ip_output.c revision 185101
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 185101 2008-11-19 19:19:30Z julian $"); 34 35#include "opt_ipfw.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_mbuf_stress_test.h" 39#include "opt_mpath.h" 40 41#include <sys/param.h> 42#include <sys/systm.h> 43#include <sys/kernel.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/priv.h> 47#include <sys/proc.h> 48#include <sys/protosw.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/sysctl.h> 52#include <sys/ucred.h> 53#include <sys/vimage.h> 54 55#include <net/if.h> 56#include <net/netisr.h> 57#include <net/pfil.h> 58#include <net/route.h> 59#ifdef RADIX_MPATH 60#include <net/radix_mpath.h> 61#endif 62 63#include <netinet/in.h> 64#include <netinet/in_systm.h> 65#include <netinet/ip.h> 66#include <netinet/in_pcb.h> 67#include <netinet/in_var.h> 68#include <netinet/ip_var.h> 69#include <netinet/ip_options.h> 70 71#ifdef IPSEC 72#include <netinet/ip_ipsec.h> 73#include <netipsec/ipsec.h> 74#endif /* IPSEC*/ 75 76#include <machine/in_cksum.h> 77 78#include <security/mac/mac_framework.h> 79 80#define print_ip(x, a, y) printf("%s %d.%d.%d.%d%s",\ 81 x, (ntohl(a.s_addr)>>24)&0xFF,\ 82 (ntohl(a.s_addr)>>16)&0xFF,\ 83 (ntohl(a.s_addr)>>8)&0xFF,\ 84 (ntohl(a.s_addr))&0xFF, y); 85 86#ifdef VIMAGE_GLOBALS 87u_short ip_id; 88#endif 89 90#ifdef MBUF_STRESS_TEST 91int mbuf_frag_size = 0; 92SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW, 93 &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size"); 94#endif 95 96static void ip_mloopback 97 (struct ifnet *, struct mbuf *, struct sockaddr_in *, int); 98 99 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 * The mbuf chain containing the packet will be freed. 106 * The mbuf opt, if present, will not be freed. 107 * In the IP forwarding case, the packet will arrive with options already 108 * inserted, so must have a NULL opt pointer. 109 */ 110int 111ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags, 112 struct ip_moptions *imo, struct inpcb *inp) 113{ 114 INIT_VNET_NET(curvnet); 115 INIT_VNET_INET(curvnet); 116 struct ip *ip; 117 struct ifnet *ifp = NULL; /* keep compiler happy */ 118 struct mbuf *m0; 119 int hlen = sizeof (struct ip); 120 int mtu; 121 int len, error = 0; 122 struct sockaddr_in *dst = NULL; /* keep compiler happy */ 123 struct in_ifaddr *ia = NULL; 124 int isbroadcast, sw_csum; 125 struct route iproute; 126 struct in_addr odst; 127#ifdef IPFIREWALL_FORWARD 128 struct m_tag *fwd_tag = NULL; 129#endif 130 M_ASSERTPKTHDR(m); 131 132 if (ro == NULL) { 133 ro = &iproute; 134 bzero(ro, sizeof (*ro)); 135 } 136 137 if (inp != NULL) { 138 M_SETFIB(m, inp->inp_inc.inc_fibnum); 139 INP_LOCK_ASSERT(inp); 140 } 141 142 if (opt) { 143 len = 0; 144 m = ip_insertoptions(m, opt, &len); 145 if (len != 0) 146 hlen = len; 147 } 148 ip = mtod(m, struct ip *); 149 150 /* 151 * Fill in IP header. If we are not allowing fragmentation, 152 * then the ip_id field is meaningless, but we don't set it 153 * to zero. Doing so causes various problems when devices along 154 * the path (routers, load balancers, firewalls, etc.) illegally 155 * disable DF on our packet. Note that a 16-bit counter 156 * will wrap around in less than 10 seconds at 100 Mbit/s on a 157 * medium with MTU 1500. See Steven M. Bellovin, "A Technique 158 * for Counting NATted Hosts", Proc. IMW'02, available at 159 * <http://www.cs.columbia.edu/~smb/papers/fnat.pdf>. 160 */ 161 if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) { 162 ip->ip_v = IPVERSION; 163 ip->ip_hl = hlen >> 2; 164 ip->ip_id = ip_newid(); 165 V_ipstat.ips_localout++; 166 } else { 167 hlen = ip->ip_hl << 2; 168 } 169 170 dst = (struct sockaddr_in *)&ro->ro_dst; 171again: 172 /* 173 * If there is a cached route, 174 * check that it is to the same destination 175 * and is still up. If not, free it and try again. 176 * The address family should also be checked in case of sharing the 177 * cache with IPv6. 