sys/netinet/ip_output.c

   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("$FreeBSD$");
  34
  35 #include "opt_inet.h"
  36 #include "opt_ipfw.h"
  37 #include "opt_ipsec.h"
  38 #include "opt_mbuf_stress_test.h"
  39 #include "opt_mpath.h"
  40 #include "opt_route.h"
  41 #include "opt_sctp.h"
  42 #include "opt_rss.h"
  43
  44 #include <sys/param.h>
  45 #include <sys/systm.h>
  46 #include <sys/kernel.h>
  47 #include <sys/lock.h>
  48 #include <sys/malloc.h>
  49 #include <sys/mbuf.h>
  50 #include <sys/priv.h>
  51 #include <sys/proc.h>
  52 #include <sys/protosw.h>
  53 #include <sys/rmlock.h>
  54 #include <sys/sdt.h>
  55 #include <sys/socket.h>
  56 #include <sys/socketvar.h>
  57 #include <sys/sysctl.h>
  58 #include <sys/ucred.h>
  59
  60 #include <net/if.h>
  61 #include <net/if_var.h>
  62 #include <net/if_llatbl.h>
  63 #include <net/netisr.h>
  64 #include <net/pfil.h>
  65 #include <net/route.h>
  66 #include <net/flowtable.h>
  67 #ifdef RADIX_MPATH
  68 #include <net/radix_mpath.h>
  69 #endif
  70 #include <net/rss_config.h>
  71 #include <net/vnet.h>
  72
  73 #include <netinet/in.h>
  74 #include <netinet/in_kdtrace.h>
  75 #include <netinet/in_systm.h>
  76 #include <netinet/ip.h>
  77 #include <netinet/in_pcb.h>
  78 #include <netinet/in_rss.h>
  79 #include <netinet/in_var.h>
  80 #include <netinet/ip_var.h>
  81 #include <netinet/ip_options.h>
  82 #ifdef SCTP
  83 #include <netinet/sctp.h>
  84 #include <netinet/sctp_crc32.h>
  85 #endif
  86
  87 #ifdef IPSEC
  88 #include <netinet/ip_ipsec.h>
  89 #include <netipsec/ipsec.h>
  90 #endif /* IPSEC*/
  91
  92 #include <machine/in_cksum.h>
  93
  94 #include <security/mac/mac_framework.h>
  95
  96 #ifdef MBUF_STRESS_TEST
  97 static int mbuf_frag_size = 0;
  98 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
  99         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 100 #endif
 101
 102 static void     ip_mloopback(struct ifnet *, const struct mbuf *, int);
 103
 104
 105 extern int in_mcast_loop;
 106 extern  struct protosw inetsw[];
 107
 108 static inline int
 109 ip_output_pfil(struct mbuf **mp, struct ifnet *ifp, struct inpcb *inp,
 110     struct sockaddr_in *dst, int *fibnum, int *error)
 111 {
 112         struct m_tag *fwd_tag = NULL;
 113         struct mbuf *m;
 114         struct in_addr odst;
 115         struct ip *ip;
 116
 117         m = *mp;
 118         ip = mtod(m, struct ip *);
 119
 120         /* Run through list of hooks for output packets. */
 121         odst.s_addr = ip->ip_dst.s_addr;
 122         *error = pfil_run_hooks(&V_inet_pfil_hook, mp, ifp, PFIL_OUT, inp);
 123         m = *mp;
 124         if ((*error) != 0 || m == NULL)
 125                 return 1; /* Finished */
 126
 127         ip = mtod(m, struct ip *);
 128
 129         /* See if destination IP address was changed by packet filter. */
 130         if (odst.s_addr != ip->ip_dst.s_addr) {
 131                 m->m_flags |= M_SKIP_FIREWALL;
 132                 /* If destination is now ourself drop to ip_input(). */
 133                 if (in_localip(ip->ip_dst)) {
 134                         m->m_flags |= M_FASTFWD_OURS;
 135                         if (m->m_pkthdr.rcvif == NULL)
 136                                 m->m_pkthdr.rcvif = V_loif;
 137                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 138                                 m->m_pkthdr.csum_flags |=
 139                                         CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 140                                 m->m_pkthdr.csum_data = 0xffff;
 141                         }
 142                         m->m_pkthdr.csum_flags |=
 143                                 CSUM_IP_CHECKED | CSUM_IP_VALID;
 144 #ifdef SCTP
 145                         if (m->m_pkthdr.csum_flags & CSUM_SCTP)
 146                                 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
 147 #endif
 148                         *error = netisr_queue(NETISR_IP, m);
 149                         return 1; /* Finished */
 150                 }
 151
 152                 bzero(dst, sizeof(*dst));
 153                 dst->sin_family = AF_INET;
 154                 dst->sin_len = sizeof(*dst);
 155                 dst->sin_addr = ip->ip_dst;
 156
 157                 return -1; /* Reloop */
 158         }
 159         /* See if fib was changed by packet filter. */
 160         if ((*fibnum) != M_GETFIB(m)) {
 161                 m->m_flags |= M_SKIP_FIREWALL;
 162                 *fibnum = M_GETFIB(m);
 163                 return -1; /* Reloop for FIB change */
 164         }
 165
 166         /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */
 167         if (m->m_flags & M_FASTFWD_OURS) {
 168                 if (m->m_pkthdr.rcvif == NULL)
 169                         m->m_pkthdr.rcvif = V_loif;
 170                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 171                         m->m_pkthdr.csum_flags |=
 172                                 CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 173                         m->m_pkthdr.csum_data = 0xffff;
 174                 }
 175 #ifdef SCTP
 176                 if (m->m_pkthdr.csum_flags & CSUM_SCTP)
 177                         m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
 178 #endif
 179                 m->m_pkthdr.csum_flags |=
 180                         CSUM_IP_CHECKED | CSUM_IP_VALID;
 181
 182                 *error = netisr_queue(NETISR_IP, m);
 183                 return 1; /* Finished */
 184         }
 185         /* Or forward to some other address? */
 186         if ((m->m_flags & M_IP_NEXTHOP) &&
 187             ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
 188                 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
 189                 m->m_flags |= M_SKIP_FIREWALL;
 190                 m->m_flags &= ~M_IP_NEXTHOP;
 191                 m_tag_delete(m, fwd_tag);
 192
 193                 return -1; /* Reloop for CHANGE of dst */
 194         }
 195
 196         return 0;
 197 }
 198
 199 /*
 200  * IP output.  The packet in mbuf chain m contains a skeletal IP
 201  * header (with len, off, ttl, proto, tos, src, dst).
