optval); 1171 break; 1172 1173 /* 1174 * Multicast socket options are processed by the in_mcast 1175 * module. 1176 */ 1177 case IP_MULTICAST_IF: 1178 case IP_MULTICAST_VIF: 1179 case IP_MULTICAST_TTL: 1180 case IP_MULTICAST_LOOP: 1181 case IP_MSFILTER: 1182 error = inp_getmoptions(inp, sopt); 1183 break; 1184 1185#ifdef IPSEC 1186 case IP_IPSEC_POLICY: 1187 { 1188 struct mbuf *m = NULL; 1189 caddr_t req = NULL; 1190 size_t len = 0; 1191 1192 if (m != 0) { 1193 req = mtod(m, caddr_t); 1194 len = m->m_len; 1195 } 1196 error = ipsec_get_policy(sotoinpcb(so), req, len, &m); 1197 if (error == 0) 1198 error = soopt_mcopyout(sopt, m); /* XXX */ 1199 if (error == 0) 1200 m_freem(m); 1201 break; 1202 } 1203#endif /* IPSEC */ 1204 1205 default: 1206 error = ENOPROTOOPT; 1207 break; 1208 } 1209 break; 1210 } 1211 return (error); 1212} 1213 1214/* 1215 * Routine called from ip_output() to loop back a copy of an IP multicast 1216 * packet to the input queue of a specified interface. Note that this 1217 * calls the output routine of the loopback "driver", but with an interface 1218 * pointer that might NOT be a loopback interface -- evil, but easier than 1219 * replicating that code here. 1220 */ 1221static void 1222ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1223 int hlen) 1224{ 1225 register struct ip *ip; 1226 struct mbuf *copym; 1227 1228 /* 1229 * Make a deep copy of the packet because we're going to 1230 * modify the pack in order to generate checksums. 1231 */ 1232 copym = m_dup(m, M_DONTWAIT); 1233 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1234 copym = m_pullup(copym, hlen); 1235 if (copym != NULL) { 1236 /* If needed, compute the checksum and mark it as valid. */ 1237 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1238 in_delayed_cksum(copym); 1239 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1240 copym->m_pkthdr.csum_flags |= 1241 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1242 copym->m_pkthdr.csum_data = 0xffff; 1243 } 1244 /* 1245 * We don't bother to fragment if the IP length is greater 1246 * than the interface's MTU. Can this possibly matter? 1247 */ 1248 ip = mtod(copym, struct ip *); 1249 ip->ip_len = htons(ip->ip_len); 1250 ip->ip_off = htons(ip->ip_off); 1251 ip->ip_sum = 0; 1252 ip->ip_sum = in_cksum(copym, hlen); 1253#if 1 /* XXX */ 1254 if (dst->sin_family != AF_INET) { 1255 printf("ip_mloopback: bad address family %d\n", 1256 dst->sin_family); 1257 dst->sin_family = AF_INET; 1258 } 1259#endif 1260 if_simloop(ifp, copym, dst->sin_family, 0); 1261 } 1262} 1263