178 */ 179 if (ro->ro_rt && ((ro->ro_rt->rt_flags & RTF_UP) == 0 || 180 dst->sin_family != AF_INET || 181 dst->sin_addr.s_addr != ip->ip_dst.s_addr)) { 182 RTFREE(ro->ro_rt); 183 ro->ro_rt = (struct rtentry *)NULL; 184 } 185#ifdef IPFIREWALL_FORWARD 186 if (ro->ro_rt == NULL && fwd_tag == NULL) { 187#else 188 if (ro->ro_rt == NULL) { 189#endif 190 bzero(dst, sizeof(*dst)); 191 dst->sin_family = AF_INET; 192 dst->sin_len = sizeof(*dst); 193 dst->sin_addr = ip->ip_dst; 194 } 195 /* 196 * If routing to interface only, short circuit routing lookup. 197 * The use of an all-ones broadcast address implies this; an 198 * interface is specified by the broadcast address of an interface, 199 * or the destination address of a ptp interface. 200 */ 201 if (flags & IP_SENDONES) { 202 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst)))) == NULL && 203 (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL) { 204 V_ipstat.ips_noroute++; 205 error = ENETUNREACH; 206 goto bad; 207 } 208 ip->ip_dst.s_addr = INADDR_BROADCAST; 209 dst->sin_addr = ip->ip_dst; 210 ifp = ia->ia_ifp; 211 ip->ip_ttl = 1; 212 isbroadcast = 1; 213 } else if (flags & IP_ROUTETOIF) { 214 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst)))) == NULL && 215 (ia = ifatoia(ifa_ifwithnet(sintosa(dst)))) == NULL) { 216 V_ipstat.ips_noroute++; 217 error = ENETUNREACH; 218 goto bad; 219 } 220 ifp = ia->ia_ifp; 221 ip->ip_ttl = 1; 222 isbroadcast = in_broadcast(dst->sin_addr, ifp); 223 } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) && 224 imo != NULL && imo->imo_multicast_ifp != NULL) { 225 /* 226 * Bypass the normal routing lookup for multicast 227 * packets if the interface is specified. 228 */ 229 ifp = imo->imo_multicast_ifp; 230 IFP_TO_IA(ifp, ia); 231 isbroadcast = 0; /* fool gcc */ 232 } else { 233 /* 234 * We want to do any cloning requested by the link layer, 235 * as this is probably required in all cases for correct 236 * operation (as it is for ARP). 237 */ 238 if (ro->ro_rt == NULL) 239#ifdef RADIX_MPATH 240 rtalloc_mpath_fib(ro, 241 ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr), 242 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 243#else 244 in_rtalloc_ign(ro, 0, 245 inp ? inp->inp_inc.inc_fibnum : M_GETFIB(m)); 246#endif 247 if (ro->ro_rt == NULL) { 248 V_ipstat.ips_noroute++; 249 error = EHOSTUNREACH; 250 goto bad; 251 } 252 ia = ifatoia(ro->ro_rt->rt_ifa); 253 ifp = ro->ro_rt->rt_ifp; 254 ro->ro_rt->rt_rmx.rmx_pksent++; 255 if (ro->ro_rt->rt_flags & RTF_GATEWAY) 256 dst = (struct sockaddr_in *)ro->ro_rt->rt_gateway; 257 if (ro->ro_rt->rt_flags & RTF_HOST) 258 isbroadcast = (ro->ro_rt->rt_flags & RTF_BROADCAST); 259 else 260 isbroadcast = in_broadcast(dst->sin_addr, ifp); 261 } 262 /* 263 * Calculate MTU. If we have a route that is up, use that, 264 * otherwise use the interface's MTU. 265 */ 266 if (ro->ro_rt != NULL && (ro->ro_rt->rt_flags & (RTF_UP|RTF_HOST))) { 267 /* 268 * This case can happen if the user changed the MTU 269 * of an interface after enabling IP on it. Because 270 * most netifs don't keep track of routes pointing to 271 * them, there is no way for one to update all its 272 * routes when the MTU is changed. 273 */ 274 if (ro->ro_rt->rt_rmx.rmx_mtu > ifp->if_mtu) 275 ro->ro_rt->rt_rmx.rmx_mtu = ifp->if_mtu; 276 mtu = ro->ro_rt->rt_rmx.rmx_mtu; 277 } else { 278 mtu = ifp->if_mtu; 279 } 280 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 281 struct in_multi *inm; 282 283 m->m_flags |= M_MCAST; 284 /* 285 * IP destination address is multicast. Make sure "dst" 286 * still points to the address in "ro". (It may have been 287 * changed to point to a gateway address, above.) 288 */ 289 dst = (struct sockaddr_in *)&ro->ro_dst; 290 /* 291 * See if the caller provided any multicast options 292 */ 293 if (imo != NULL) { 294 ip->ip_ttl = imo->imo_multicast_ttl; 295 if (imo->imo_multicast_vif != -1) 296 ip->ip_src.s_addr = 297 ip_mcast_src ? 