 202  * The mbuf chain containing the packet will be freed.
 203  * The mbuf opt, if present, will not be freed.
 204  * If route ro is present and has ro_rt initialized, route lookup would be
 205  * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL,
 206  * then result of route lookup is stored in ro->ro_rt.
 207  *
 208  * In the IP forwarding case, the packet will arrive with options already
 209  * inserted, so must have a NULL opt pointer.
 210  */
 211 int
 212 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
 213     struct ip_moptions *imo, struct inpcb *inp)
 214 {
 215         struct rm_priotracker in_ifa_tracker;
 216         struct ip *ip;
 217         struct ifnet *ifp = NULL;       /* keep compiler happy */
 218         struct mbuf *m0;
 219         int hlen = sizeof (struct ip);
 220         int mtu;
 221         int error = 0;
 222         struct sockaddr_in *dst;
 223         const struct sockaddr_in *gw;
 224         struct in_ifaddr *ia;
 225         int isbroadcast;
 226         uint16_t ip_len, ip_off;
 227         struct route iproute;
 228         struct rtentry *rte;    /* cache for ro->ro_rt */
 229         uint32_t fibnum;
 230         int have_ia_ref;
 231 #ifdef IPSEC
 232         int no_route_but_check_spd = 0;
 233 #endif
 234         M_ASSERTPKTHDR(m);
 235
 236         if (inp != NULL) {
 237                 INP_LOCK_ASSERT(inp);
 238                 M_SETFIB(m, inp->inp_inc.inc_fibnum);
 239                 if ((flags & IP_NODEFAULTFLOWID) == 0) {
 240                         m->m_pkthdr.flowid = inp->inp_flowid;
 241                         M_HASHTYPE_SET(m, inp->inp_flowtype);
 242                 }
 243         }
 244
 245         if (ro == NULL) {
 246                 ro = &iproute;
 247                 bzero(ro, sizeof (*ro));
 248         }
 249
 250 #ifdef FLOWTABLE
 251         if (ro->ro_rt == NULL)
 252                 (void )flowtable_lookup(AF_INET, m, ro);
 253 #endif
 254
 255         if (opt) {
 256                 int len = 0;
 257                 m = ip_insertoptions(m, opt, &len);
 258                 if (len != 0)
 259                         hlen = len; /* ip->ip_hl is updated above */
 260         }
 261         ip = mtod(m, struct ip *);
 262         ip_len = ntohs(ip->ip_len);
 263         ip_off = ntohs(ip->ip_off);
 264
 265         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
 266                 ip->ip_v = IPVERSION;
 267                 ip->ip_hl = hlen >> 2;
 268                 ip_fillid(ip);
 269                 IPSTAT_INC(ips_localout);
 270         } else {
 271                 /* Header already set, fetch hlen from there */
 272                 hlen = ip->ip_hl << 2;
 273         }
 274
 275         /*
 276          * dst/gw handling:
 277          *
 278          * dst can be rewritten but always points to &ro->ro_dst.
 279          * gw is readonly but can point either to dst OR rt_gateway,
 280          * therefore we need restore gw if we're redoing lookup.
 281          */
 282         gw = dst = (struct sockaddr_in *)&ro->ro_dst;
 283         fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : M_GETFIB(m);
 284         rte = ro->ro_rt;
 285         /*
 286          * The address family should also be checked in case of sharing
 287          * the cache with IPv6.
 288          */
 289         if (rte == NULL || dst->sin_family != AF_INET) {
 290                 bzero(dst, sizeof(*dst));
 291                 dst->sin_family = AF_INET;
 292                 dst->sin_len = sizeof(*dst);
 293                 dst->sin_addr = ip->ip_dst;
 294         }
 295 again:
 296         ia = NULL;
 297         have_ia_ref = 0;
 298         /*
 299          * If routing to interface only, short circuit routing lookup.
 300          * The use of an all-ones broadcast address implies this; an
 301          * interface is specified by the broadcast address of an interface,
 302          * or the destination address of a ptp interface.
 303          */
 304         if (flags & IP_SENDONES) {
 305                 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst),
 306                                                       M_GETFIB(m)))) == NULL &&
 307                     (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
 308                                                     M_GETFIB(m)))) == NULL) {
 309                         IPSTAT_INC(ips_noroute);
 310                         error = ENETUNREACH;
 311                         goto bad;
 312                 }
 313                 have_ia_ref = 1;
 314                 ip->ip_dst.s_addr = INADDR_BROADCAST;
 315                 dst->sin_addr = ip->ip_dst;
 316                 ifp = ia->ia_ifp;
 317                 ip->ip_ttl = 1;
 318                 isbroadcast = 1;
 319         } else if (flags & IP_ROUTETOIF) {
 320                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
 321                                                     M_GETFIB(m)))) == NULL &&
 322                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst), 0,
 323                                                 M_GETFIB(m)))) == NULL) {
 324                         IPSTAT_INC(ips_noroute);
 325                         error = ENETUNREACH;
 326                         goto bad;
 327                 }
 328                 have_ia_ref = 1;
 329                 ifp = ia->ia_ifp;
 330                 ip->ip_ttl = 1;
 331                 isbroadcast = in_broadcast(dst->sin_addr, ifp);
 332         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 333             imo != NULL && imo->imo_multicast_ifp != NULL) {
 334                 /*
 335                  * Bypass the normal routing lookup for multicast
 336                  * packets if the interface is specified.
 337                  */
 338                 ifp = imo->imo_multicast_ifp;
 339                 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
 340                 if (ia)
 341                         have_ia_ref = 1;
 342                 isbroadcast = 0;        /* fool gcc */
 343         } else {
 344                 /*
 345                  * We want to do any cloning requested by the link layer,
 346                  * as this is probably required in all cases for correct
 347                  * operation (as it is for ARP).