298 ip_mcast_src(imo->imo_multicast_vif) : 299 INADDR_ANY; 300 } else 301 ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL; 302 /* 303 * Confirm that the outgoing interface supports multicast. 304 */ 305 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) { 306 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 307 V_ipstat.ips_noroute++; 308 error = ENETUNREACH; 309 goto bad; 310 } 311 } 312 /* 313 * If source address not specified yet, use address 314 * of outgoing interface. 315 */ 316 if (ip->ip_src.s_addr == INADDR_ANY) { 317 /* Interface may have no addresses. */ 318 if (ia != NULL) 319 ip->ip_src = IA_SIN(ia)->sin_addr; 320 } 321 322 IN_MULTI_LOCK(); 323 IN_LOOKUP_MULTI(ip->ip_dst, ifp, inm); 324 if (inm != NULL && 325 (imo == NULL || imo->imo_multicast_loop)) { 326 IN_MULTI_UNLOCK(); 327 /* 328 * If we belong to the destination multicast group 329 * on the outgoing interface, and the caller did not 330 * forbid loopback, loop back a copy. 331 */ 332 ip_mloopback(ifp, m, dst, hlen); 333 } 334 else { 335 IN_MULTI_UNLOCK(); 336 /* 337 * If we are acting as a multicast router, perform 338 * multicast forwarding as if the packet had just 339 * arrived on the interface to which we are about 340 * to send. The multicast forwarding function 341 * recursively calls this function, using the 342 * IP_FORWARDING flag to prevent infinite recursion. 343 * 344 * Multicasts that are looped back by ip_mloopback(), 345 * above, will be forwarded by the ip_input() routine, 346 * if necessary. 347 */ 348 if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) { 349 /* 350 * If rsvp daemon is not running, do not 351 * set ip_moptions. This ensures that the packet 352 * is multicast and not just sent down one link 353 * as prescribed by rsvpd. 354 */ 355 if (!V_rsvp_on) 356 imo = NULL; 357 if (ip_mforward && 358 ip_mforward(ip, ifp, m, imo) != 0) { 359 m_freem(m); 360 goto done; 361 } 362 } 363 } 364 365 /* 366 * Multicasts with a time-to-live of zero may be looped- 367 * back, above, but must not be transmitted on a network. 368 * Also, multicasts addressed to the loopback interface 369 * are not sent -- the above call to ip_mloopback() will 370 * loop back a copy if this host actually belongs to the 371 * destination group on the loopback interface. 372 */ 373 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) { 374 m_freem(m); 375 goto done; 376 } 377 378 goto sendit; 379 } 380 381 /* 382 * If the source address is not specified yet, use the address 383 * of the outoing interface. 384 */ 385 if (ip->ip_src.s_addr == INADDR_ANY) { 386 /* Interface may have no addresses. */ 387 if (ia != NULL) { 388 ip->ip_src = IA_SIN(ia)->sin_addr; 389 } 390 } 391 392 /* 393 * Verify that we have any chance at all of being able to queue the 394 * packet or packet fragments, unless ALTQ is enabled on the given 395 * interface in which case packetdrop should be done by queueing. 396 */ 397#ifdef ALTQ 398 if ((!ALTQ_IS_ENABLED(&ifp->if_snd)) && 399 ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 400 ifp->if_snd.ifq_maxlen)) 401#else 402 if ((ifp->if_snd.ifq_len + ip->ip_len / mtu + 1) >= 403 ifp->if_snd.ifq_maxlen) 404#endif /* ALTQ */ 405 { 406 error = ENOBUFS; 407 V_ipstat.ips_odropped++; 408 ifp->if_snd.ifq_drops += (ip->ip_len / ifp->if_mtu + 1); 409 goto bad; 410 } 411 412 /* 413 * Look for broadcast address and 414 * verify user is allowed to send 415 * such a packet. 416 */ 417 if (isbroadcast) { 418 if ((ifp->if_flags & IFF_BROADCAST) == 0) { 419 error = EADDRNOTAVAIL; 420 goto bad; 421 } 422 if ((flags & IP_ALLOWBROADCAST) == 0) { 423 error = EACCES; 424 goto bad; 425 } 426 /* don't allow broadcast messages to be fragmented */ 427 if (ip->ip_len > mtu) { 428 error = EMSGSIZE; 429 goto bad; 430 } 431 m->m_flags |= M_BCAST; 432 } else { 433 m->m_flags &= ~M_BCAST; 434 } 435 436sendit: 437#ifdef IPSEC 438 switch(ip_ipsec_output(&m, inp, &flags, &error, &ro, &iproute, &dst, &ia, &ifp)) { 439 case 1: 440 goto bad; 441 case -1: 442 goto done; 443 case 0: 444 default: 445 break; /* Continue with packet processing. */ 446 } 447 /* Update variables that are affected by ipsec4_output(). */ 448 ip = mtod(m, struct ip *); 449 hlen = ip->ip_hl << 2; 450#endif /* IPSEC */ 451 452 /* Jump over all PFIL processing if hooks are not active. */ 453 if (!PFIL_HOOKED(&inet_pfil_hook)) 454 goto passout; 455 456 /* Run through list of hooks for output packets. */ 457 odst.s_addr = ip->ip_dst.s_addr; 458 error = pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, inp); 459 if (error != 0 || m == NULL) 460 goto done; 461 462 ip = mtod(m, struct ip *); 463 464 /* See if destination IP address was changed by packet filter. */ 465 if (odst.s_addr != ip->ip_dst.s_addr) { 466 m->m_flags |= M_SKIP_FIREWALL; 467 /* If destination is now ourself drop to ip_input(). */ 468 if (in_localip(ip->ip_dst)) { 469 m->m_flags |= M_FASTFWD_OURS; 470 if (m->m_pkthdr.rcvif == NULL) 471 m->m_pkthdr.rcvif = V_loif; 472 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 473 m->m_pkthdr.csum_flags |= 474 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 475 m->m_pkthdr.csum_data = 0xffff; 476 } 477 m->m_pkthdr.csum_flags |= 478 CSUM_IP_CHECKED | CSUM_IP_VALID; 479 480 error = netisr_queue(NETISR_IP, m); 481 goto done; 482 } else 483 goto again; /* Redo the routing table lookup. */ 484 } 485 486#ifdef IPFIREWALL_FORWARD 487 /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */ 488 if (m->m_flags & M_FASTFWD_OURS) { 489 if (m->m_pkthdr.rcvif == NULL) 490 m->m_pkthdr.rcvif = V_loif; 491 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 492 m->m_pkthdr.csum_flags |= 493 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 494 m->m_pkthdr.csum_data = 0xffff; 495 } 496 m->m_pkthdr.csum_flags |= 497 CSUM_IP_CHECKED | CSUM_IP_VALID; 498 499 error = netisr_queue(NETISR_IP, m); 500 goto done; 501 } 502 /* Or forward to some other address? */ 503 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 504 if (fwd_tag) { 505 dst = (struct sockaddr_in *)&ro->ro_dst; 506 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in)); 507 m->m_flags |= M_SKIP_FIREWALL; 508 m_tag_delete(m, fwd_tag); 509 goto again; 510 } 511#endif /* IPFIREWALL_FORWARD */ 512 513passout: 514 /* 127/8 must not appear on wire - RFC1122. */ 515 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 516 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 517 if ((ifp->if_flags & IFF_LOOPBACK) == 0) { 518 V_ipstat.ips_badaddr++; 519 error = EADDRNOTAVAIL; 520 goto bad; 521 } 522 } 523 524 m->m_pkthdr.csum_flags |= CSUM_IP; 525 sw_csum = m->m_pkthdr.csum_flags & ~ifp->if_hwassist; 526 if (sw_csum & CSUM_DELAY_DATA) { 527 in_delayed_cksum(m); 528 sw_csum &= ~CSUM_DELAY_DATA; 529 } 530 m->m_pkthdr.csum_flags &= ifp->if_hwassist; 531 532 /* 533 * If small enough for interface, or the interface will take 534 * care of the fragmentation for us, we can just send directly. 535 */ 536 if (ip->ip_len <= mtu || 537 (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0 || 538 ((ip->ip_off & IP_DF) == 0 && (ifp->if_hwassist & CSUM_FRAGMENT))) { 539 ip->ip_len = htons(ip->ip_len); 540 ip->ip_off = htons(ip->ip_off); 541 ip->ip_sum = 0; 542 if (sw_csum & CSUM_DELAY_IP) 543 ip->ip_sum = in_cksum(m, hlen); 544 545 /* 546 * Record statistics for this interface address. 547 * With CSUM_TSO the byte/packet count will be slightly 548 * incorrect because we count the IP+TCP headers only 549 * once instead of for every generated packet. 550 */ 551 if (!(flags & IP_FORWARDING) && ia) { 552 if (m->m_pkthdr.csum_flags & CSUM_TSO) 553 ia->ia_ifa.if_opackets += 554 m->m_pkthdr.len / m->m_pkthdr.tso_segsz; 555 else 556 ia->ia_ifa.if_opackets++; 557 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 558 } 559#ifdef MBUF_STRESS_TEST 560 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size) 561 m = m_fragment(m, M_DONTWAIT, mbuf_frag_size); 562#endif 563 /* 564 * Reset layer specific mbuf flags 565 * to avoid confusing lower layers. 