 348                  */
 349                 if (rte == NULL) {
 350 #ifdef RADIX_MPATH
 351                         rtalloc_mpath_fib(ro,
 352                             ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr),
 353                             fibnum);
 354 #else
 355                         in_rtalloc_ign(ro, 0, fibnum);
 356 #endif
 357                         rte = ro->ro_rt;
 358                 }
 359                 if (rte == NULL ||
 360                     (rte->rt_flags & RTF_UP) == 0 ||
 361                     rte->rt_ifp == NULL ||
 362                     !RT_LINK_IS_UP(rte->rt_ifp)) {
 363 #ifdef IPSEC
 364                         /*
 365                          * There is no route for this packet, but it is
 366                          * possible that a matching SPD entry exists.
 367                          */
 368                         no_route_but_check_spd = 1;
 369                         mtu = 0; /* Silence GCC warning. */
 370                         goto sendit;
 371 #endif
 372                         IPSTAT_INC(ips_noroute);
 373                         error = EHOSTUNREACH;
 374                         goto bad;
 375                 }
 376                 ia = ifatoia(rte->rt_ifa);
 377                 ifp = rte->rt_ifp;
 378                 counter_u64_add(rte->rt_pksent, 1);
 379                 rt_update_ro_flags(ro);
 380                 if (rte->rt_flags & RTF_GATEWAY)
 381                         gw = (struct sockaddr_in *)rte->rt_gateway;
 382                 if (rte->rt_flags & RTF_HOST)
 383                         isbroadcast = (rte->rt_flags & RTF_BROADCAST);
 384                 else
 385                         isbroadcast = in_broadcast(gw->sin_addr, ifp);
 386         }
 387
 388         /*
 389          * Calculate MTU.  If we have a route that is up, use that,
 390          * otherwise use the interface's MTU.
 391          */
 392         if (rte != NULL && (rte->rt_flags & (RTF_UP|RTF_HOST)))
 393                 mtu = rte->rt_mtu;
 394         else
 395                 mtu = ifp->if_mtu;
 396         /* Catch a possible divide by zero later. */
 397         KASSERT(mtu > 0, ("%s: mtu %d <= 0, rte=%p (rt_flags=0x%08x) ifp=%p",
 398             __func__, mtu, rte, (rte != NULL) ? rte->rt_flags : 0, ifp));
 399
 400         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 401                 m->m_flags |= M_MCAST;
 402                 /*
 403                  * IP destination address is multicast.  Make sure "gw"
 404                  * still points to the address in "ro".  (It may have been
 405                  * changed to point to a gateway address, above.)
 406                  */
 407                 gw = dst;
 408                 /*
 409                  * See if the caller provided any multicast options
 410                  */
 411                 if (imo != NULL) {
 412                         ip->ip_ttl = imo->imo_multicast_ttl;
 413                         if (imo->imo_multicast_vif != -1)
 414                                 ip->ip_src.s_addr =
 415                                     ip_mcast_src ?
 416                                     ip_mcast_src(imo->imo_multicast_vif) :
 417                                     INADDR_ANY;
 418                 } else
 419                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 420                 /*
 421                  * Confirm that the outgoing interface supports multicast.
 422                  */
 423                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 424                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 425                                 IPSTAT_INC(ips_noroute);
 426                                 error = ENETUNREACH;
 427                                 goto bad;
 428                         }
 429                 }
 430                 /*
 431                  * If source address not specified yet, use address
 432                  * of outgoing interface.
 433                  */
 434                 if (ip->ip_src.s_addr == INADDR_ANY) {
 435                         /* Interface may have no addresses. */
 436                         if (ia != NULL)
 437                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 438                 }
 439
 440                 if ((imo == NULL && in_mcast_loop) ||
 441                     (imo && imo->imo_multicast_loop)) {
 442                         /*
 443                          * Loop back multicast datagram if not expressly
 444                          * forbidden to do so, even if we are not a member
 445                          * of the group; ip_input() will filter it later,
 446                          * thus deferring a hash lookup and mutex acquisition
 447                          * at the expense of a cheap copy using m_copym().
 448                          */
 449                         ip_mloopback(ifp, m, hlen);
 450                 } else {
 451                         /*
 452                          * If we are acting as a multicast router, perform
 453                          * multicast forwarding as if the packet had just
 454                          * arrived on the interface to which we are about
 455                          * to send.  The multicast forwarding function
 456                          * recursively calls this function, using the
 457                          * IP_FORWARDING flag to prevent infinite recursion.
 458                          *
 459                          * Multicasts that are looped back by ip_mloopback(),
 460                          * above, will be forwarded by the ip_input() routine,
 461                          * if necessary.
 462                          */
 463                         if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) {
 464                                 /*
 465                                  * If rsvp daemon is not running, do not
 466                                  * set ip_moptions. This ensures that the packet
 467                                  * is multicast and not just sent down one link
 468                                  * as prescribed by rsvpd.
 469                                  */
 470                                 if (!V_rsvp_on)
 471                                         imo = NULL;
 472                                 if (ip_mforward &&
 473                                     ip_mforward(ip, ifp, m, imo) != 0) {
 474                                         m_freem(m);
 475                                         goto done;
 476                                 }
 477                         }
 478                 }
 479
 480                 /*
 481                  * Multicasts with a time-to-live of zero may be looped-
 482                  * back, above, but must not be transmitted on a network.
 483                  * Also, multicasts addressed to the loopback interface
 484                  * are not sent -- the above call to ip_mloopback() will
 485                  * loop back a copy. ip_input() will drop the copy if
 486                  * this host does not belong to the destination group on
 487                  * the loopback interface.
 488                  */
 489                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 490                         m_freem(m);
 491                         goto done;
 492                 }
 493
 494                 goto sendit;
 495         }
 496
 497         /*
 498          * If the source address is not specified yet, use the address
 499          * of the outoing interface.
 500          */
 501         if (ip->ip_src.s_addr == INADDR_ANY) {
 502                 /* Interface may have no addresses. */
 503                 if (ia != NULL) {
 504                         ip->ip_src = IA_SIN(ia)->sin_addr;
 505                 }
 506         }
 507
 508         /*
 509          * Look for broadcast address and
 510          * verify user is allowed to send
 511          * such a packet.