566 */ 567 m->m_flags &= ~(M_PROTOFLAGS); 568 569 error = (*ifp->if_output)(ifp, m, 570 (struct sockaddr *)dst, ro->ro_rt); 571 goto done; 572 } 573 574 /* Balk when DF bit is set or the interface didn't support TSO. */ 575 if ((ip->ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) { 576 error = EMSGSIZE; 577 V_ipstat.ips_cantfrag++; 578 goto bad; 579 } 580 581 /* 582 * Too large for interface; fragment if possible. If successful, 583 * on return, m will point to a list of packets to be sent. 584 */ 585 error = ip_fragment(ip, &m, mtu, ifp->if_hwassist, sw_csum); 586 if (error) 587 goto bad; 588 for (; m; m = m0) { 589 m0 = m->m_nextpkt; 590 m->m_nextpkt = 0; 591 if (error == 0) { 592 /* Record statistics for this interface address. */ 593 if (ia != NULL) { 594 ia->ia_ifa.if_opackets++; 595 ia->ia_ifa.if_obytes += m->m_pkthdr.len; 596 } 597 /* 598 * Reset layer specific mbuf flags 599 * to avoid confusing upper layers. 600 */ 601 m->m_flags &= ~(M_PROTOFLAGS); 602 603 error = (*ifp->if_output)(ifp, m, 604 (struct sockaddr *)dst, ro->ro_rt); 605 } else 606 m_freem(m); 607 } 608 609 if (error == 0) 610 V_ipstat.ips_fragmented++; 611 612done: 613 if (ro == &iproute && ro->ro_rt) { 614 RTFREE(ro->ro_rt); 615 } 616 return (error); 617bad: 618 m_freem(m); 619 goto done; 620} 621 622/* 623 * Create a chain of fragments which fit the given mtu. m_frag points to the 624 * mbuf to be fragmented; on return it points to the chain with the fragments. 625 * Return 0 if no error. If error, m_frag may contain a partially built 626 * chain of fragments that should be freed by the caller. 627 * 628 * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist) 629 * sw_csum contains the delayed checksums flags (e.g., CSUM_DELAY_IP). 630 */ 631int 632ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu, 633 u_long if_hwassist_flags, int sw_csum) 634{ 635 INIT_VNET_INET(curvnet); 636 int error = 0; 637 int hlen = ip->ip_hl << 2; 638 int len = (mtu - hlen) & ~7; /* size of payload in each fragment */ 639 int off; 640 struct mbuf *m0 = *m_frag; /* the original packet */ 641 int firstlen; 642 struct mbuf **mnext; 643 int nfrags; 644 645 if (ip->ip_off & IP_DF) { /* Fragmentation not allowed */ 646 V_ipstat.ips_cantfrag++; 647 return EMSGSIZE; 648 } 649 650 /* 651 * Must be able to put at least 8 bytes per fragment. 652 */ 653 if (len < 8) 654 return EMSGSIZE; 655 656 /* 657 * If the interface will not calculate checksums on 658 * fragmented packets, then do it here. 659 */ 660 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA && 661 (if_hwassist_flags & CSUM_IP_FRAGS) == 0) { 662 in_delayed_cksum(m0); 663 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 664 } 665 666 if (len > PAGE_SIZE) { 667 /* 668 * Fragment large datagrams such that each segment 669 * contains a multiple of PAGE_SIZE amount of data, 670 * plus headers. This enables a receiver to perform 671 * page-flipping zero-copy optimizations. 672 * 673 * XXX When does this help given that sender and receiver 674 * could have different page sizes, and also mtu could 675 * be less than the receiver's page size ? 676 */ 677 int newlen; 678 struct mbuf *m; 679 680 for (m = m0, off = 0; m && (off+m->m_len) <= mtu; m = m->m_next) 681 off += m->m_len; 682 683 /* 684 * firstlen (off - hlen) must be aligned on an 685 * 8-byte boundary 686 */ 687 if (off < hlen) 688 goto smart_frag_failure; 689 off = ((off - hlen) & ~7) + hlen; 690 newlen = (~PAGE_MASK) & mtu; 691 if ((newlen + sizeof (struct ip)) > mtu) { 692 /* we failed, go back the default */ 693smart_frag_failure: 694 newlen = len; 695 off = hlen + len; 696 } 697 len = newlen; 698 699 } else { 700 off = hlen + len; 701 } 702 703 firstlen = off - hlen; 704 mnext = &m0->m_nextpkt; /* pointer to next packet */ 705 706 /* 707 * Loop through length of segment after first fragment, 708 * make new header and copy data of each part and link onto chain. 709 * Here, m0 is the original packet, m is the fragment being created. 710 * The fragments are linked off the m_nextpkt of the original 711 * packet, which after processing serves as the first fragment. 712 */ 713 for (nfrags = 1; off < ip->ip_len; off += len, nfrags++) { 714 struct ip *mhip; /* ip header on the fragment */ 715 struct mbuf *m; 716 int mhlen = sizeof (struct ip); 717 718 MGETHDR(m, M_DONTWAIT, MT_DATA); 719 if (m == NULL) { 720 error = ENOBUFS; 721 V_ipstat.ips_odropped++; 722 goto done; 723 } 724 m->m_flags |= (m0->m_flags & M_MCAST) | M_FRAG; 725 /* 726 * In the first mbuf, leave room for the link header, then 727 * copy the original IP header including options. The payload 728 * goes into an additional mbuf chain returned by m_copy(). 729 */ 730 m->m_data += max_linkhdr; 731 mhip = mtod(m, struct ip *); 732 *mhip = *ip; 733 if (hlen > sizeof (struct ip)) { 734 mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip); 735 mhip->ip_v = IPVERSION; 736 mhip->ip_hl = mhlen >> 2; 737 } 738 m->m_len = mhlen; 739 /* XXX do we need to add ip->ip_off below ? */ 740 mhip->ip_off = ((off - hlen) >> 3) + ip->ip_off; 741 if (off + len >= ip->ip_len) { /* last fragment */ 742 len = ip->ip_len - off; 743 m->m_flags |= M_LASTFRAG; 744 } else 745 mhip->ip_off |= IP_MF; 746 mhip->ip_len = htons((u_short)(len + mhlen)); 747 m->m_next = m_copy(m0, off, len); 748 if (m->m_next == NULL) { /* copy failed */ 749 m_free(m); 750 error = ENOBUFS; /* ??? */ 751 V_ipstat.ips_odropped++; 752 goto done; 753 } 754 m->m_pkthdr.len = mhlen + len; 755 m->m_pkthdr.rcvif = NULL; 756#ifdef MAC 757 mac_netinet_fragment(m0, m); 758#endif 759 m->m_pkthdr.csum_flags = m0->m_pkthdr.csum_flags; 760 mhip->ip_off = htons(mhip->ip_off); 761 mhip->ip_sum = 0; 762 if (sw_csum & CSUM_DELAY_IP) 763 mhip->ip_sum = in_cksum(m, mhlen); 764 *mnext = m; 765 mnext = &m->m_nextpkt; 766 } 767 V_ipstat.ips_ofragments += nfrags; 768 769 /* set first marker for fragment chain */ 770 m0->m_flags |= M_FIRSTFRAG | M_FRAG; 771 m0->m_pkthdr.csum_data = nfrags; 772 773 /* 774 * Update first fragment by trimming what's been copied out 775 * and updating header. 776 */ 777 m_adj(m0, hlen + firstlen - ip->ip_len); 778 m0->m_pkthdr.len = hlen + firstlen; 779 ip->ip_len = htons((u_short)m0->m_pkthdr.len); 780 ip->ip_off |= IP_MF; 781 ip->ip_off = htons(ip->ip_off); 782 ip->ip_sum = 0; 783 if (sw_csum & CSUM_DELAY_IP) 784 ip->ip_sum = in_cksum(m0, hlen); 785 786done: 787 *m_frag = m0; 788 return error; 789} 790 791void 792in_delayed_cksum(struct mbuf *m) 793{ 794 struct ip *ip; 795 u_short csum, offset; 796 797 ip = mtod(m, struct ip *); 798 offset = ip->ip_hl << 2 ; 799 csum = in_cksum_skip(m, ip->ip_len, offset); 800 if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0) 801 csum = 0xffff; 802 offset += m->m_pkthdr.csum_data; /* checksum offset */ 803 804 if (offset + sizeof(u_short) > m->m_len) { 805 printf("delayed m_pullup, m->len: %d off: %d p: %d\n", 806 m->m_len, offset, ip->ip_p); 807 /* 808 * XXX 809 * this shouldn't happen, but if it does, the 810 * correct behavior may be to insert the checksum 811 * in the appropriate next mbuf in the chain. 812 */ 813 return; 814 } 815 *(u_short *)(m->m_data + offset) = csum; 816} 817 818/* 819 * IP socket option processing. 820 */ 821int 822ip_ctloutput(struct socket *so, struct sockopt *sopt) 823{ 824 struct inpcb *inp = sotoinpcb(so); 825 int error, optval; 826 827 error = optval = 0; 828 if (sopt->sopt_level != IPPROTO_IP) { 829 if ((sopt->sopt_level == SOL_SOCKET) && 830 (sopt->sopt_name == SO_SETFIB)) { 831 inp->inp_inc.inc_fibnum = so->so_fibnum; 832 return (0); 833 } 834 return (EINVAL); 835 } 836 837 switch (sopt->sopt_dir) { 838 case SOPT_SET: 839 switch (sopt->sopt_name) { 840 case IP_OPTIONS: 841#ifdef notyet 842 case IP_RETOPTS: 843#endif 844 { 845 struct mbuf *m; 846 if (sopt->sopt_valsize > MLEN) { 847 error = EMSGSIZE; 848 break; 849 } 850 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA); 851 if (m == NULL) { 852 error = ENOBUFS; 853 break; 854 } 855 m->m_len = sopt->sopt_valsize; 856 error = sooptcopyin(sopt, mtod(m, char *), m->m_len, 857 m->m_len); 858 if (error) { 859 m_free(m); 860 break; 861 } 862 INP_WLOCK(inp); 863 error = ip_pcbopts(inp, sopt->sopt_name, m); 864 INP_WUNLOCK(inp); 865 return (error); 866 } 867 868 case IP_TOS: 869 case IP_TTL: 870 case IP_MINTTL: 871 case IP_RECVOPTS: 872 case IP_RECVRETOPTS: 873 case IP_RECVDSTADDR: 874 case IP_RECVTTL: 875 case IP_RECVIF: 876 case IP_FAITH: 877 case IP_ONESBCAST: 878 case IP_DONTFRAG: 879 error = sooptcopyin(sopt, &optval, sizeof optval, 880 sizeof optval); 881 if (error) 882 break; 883 884 switch (sopt->sopt_name) { 885 case IP_TOS: 886 inp->inp_ip_tos = optval; 887 break; 888 889 case IP_TTL: 890 inp->inp_ip_ttl = optval; 891 break; 892 893 case IP_MINTTL: 894 if (optval > 0 && optval <= MAXTTL) 895 inp->inp_ip_minttl = optval; 896 else 897 error = EINVAL; 898 break; 899 900#define OPTSET(bit) do { \ 901 INP_WLOCK(inp); \ 902 if (optval) \ 903 inp->inp_flags |= bit; \ 904 else \ 905 inp->inp_flags &= ~bit; \ 906 INP_WUNLOCK(inp); \ 907} while (0) 908 909 case IP_RECVOPTS: 910 OPTSET(INP_RECVOPTS); 911 break; 912 913 case IP_RECVRETOPTS: 914 OPTSET(INP_RECVRETOPTS); 915 break; 916 917 case IP_RECVDSTADDR: 918 OPTSET(INP_RECVDSTADDR); 919 break; 920 921 case IP_RECVTTL: 922 OPTSET(INP_RECVTTL); 923 break; 924 925 case IP_RECVIF: 926 OPTSET(INP_RECVIF); 927 break; 928 929 case IP_FAITH: 930 OPTSET(INP_FAITH); 931 break; 932 933 case IP_ONESBCAST: 934 OPTSET(INP_ONESBCAST); 935 break; 936 case IP_DONTFRAG: 937 OPTSET(INP_DONTFRAG); 938 break; 939 } 940 break; 941#undef OPTSET 942 943 /* 944 * Multicast socket options are processed by the in_mcast 945 * module. 946 */ 947 case IP_MULTICAST_IF: 948 case IP_MULTICAST_VIF: 949 case IP_MULTICAST_TTL: 950 case IP_MULTICAST_LOOP: 951 case IP_ADD_MEMBERSHIP: 952 case IP_DROP_MEMBERSHIP: 953 case IP_ADD_SOURCE_MEMBERSHIP: 954 case IP_DROP_SOURCE_MEMBERSHIP: 955 case IP_BLOCK_SOURCE: 956 case IP_UNBLOCK_SOURCE: 957 case IP_MSFILTER: 958 case MCAST_JOIN_GROUP: 959 case MCAST_LEAVE_GROUP: 960 case MCAST_JOIN_SOURCE_GROUP: 961 case MCAST_LEAVE_SOURCE_GROUP: 962 case MCAST_BLOCK_SOURCE: 963 case MCAST_UNBLOCK_SOURCE: 964 error = inp_setmoptions(inp, sopt); 965 break; 966 967 case IP_PORTRANGE: 968 error = sooptcopyin(sopt, &optval, sizeof optval, 969 sizeof optval); 970 if (error) 971 break; 972 973 INP_WLOCK(inp); 974 switch (optval) { 975 case IP_PORTRANGE_DEFAULT: 976 inp->inp_flags &= ~(INP_LOWPORT); 977 inp->inp_flags &= ~(INP_HIGHPORT); 978 break; 979 980 case IP_PORTRANGE_HIGH: 981 inp->inp_flags &= ~(INP_LOWPORT); 982 inp->inp_flags |= INP_HIGHPORT; 983 break; 984 985 case IP_PORTRANGE_LOW: 986 inp->inp_flags &= ~(INP_HIGHPORT); 987 inp->inp_flags |= INP_LOWPORT; 988 break; 989 990 default: 991 error = EINVAL; 992 break; 993 } 994 INP_WUNLOCK(inp); 995 break; 996 997#ifdef IPSEC 998 case IP_IPSEC_POLICY: 999 { 1000 caddr_t req; 1001 struct mbuf *m; 1002 1003 if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */ 1004 break; 1005 if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */ 1006 break; 1007 req = mtod(m, caddr_t); 1008 error = ipsec4_set_policy(inp, sopt->sopt_name, req, 1009 m->m_len, (sopt->sopt_td != NULL) ? 1010 sopt->sopt_td->td_ucred : NULL); 1011 m_freem(m); 1012 break; 1013 } 1014#endif /* IPSEC */ 1015 1016 default: 1017 error = ENOPROTOOPT; 1018 break; 1019 } 1020 break; 1021 1022 case SOPT_GET: 1023 switch (sopt->sopt_name) { 1024 case IP_OPTIONS: 1025 case IP_RETOPTS: 1026 if (inp->inp_options) 1027 error = sooptcopyout(sopt, 1028 mtod(inp->inp_options, 1029 char *), 1030 inp->inp_options->m_len); 1031 else 1032 sopt->sopt_valsize = 0; 1033 break; 1034 1035 case IP_TOS: 1036 case IP_TTL: 1037 case IP_MINTTL: 1038 case IP_RECVOPTS: 1039 case IP_RECVRETOPTS: 1040 case IP_RECVDSTADDR: 1041 case IP_RECVTTL: 1042 case IP_RECVIF: 1043 case IP_PORTRANGE: 1044 case IP_FAITH: 1045 case IP_ONESBCAST: 1046 case IP_DONTFRAG: 1047 switch (sopt->sopt_name) { 1048 1049 case IP_TOS: 1050 optval = inp->inp_ip_tos; 1051 break; 1052 1053 case IP_TTL: 1054 optval = inp->inp_ip_ttl; 1055 break; 1056 1057 case IP_MINTTL: 1058 optval = inp->inp_ip_minttl; 1059 break; 1060 1061#define OPTBIT(bit) (inp->inp_flags & bit ? 