 512          */
 513         if (isbroadcast) {
 514                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 515                         error = EADDRNOTAVAIL;
 516                         goto bad;
 517                 }
 518                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 519                         error = EACCES;
 520                         goto bad;
 521                 }
 522                 /* don't allow broadcast messages to be fragmented */
 523                 if (ip_len > mtu) {
 524                         error = EMSGSIZE;
 525                         goto bad;
 526                 }
 527                 m->m_flags |= M_BCAST;
 528         } else {
 529                 m->m_flags &= ~M_BCAST;
 530         }
 531
 532 sendit:
 533 #ifdef IPSEC
 534         switch(ip_ipsec_output(&m, inp, &error)) {
 535         case 1:
 536                 goto bad;
 537         case -1:
 538                 goto done;
 539         case 0:
 540         default:
 541                 break;  /* Continue with packet processing. */
 542         }
 543         /*
 544          * Check if there was a route for this packet; return error if not.
 545          */
 546         if (no_route_but_check_spd) {
 547                 IPSTAT_INC(ips_noroute);
 548                 error = EHOSTUNREACH;
 549                 goto bad;
 550         }
 551         /* Update variables that are affected by ipsec4_output(). */
 552         ip = mtod(m, struct ip *);
 553         hlen = ip->ip_hl << 2;
 554 #endif /* IPSEC */
 555
 556         /* Jump over all PFIL processing if hooks are not active. */
 557         if (PFIL_HOOKED(&V_inet_pfil_hook)) {
 558                 switch (ip_output_pfil(&m, ifp, inp, dst, &fibnum, &error)) {
 559                 case 1: /* Finished */
 560                         goto done;
 561
 562                 case 0: /* Continue normally */
 563                         ip = mtod(m, struct ip *);
 564                         break;
 565
 566                 case -1: /* Need to try again */
 567                         /* Reset everything for a new round */
 568                         RO_RTFREE(ro);
 569                         if (have_ia_ref)
 570                                 ifa_free(&ia->ia_ifa);
 571                         ro->ro_prepend = NULL;
 572                         rte = NULL;
 573                         gw = dst;
 574                         ip = mtod(m, struct ip *);
 575                         goto again;
 576
 577                 }
 578         }
 579
 580         /* 127/8 must not appear on wire - RFC1122. */
 581         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 582             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 583                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 584                         IPSTAT_INC(ips_badaddr);
 585                         error = EADDRNOTAVAIL;
 586                         goto bad;
 587                 }
 588         }
 589
 590         m->m_pkthdr.csum_flags |= CSUM_IP;
 591         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
 592                 in_delayed_cksum(m);
 593                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 594         }
 595 #ifdef SCTP
 596         if (m->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
 597                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
 598                 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
 599         }
 600 #endif
 601
 602         /*
 603          * If small enough for interface, or the interface will take
 604          * care of the fragmentation for us, we can just send directly.
 605          */
 606         if (ip_len <= mtu ||
 607             (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
 608                 ip->ip_sum = 0;
 609                 if (m->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
 610                         ip->ip_sum = in_cksum(m, hlen);
 611                         m->m_pkthdr.csum_flags &= ~CSUM_IP;
 612                 }
 613
 614                 /*
 615                  * Record statistics for this interface address.
 616                  * With CSUM_TSO the byte/packet count will be slightly
 617                  * incorrect because we count the IP+TCP headers only
 618                  * once instead of for every generated packet.
 619                  */
 620                 if (!(flags & IP_FORWARDING) && ia) {
 621                         if (m->m_pkthdr.csum_flags & CSUM_TSO)
 622                                 counter_u64_add(ia->ia_ifa.ifa_opackets,
 623                                     m->m_pkthdr.len / m->m_pkthdr.tso_segsz);
 624                         else
 625                                 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
 626
 627                         counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len);
 628                 }
 629 #ifdef MBUF_STRESS_TEST
 630                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
 631                         m = m_fragment(m, M_NOWAIT, mbuf_frag_size);
 632 #endif
 633                 /*
 634                  * Reset layer specific mbuf flags
 635                  * to avoid confusing lower layers.
 636                  */
 637                 m_clrprotoflags(m);
 638                 IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
 639                 error = (*ifp->if_output)(ifp, m,
 640                     (const struct sockaddr *)gw, ro);
 641                 goto done;
 642         }
 643
 644         /* Balk when DF bit is set or the interface didn't support TSO. */
 645         if ((ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) {
 646                 error = EMSGSIZE;
 647                 IPSTAT_INC(ips_cantfrag);
 648                 goto bad;
 649         }
 650
 651         /*
 652          * Too large for interface; fragment if possible. If successful,
 653          * on return, m will point to a list of packets to be sent.
 654          */
 655         error = ip_fragment(ip, &m, mtu, ifp->if_hwassist);
 656         if (error)
 657                 goto bad;
 658         for (; m; m = m0) {
 659                 m0 = m->m_nextpkt;
 660                 m->m_nextpkt = 0;
 661                 if (error == 0) {
 662                         /* Record statistics for this interface address. */
 663                         if (ia != NULL) {
 664                                 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
 665                                 counter_u64_add(ia->ia_ifa.ifa_obytes,
 666                                     m->m_pkthdr.len);
 667                         }
 668                         /*
 669                          * Reset layer specific mbuf flags
 670                          * to avoid confusing upper layers.
 671                          */
 672                         m_clrprotoflags(m);
 673
 674                         IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
 675                         error = (*ifp->if_output)(ifp, m,
 676                             (const struct sockaddr *)gw, ro);
 677                 } else
 678                         m_freem(m);
 679         }
 680
 681         if (error == 0)
 682                 IPSTAT_INC(ips_fragmented);
 683
 684 done:
 685         if (ro == &iproute)
 686                 RO_RTFREE(ro);
 687         else if (rte == NULL)
 688                 /*
 689                  * If the caller supplied a route but somehow the reference
 690                  * to it has been released need to prevent the caller
 691                  * calling RTFREE on it again.
 692                  */
 693                 ro->ro_rt = NULL;
 694         if (have_ia_ref)
 695                 ifa_free(&ia->ia_ifa);
 696         return (error);
 697 bad:
 698         m_freem(m);
 699         goto done;
 700 }
 701
 702 /*
 703  * Create a chain of fragments which fit the given mtu. m_frag points to the
 704  * mbuf to be fragmented; on return it points to the chain with the fragments.
 705  * Return 0 if no error. If error, m_frag may contain a partially built
 706  * chain of fragments that should be freed by the caller.