1 : 0) 1062 1063 case IP_RECVOPTS: 1064 optval = OPTBIT(INP_RECVOPTS); 1065 break; 1066 1067 case IP_RECVRETOPTS: 1068 optval = OPTBIT(INP_RECVRETOPTS); 1069 break; 1070 1071 case IP_RECVDSTADDR: 1072 optval = OPTBIT(INP_RECVDSTADDR); 1073 break; 1074 1075 case IP_RECVTTL: 1076 optval = OPTBIT(INP_RECVTTL); 1077 break; 1078 1079 case IP_RECVIF: 1080 optval = OPTBIT(INP_RECVIF); 1081 break; 1082 1083 case IP_PORTRANGE: 1084 if (inp->inp_flags & INP_HIGHPORT) 1085 optval = IP_PORTRANGE_HIGH; 1086 else if (inp->inp_flags & INP_LOWPORT) 1087 optval = IP_PORTRANGE_LOW; 1088 else 1089 optval = 0; 1090 break; 1091 1092 case IP_FAITH: 1093 optval = OPTBIT(INP_FAITH); 1094 break; 1095 1096 case IP_ONESBCAST: 1097 optval = OPTBIT(INP_ONESBCAST); 1098 break; 1099 case IP_DONTFRAG: 1100 optval = OPTBIT(INP_DONTFRAG); 1101 break; 1102 } 1103 error = sooptcopyout(sopt, &optval, sizeof optval); 1104 break; 1105 1106 /* 1107 * Multicast socket options are processed by the in_mcast 1108 * module. 1109 */ 1110 case IP_MULTICAST_IF: 1111 case IP_MULTICAST_VIF: 1112 case IP_MULTICAST_TTL: 1113 case IP_MULTICAST_LOOP: 1114 case IP_MSFILTER: 1115 error = inp_getmoptions(inp, sopt); 1116 break; 1117 1118#ifdef IPSEC 1119 case IP_IPSEC_POLICY: 1120 { 1121 struct mbuf *m = NULL; 1122 caddr_t req = NULL; 1123 size_t len = 0; 1124 1125 if (m != 0) { 1126 req = mtod(m, caddr_t); 1127 len = m->m_len; 1128 } 1129 error = ipsec4_get_policy(sotoinpcb(so), req, len, &m); 1130 if (error == 0) 1131 error = soopt_mcopyout(sopt, m); /* XXX */ 1132 if (error == 0) 1133 m_freem(m); 1134 break; 1135 } 1136#endif /* IPSEC */ 1137 1138 default: 1139 error = ENOPROTOOPT; 1140 break; 1141 } 1142 break; 1143 } 1144 return (error); 1145} 1146 1147/* 1148 * Routine called from ip_output() to loop back a copy of an IP multicast 1149 * packet to the input queue of a specified interface. Note that this 1150 * calls the output routine of the loopback "driver", but with an interface 1151 * pointer that might NOT be a loopback interface -- evil, but easier than 1152 * replicating that code here. 1153 */ 1154static void 1155ip_mloopback(struct ifnet *ifp, struct mbuf *m, struct sockaddr_in *dst, 1156 int hlen) 1157{ 1158 register struct ip *ip; 1159 struct mbuf *copym; 1160 1161 /* 1162 * Make a deep copy of the packet because we're going to 1163 * modify the pack in order to generate checksums. 1164 */ 1165 copym = m_dup(m, M_DONTWAIT); 1166 if (copym != NULL && (copym->m_flags & M_EXT || copym->m_len < hlen)) 1167 copym = m_pullup(copym, hlen); 1168 if (copym != NULL) { 1169 /* If needed, compute the checksum and mark it as valid. */ 1170 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 1171 in_delayed_cksum(copym); 1172 copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 1173 copym->m_pkthdr.csum_flags |= 1174 CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1175 copym->m_pkthdr.csum_data = 0xffff; 1176 } 1177 /* 1178 * We don't bother to fragment if the IP length is greater 1179 * than the interface's MTU. Can this possibly matter? 1180 */ 1181 ip = mtod(copym, struct ip *); 1182 ip->ip_len = htons(ip->ip_len); 1183 ip->ip_off = htons(ip->ip_off); 1184 ip->ip_sum = 0; 1185 ip->ip_sum = in_cksum(copym, hlen); 1186#if 1 /* XXX */ 1187 if (dst->sin_family != AF_INET) { 1188 printf("ip_mloopback: bad address family %d\n", 1189 dst->sin_family); 1190 dst->sin_family = AF_INET; 1191 } 1192#endif 1193 if_simloop(ifp, copym, dst->sin_family, 0); 1194 } 1195} 1196