 707  *
 708  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
 709  */
 710 int
 711 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
 712     u_long if_hwassist_flags)
 713 {
 714         int error = 0;
 715         int hlen = ip->ip_hl << 2;
 716         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
 717         int off;
 718         struct mbuf *m0 = *m_frag;      /* the original packet          */
 719         int firstlen;
 720         struct mbuf **mnext;
 721         int nfrags;
 722         uint16_t ip_len, ip_off;
 723
 724         ip_len = ntohs(ip->ip_len);
 725         ip_off = ntohs(ip->ip_off);
 726
 727         if (ip_off & IP_DF) {   /* Fragmentation not allowed */
 728                 IPSTAT_INC(ips_cantfrag);
 729                 return EMSGSIZE;
 730         }
 731
 732         /*
 733          * Must be able to put at least 8 bytes per fragment.
 734          */
 735         if (len < 8)
 736                 return EMSGSIZE;
 737
 738         /*
 739          * If the interface will not calculate checksums on
 740          * fragmented packets, then do it here.
 741          */
 742         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 743                 in_delayed_cksum(m0);
 744                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 745         }
 746 #ifdef SCTP
 747         if (m0->m_pkthdr.csum_flags & CSUM_SCTP) {
 748                 sctp_delayed_cksum(m0, hlen);
 749                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
 750         }
 751 #endif
 752         if (len > PAGE_SIZE) {
 753                 /*
 754                  * Fragment large datagrams such that each segment
 755                  * contains a multiple of PAGE_SIZE amount of data,
 756                  * plus headers. This enables a receiver to perform
 757                  * page-flipping zero-copy optimizations.
 758                  *
 759                  * XXX When does this help given that sender and receiver
 760                  * could have different page sizes, and also mtu could
 761                  * be less than the receiver's page size ?
 762                  */
 763                 int newlen;
 764
 765                 off = MIN(mtu, m0->m_pkthdr.len);
 766
 767                 /*
 768                  * firstlen (off - hlen) must be aligned on an
 769                  * 8-byte boundary
 770                  */
 771                 if (off < hlen)
 772                         goto smart_frag_failure;
 773                 off = ((off - hlen) & ~7) + hlen;
 774                 newlen = (~PAGE_MASK) & mtu;
 775                 if ((newlen + sizeof (struct ip)) > mtu) {
 776                         /* we failed, go back the default */
 777 smart_frag_failure:
 778                         newlen = len;
 779                         off = hlen + len;
 780                 }
 781                 len = newlen;
 782
 783         } else {
 784                 off = hlen + len;
 785         }
 786
 787         firstlen = off - hlen;
 788         mnext = &m0->m_nextpkt;         /* pointer to next packet */
 789
 790         /*
 791          * Loop through length of segment after first fragment,
 792          * make new header and copy data of each part and link onto chain.
 793          * Here, m0 is the original packet, m is the fragment being created.
 794          * The fragments are linked off the m_nextpkt of the original
 795          * packet, which after processing serves as the first fragment.
 796          */
 797         for (nfrags = 1; off < ip_len; off += len, nfrags++) {
 798                 struct ip *mhip;        /* ip header on the fragment */
 799                 struct mbuf *m;
 800                 int mhlen = sizeof (struct ip);
 801
 802                 m = m_gethdr(M_NOWAIT, MT_DATA);
 803                 if (m == NULL) {
 804                         error = ENOBUFS;
 805                         IPSTAT_INC(ips_odropped);
 806                         goto done;
 807                 }
 808                 /*
 809                  * Make sure the complete packet header gets copied
 810                  * from the originating mbuf to the newly created
 811                  * mbuf. This also ensures that existing firewall
 812                  * classification(s), VLAN tags and so on get copied
 813                  * to the resulting fragmented packet(s):
 814                  */
 815                 if (m_dup_pkthdr(m, m0, M_NOWAIT) == 0) {
 816                         m_free(m);
 817                         error = ENOBUFS;
 818                         IPSTAT_INC(ips_odropped);
 819                         goto done;
 820                 }
 821                 /*
 822                  * In the first mbuf, leave room for the link header, then
 823                  * copy the original IP header including options. The payload
 824                  * goes into an additional mbuf chain returned by m_copym().
 825                  */
 826                 m->m_data += max_linkhdr;
 827                 mhip = mtod(m, struct ip *);
 828                 *mhip = *ip;
 829                 if (hlen > sizeof (struct ip)) {
 830                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
 831                         mhip->ip_v = IPVERSION;
 832                         mhip->ip_hl = mhlen >> 2;
 833                 }
 834                 m->m_len = mhlen;
 835                 /* XXX do we need to add ip_off below ? */
 836                 mhip->ip_off = ((off - hlen) >> 3) + ip_off;
 837                 if (off + len >= ip_len)
 838                         len = ip_len - off;
 839                 else
 840                         mhip->ip_off |= IP_MF;
 841                 mhip->ip_len = htons((u_short)(len + mhlen));
 842                 m->m_next = m_copym(m0, off, len, M_NOWAIT);
 843                 if (m->m_next == NULL) {        /* copy failed */
 844                         m_free(m);
 845                         error = ENOBUFS;        /* ??? */
 846                         IPSTAT_INC(ips_odropped);
 847                         goto done;
 848                 }
 849                 m->m_pkthdr.len = mhlen + len;
 850 #ifdef MAC
 851                 mac_netinet_fragment(m0, m);
 852 #endif
 853                 mhip->ip_off = htons(mhip->ip_off);
 854                 mhip->ip_sum = 0;
 855                 if (m->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
 856                         mhip->ip_sum = in_cksum(m, mhlen);
 857                         m->m_pkthdr.csum_flags &= ~CSUM_IP;
 858                 }
 859                 *mnext = m;
 860                 mnext = &m->m_nextpkt;
 861         }
 862         IPSTAT_ADD(ips_ofragments, nfrags);
 863
 864         /*
 865          * Update first fragment by trimming what's been copied out
 866          * and updating header.
 867          */
 868         m_adj(m0, hlen + firstlen - ip_len);
 869         m0->m_pkthdr.len = hlen + firstlen;
 870         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
 871         ip->ip_off = htons(ip_off | IP_MF);
 872         ip->ip_sum = 0;
 873         if (m0->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
 874                 ip->ip_sum = in_cksum(m0, hlen);
 875                 m0->m_pkthdr.csum_flags &= ~CSUM_IP;
 876         }
 877
 878 done:
 879         *m_frag = m0;
 880         return error;
 881 }
 882
 883 void
 884 in_delayed_cksum(struct mbuf *m)
 885 {
 886         struct ip *ip;
 887         uint16_t csum, offset, ip_len;
 888
 889         ip = mtod(m, struct ip *);
 890         offset = ip->ip_hl << 2 ;
 891         ip_len = ntohs(ip->ip_len);
 892         csum = in_cksum_skip(m, ip_len, offset);
 893         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
 894                 csum = 0xffff;
 895         offset += m->m_pkthdr.csum_data;        /* checksum offset */
 896
 897         /* find the mbuf in the chain where the checksum starts*/
 898         while ((m != NULL) && (offset >= m->m_len)) {
 899                 offset -= m->m_len;
 900                 m = m->m_next;
 901         }
 902         KASSERT(m != NULL, ("in_delayed_cksum: checksum outside mbuf chain."));
 903         KASSERT(offset + sizeof(u_short) <= m->m_len, ("in_delayed_cksum: checksum split between mbufs."));
 904         *(u_short *)(m->m_data + offset) = csum;
 905 }
 906
 907 /*
 908  * IP socket option processing.
 909  */
 910 int
 911 ip_ctloutput(struct socket *so, struct sockopt *sopt)
 912 {
 913         struct  inpcb *inp = sotoinpcb(so);
 914         int     error, optval;
 915 #ifdef  RSS
 916         uint32_t rss_bucket;
 917         int retval;
 918 #endif
 919
 920         error = optval = 0;
 921         if (sopt->sopt_level != IPPROTO_IP) {
 922                 error = EINVAL;
 923
 924                 if (sopt->sopt_level == SOL_SOCKET &&
 925                     sopt->sopt_dir == SOPT_SET) {
 926                         switch (sopt->sopt_name) {
 927                         case SO_REUSEADDR:
 928                                 INP_WLOCK(inp);
 929                                 if ((so->so_options & SO_REUSEADDR) != 0)
 930                                         inp->inp_flags2 |= INP_REUSEADDR;
 931                                 else
 932                                         inp->inp_flags2 &= ~INP_REUSEADDR;
 933                                 INP_WUNLOCK(inp);
 934                                 error = 0;
 935                                 break;
 936                         case SO_REUSEPORT:
 937                                 INP_WLOCK(inp);
 938                                 if ((so->so_options & SO_REUSEPORT) != 0)
 939                                         inp->inp_flags2 |= INP_REUSEPORT;
 940                                 else
 941                                         inp->inp_flags2 &= ~INP_REUSEPORT;
 942                                 INP_WUNLOCK(inp);
 943                                 error = 0;
 944                                 break;
 945                         case SO_SETFIB:
 946                                 INP_WLOCK(inp);
 947                                 inp->inp_inc.inc_fibnum = so->so_fibnum;
 948                                 INP_WUNLOCK(inp);
 949                                 error = 0;
 950                                 break;
 951                         default:
 952                                 break;
 953                         }
 954                 }
 955                 return (error);
 956         }
 957
 958         switch (sopt->sopt_dir) {
 959         case SOPT_SET:
 960                 switch (sopt->sopt_name) {
 961                 case IP_OPTIONS:
 962 #ifdef notyet
 963                 case IP_RETOPTS:
 964 #endif
 965                 {
 966                         struct mbuf *m;
 967                         if (sopt->sopt_valsize > MLEN) {
 968                                 error = EMSGSIZE;
 969                                 break;
 970                         }
 971                         m = m_get(sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
 972                         if (m == NULL) {
 973                                 error = ENOBUFS;
 974                                 break;
 975                         }
 976                         m->m_len = sopt->sopt_valsize;
 977                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
 978                                             m->m_len);
 979                         if (error) {
 980                                 m_free(m);
 981                                 break;
 982                         }
 983                         INP_WLOCK(inp);
 984                         error = ip_pcbopts(inp, sopt->sopt_name, m);
 985                         INP_WUNLOCK(inp);
 986                         return (error);
 987                 }
 988
 989                 case IP_BINDANY:
 990                         if (sopt->sopt_td != NULL) {
 991                                 error = priv_check(sopt->sopt_td,
 992                                     PRIV_NETINET_BINDANY);
 993                                 if (error)
 994                                         break;
 995                         }
 996                         /* FALLTHROUGH */
 997                 case IP_BINDMULTI:
 998 #ifdef  RSS
 999                 case IP_RSS_LISTEN_BUCKET:
1000 #endif
1001                 case IP_TOS:
1002                 case IP_TTL:
1003                 case IP_MINTTL:
1004                 case IP_RECVOPTS:
1005                 case IP_RECVRETOPTS:
1006                 case IP_RECVDSTADDR:
1007                 case IP_RECVTTL:
1008                 case IP_RECVIF:
1009                 case IP_ONESBCAST:
1010                 case IP_DONTFRAG:
1011                 case IP_RECVTOS:
1012                 case IP_RECVFLOWID:
1013 #ifdef  RSS
1014                 case IP_RECVRSSBUCKETID:
1015 #endif
1016                         error = sooptcopyin(sopt, &optval, sizeof optval,
1017                                             sizeof optval);
1018                         if (error)
1019                                 break;
1020
1021                         switch (sopt->sopt_name) {
1022                         case IP_TOS:
1023                                 inp->inp_ip_tos = optval;
1024                                 break;
1025
1026                         case IP_TTL:
1027                                 inp->inp_ip_ttl = optval;
1028                                 break;
1029
1030                         case IP_MINTTL:
1031                                 if (optval >= 0 && optval <= MAXTTL)
1032                                         inp->inp_ip_minttl = optval;
1033                                 else
1034                                         error = EINVAL;
1035                                 break;
1036
1037 #define OPTSET(bit) do {                                                \
1038         INP_WLOCK(inp);                                                 \
1039         if (optval)                                                     \
1040                 inp->inp_flags |= bit;                                  \
1041         else                                                            \
1042                 inp->inp_flags &= ~bit;                                 \
1043         INP_WUNLOCK(inp);                                               \
1044 } while (0)
1045
1046 #define OPTSET2(bit, val) do {                                          \
1047         INP_WLOCK(inp);                                                 \
1048         if (val)                                                        \
1049                 inp->inp_flags2 |= bit;                                 \
1050         else                                                            \
1051                 inp->inp_flags2 &= ~bit;                                \
1052         INP_WUNLOCK(inp);                                               \
1053 } while (0)
1054
1055                         case IP_RECVOPTS:
1056                                 OPTSET(INP_RECVOPTS);
1057                                 break;
1058
1059                         case IP_RECVRETOPTS:
1060                                 OPTSET(INP_RECVRETOPTS);
1061                                 break;
1062
1063                         case IP_RECVDSTADDR:
1064                                 OPTSET(INP_RECVDSTADDR);
1065                                 break;
1066
1067                         case IP_RECVTTL:
1068                                 OPTSET(INP_RECVTTL);
1069                                 break;
1070
1071                         case IP_RECVIF:
1072                                 OPTSET(INP_RECVIF);
1073                                 break;
1074
1075                         case IP_ONESBCAST:
1076                                 OPTSET(INP_ONESBCAST);
1077                                 break;
1078                         case IP_DONTFRAG:
1079                                 OPTSET(INP_DONTFRAG);
1080                                 break;
1081                         case IP_BINDANY:
1082                                 OPTSET(INP_BINDANY);
1083                                 break;
1084                         case IP_RECVTOS:
1085                                 OPTSET(INP_RECVTOS);
1086                                 break;
1087                         case IP_BINDMULTI:
1088                                 OPTSET2(INP_BINDMULTI, optval);
1089                                 break;
1090                         case IP_RECVFLOWID:
1091                                 OPTSET2(INP_RECVFLOWID, optval);
1092                                 break;
1093 #ifdef  RSS
1094                         case IP_RSS_LISTEN_BUCKET:
1095                                 if ((optval >= 0) &&
1096                                     (optval < rss_getnumbuckets())) {
1097                                         inp->inp_rss_listen_bucket = optval;
1098                                         OPTSET2(INP_RSS_BUCKET_SET, 1);
1099                                 } else {
1100                                         error = EINVAL;
1101                                 }
1102                                 break;
1103                         case IP_RECVRSSBUCKETID:
1104                                 OPTSET2(INP_RECVRSSBUCKETID, optval);
1105                                 break;
1106 #endif
1107                         }
1108                         break;
1109 #undef OPTSET
1110 #undef OPTSET2
1111
1112                 /*
1113                  * Multicast socket options are processed by the in_mcast
1114                  * module.
1115                  */
1116                 case IP_MULTICAST_IF:
1117                 case IP_MULTICAST_VIF:
1118                 case IP_MULTICAST_TTL:
1119                 case IP_MULTICAST_LOOP:
1120                 case IP_ADD_MEMBERSHIP:
1121                 case IP_DROP_MEMBERSHIP:
1122                 case IP_ADD_SOURCE_MEMBERSHIP:
1123                 case IP_DROP_SOURCE_MEMBERSHIP:
1124                 case IP_BLOCK_SOURCE:
1125                 case IP_UNBLOCK_SOURCE:
1126                 case IP_MSFILTER:
1127                 case MCAST_JOIN_GROUP:
1128                 case MCAST_LEAVE_GROUP:
1129                 case MCAST_JOIN_SOURCE_GROUP:
1130                 case MCAST_LEAVE_SOURCE_GROUP:
1131                 case MCAST_BLOCK_SOURCE:
1132                 case MCAST_UNBLOCK_SOURCE:
1133                         error = inp_setmoptions(inp, sopt);
1134                         break;
1135
1136                 case IP_PORTRANGE:
1137                         error = sooptcopyin(sopt, &optval, sizeof optval,
1138                                             sizeof optval);
1139                         if (error)
1140                                 break;
1141
1142                         INP_WLOCK(inp);
1143                         switch (optval) {
1144                         case IP_PORTRANGE_DEFAULT:
1145                                 inp->inp_flags &= ~(INP_LOWPORT);
1146                                 inp->inp_flags &= ~(INP_HIGHPORT);
1147                                 break;
1148
1149                         case IP_PORTRANGE_HIGH:
1150                                 inp->inp_flags &= ~(INP_LOWPORT);
1151                                 inp->inp_flags |= INP_HIGHPORT;
1152                                 break;
1153
1154                         case IP_PORTRANGE_LOW:
1155                                 inp->inp_flags &= ~(INP_HIGHPORT);
1156                                 inp->inp_flags |= INP_LOWPORT;
1157                                 break;
1158
1159                         default:
1160                                 error = EINVAL;
1161                                 break;
1162                         }
1163                         INP_WUNLOCK(inp);
1164                         break;
1165
1166 #ifdef IPSEC
1167                 case IP_IPSEC_POLICY:
1168                 {
1169                         caddr_t req;
1170                         struct mbuf *m;
1171
1172                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1173                                 break;
1174                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1175                                 break;
1176                         req = mtod(m, caddr_t);
1177                         error = ipsec_set_policy(inp, sopt->sopt_name, req,
1178                             m->m_len, (sopt->sopt_td != NULL) ?
1179                             sopt->sopt_td->td_ucred : NULL);
1180                         m_freem(m);
1181                         break;
1182                 }
1183 #endif /* IPSEC */
1184
1185                 default:
1186                         error = ENOPROTOOPT;
1187                         break;
1188                 }
1189                 break;
1190
1191         case SOPT_GET:
1192                 switch (sopt->sopt_name) {
1193                 case IP_OPTIONS:
1194                 case IP_RETOPTS:
1195                         if (inp->inp_options)
1196                                 error = sooptcopyout(sopt,
1197                                                      mtod(inp->inp_options,
1198                                                           char *),
1199                                                      inp->inp_options->m_len);
1200                         else
1201                                 sopt->sopt_valsize = 0;
1202                         break;
1203
1204                 case IP_TOS:
1205                 case IP_TTL:
1206                 case IP_MINTTL:
1207                 case IP_RECVOPTS:
1208                 case IP_RECVRETOPTS:
1209                 case IP_RECVDSTADDR:
1210                 case IP_RECVTTL:
1211                 case IP_RECVIF:
1212                 case IP_PORTRANGE:
1213                 case IP_ONESBCAST:
1214                 case IP_DONTFRAG:
1215                 case IP_BINDANY:
1216                 case IP_RECVTOS:
1217                 case IP_BINDMULTI:
1218                 case IP_FLOWID:
1219                 case IP_FLOWTYPE:
1220                 case IP_RECVFLOWID:
1221 #ifdef  RSS
1222                 case IP_RSSBUCKETID:
1223                 case IP_RECVRSSBUCKETID:
1224 #endif
1225                         switch (sopt->sopt_name) {
1226
1227                         case IP_TOS:
1228                                 optval = inp->inp_ip_tos;
1229                                 break;
1230
1231                         case IP_TTL:
1232                                 optval = inp->inp_ip_ttl;
1233                                 break;
1234
1235                         case IP_MINTTL:
1236                                 optval = inp->inp_ip_minttl;
1237                                 break;
1238
1239 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1240 #define OPTBIT2(bit)    (inp->inp_flags2 & bit ? 1 : 0)
1241
1242                         case IP_RECVOPTS:
1243                                 optval = OPTBIT(INP_RECVOPTS);
1244                                 break;
1245
1246                         case IP_RECVRETOPTS:
1247                                 optval = OPTBIT(INP_RECVRETOPTS);
1248                                 break;
1249
1250                         case IP_RECVDSTADDR:
1251                                 optval = OPTBIT(INP_RECVDSTADDR);
1252                                 break;
1253
1254                         case IP_RECVTTL:
1255                                 optval = OPTBIT(INP_RECVTTL);
1256                                 break;
1257
1258                         case IP_RECVIF:
1259                                 optval = OPTBIT(INP_RECVIF);
1260                                 break;
1261
1262                         case IP_PORTRANGE:
1263                                 if (inp->inp_flags & INP_HIGHPORT)
1264                                         optval = IP_PORTRANGE_HIGH;
1265                                 else if (inp->inp_flags & INP_LOWPORT)
1266                                         optval = IP_PORTRANGE_LOW;
1267                                 else
1268                                         optval = 0;
1269                                 break;
1270
1271                         case IP_ONESBCAST:
1272                                 optval = OPTBIT(INP_ONESBCAST);
1273                                 break;
1274                         case IP_DONTFRAG:
1275                                 optval = OPTBIT(INP_DONTFRAG);
1276                                 break;
1277                         case IP_BINDANY:
1278                                 optval = OPTBIT(INP_BINDANY);
1279                                 break;
1280                         case IP_RECVTOS:
1281                                 optval = OPTBIT(INP_RECVTOS);
1282                                 break;
1283                         case IP_FLOWID:
1284                                 optval = inp->inp_flowid;
1285                                 break;
1286                         case IP_FLOWTYPE:
1287                                 optval = inp->inp_flowtype;
1288                                 break;
1289                         case IP_RECVFLOWID:
1290                                 optval = OPTBIT2(INP_RECVFLOWID);
1291                                 break;
1292 #ifdef  RSS
1293                         case IP_RSSBUCKETID:
1294                                 retval = rss_hash2bucket(inp->inp_flowid,
1295                                     inp->inp_flowtype,
1296                                     &rss_bucket);
1297                                 if (retval == 0)
1298                                         optval = rss_bucket;
1299                                 else
1300                                         error = EINVAL;
1301                                 break;
1302                         case IP_RECVRSSBUCKETID:
1303                                 optval = OPTBIT2(INP_RECVRSSBUCKETID);
1304                                 break;
1305 #endif
1306                         case IP_BINDMULTI:
1307                                 optval = OPTBIT2(INP_BINDMULTI);
1308                                 break;
1309                         }
1310                         error = sooptcopyout(sopt, &optval, sizeof optval);
1311                         break;
1312
1313                 /*
1314                  * Multicast socket options are processed by the in_mcast
1315                  * module.
1316                  */
1317                 case IP_MULTICAST_IF:
1318                 case IP_MULTICAST_VIF:
1319                 case IP_MULTICAST_TTL:
1320                 case IP_MULTICAST_LOOP:
1321                 case IP_MSFILTER:
1322                         error = inp_getmoptions(inp, sopt);
1323                         break;
1324
1325 #ifdef IPSEC
1326                 case IP_IPSEC_POLICY:
1327                 {
1328                         struct mbuf *m = NULL;
1329                         caddr_t req = NULL;
1330                         size_t len = 0;
1331
1332                         if (m != 0) {
1333                                 req = mtod(m, caddr_t);
1334                                 len = m->m_len;
1335                         }
1336                         error = ipsec_get_policy(sotoinpcb(so), req, len, &m);
1337                         if (error == 0)
1338                                 error = soopt_mcopyout(sopt, m); /* XXX */
1339                         if (error == 0)
1340                                 m_freem(m);
1341                         break;
1342                 }
1343 #endif /* IPSEC */
1344
1345                 default:
1346                         error = ENOPROTOOPT;
1347                         break;
1348                 }
1349                 break;
1350         }
1351         return (error);
1352 }
1353
1354 /*
1355  * Routine called from ip_output() to loop back a copy of an IP multicast
1356  * packet to the input queue of a specified interface.  Note that this
1357  * calls the output routine of the loopback "driver", but with an interface
1358  * pointer that might NOT be a loopback interface -- evil, but easier than
1359  * replicating that code here.
1360  */
1361 static void
1362 ip_mloopback(struct ifnet *ifp, const struct mbuf *m, int hlen)
1363 {
1364         struct ip *ip;
1365         struct mbuf *copym;
1366
1367         /*
1368          * Make a deep copy of the packet because we're going to
1369          * modify the pack in order to generate checksums.
1370          */
1371         copym = m_dup(m, M_NOWAIT);
1372         if (copym != NULL && (!M_WRITABLE(copym) || copym->m_len < hlen))
1373                 copym = m_pullup(copym, hlen);
1374         if (copym != NULL) {
1375                 /* If needed, compute the checksum and mark it as valid. */
1376                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1377                         in_delayed_cksum(copym);
1378                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1379                         copym->m_pkthdr.csum_flags |=
1380                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1381                         copym->m_pkthdr.csum_data = 0xffff;
1382                 }
1383                 /*
1384                  * We don't bother to fragment if the IP length is greater
1385                  * than the interface's MTU.  Can this possibly matter?
1386                  */
1387                 ip = mtod(copym, struct ip *);
1388                 ip->ip_sum = 0;
1389                 ip->ip_sum = in_cksum(copym, hlen);
1390                 if_simloop(ifp, copym, AF_INET, 0);
1391         }
1392 }