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_ipsec.h"
  37 #include "opt_mbuf_stress_test.h"
  38 #include "opt_mpath.h"
  39 #include "opt_route.h"
  40 #include "opt_sctp.h"
  41 #include "opt_rss.h"
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/kernel.h>
  46 #include <sys/lock.h>
  47 #include <sys/malloc.h>
  48 #include <sys/mbuf.h>
  49 #include <sys/priv.h>
  50 #include <sys/proc.h>
  51 #include <sys/protosw.h>
  52 #include <sys/rmlock.h>
  53 #include <sys/sdt.h>
  54 #include <sys/socket.h>
  55 #include <sys/socketvar.h>
  56 #include <sys/sysctl.h>
  57 #include <sys/ucred.h>
  58
  59 #include <net/if.h>
  60 #include <net/if_var.h>
  61 #include <net/if_llatbl.h>
  62 #include <net/netisr.h>
  63 #include <net/pfil.h>
  64 #include <net/route.h>
  65 #include <net/flowtable.h>
  66 #ifdef RADIX_MPATH
  67 #include <net/radix_mpath.h>
  68 #endif
  69 #include <net/rss_config.h>
  70 #include <net/vnet.h>
  71
  72 #include <netinet/in.h>
  73 #include <netinet/in_kdtrace.h>
  74 #include <netinet/in_systm.h>
  75 #include <netinet/ip.h>
  76 #include <netinet/in_pcb.h>
  77 #include <netinet/in_rss.h>
  78 #include <netinet/in_var.h>
  79 #include <netinet/ip_var.h>
  80 #include <netinet/ip_options.h>
  81 #ifdef SCTP
  82 #include <netinet/sctp.h>
  83 #include <netinet/sctp_crc32.h>
  84 #endif
  85
  86 #ifdef IPSEC
  87 #include <netinet/ip_ipsec.h>
  88 #include <netipsec/ipsec.h>
  89 #endif /* IPSEC*/
  90
  91 #include <machine/in_cksum.h>
  92
  93 #include <security/mac/mac_framework.h>
  94
  95 #ifdef MBUF_STRESS_TEST
  96 static int mbuf_frag_size = 0;
  97 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
  98         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
  99 #endif
 100
 101 static void     ip_mloopback(struct ifnet *, const struct mbuf *, int);
 102
 103
 104 extern int in_mcast_loop;
 105 extern  struct protosw inetsw[];
 106
 107 static inline int
 108 ip_output_pfil(struct mbuf **mp, struct ifnet *ifp, struct inpcb *inp,
 109     struct sockaddr_in *dst, int *fibnum, int *error)
 110 {
 111         struct m_tag *fwd_tag = NULL;
 112         struct mbuf *m;
 113         struct in_addr odst;
 114         struct ip *ip;
 115
 116         m = *mp;
 117         ip = mtod(m, struct ip *);
 118
 119         /* Run through list of hooks for output packets. */
 120         odst.s_addr = ip->ip_dst.s_addr;
 121         *error = pfil_run_hooks(&V_inet_pfil_hook, mp, ifp, PFIL_OUT, inp);
 122         m = *mp;
 123         if ((*error) != 0 || m == NULL)
 124                 return 1; /* Finished */
 125
 126         ip = mtod(m, struct ip *);
 127
 128         /* See if destination IP address was changed by packet filter. */
 129         if (odst.s_addr != ip->ip_dst.s_addr) {
 130                 m->m_flags |= M_SKIP_FIREWALL;
 131                 /* If destination is now ourself drop to ip_input(). */
 132                 if (in_localip(ip->ip_dst)) {
 133                         m->m_flags |= M_FASTFWD_OURS;
 134                         if (m->m_pkthdr.rcvif == NULL)
 135                                 m->m_pkthdr.rcvif = V_loif;
 136                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 137                                 m->m_pkthdr.csum_flags |=
 138                                         CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 139                                 m->m_pkthdr.csum_data = 0xffff;
 140                         }
 141                         m->m_pkthdr.csum_flags |=
 142                                 CSUM_IP_CHECKED | CSUM_IP_VALID;
 143 #ifdef SCTP
 144                         if (m->m_pkthdr.csum_flags & CSUM_SCTP)
 145                                 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
 146 #endif
 147                         *error = netisr_queue(NETISR_IP, m);
 148                         return 1; /* Finished */
 149                 }
 150
 151                 bzero(dst, sizeof(*dst));
 152                 dst->sin_family = AF_INET;
 153                 dst->sin_len = sizeof(*dst);
 154                 dst->sin_addr = ip->ip_dst;
 155
 156                 return -1; /* Reloop */
 157         }
 158         /* See if fib was changed by packet filter. */
 159         if ((*fibnum) != M_GETFIB(m)) {
 160                 m->m_flags |= M_SKIP_FIREWALL;
 161                 *fibnum = M_GETFIB(m);
 162                 return -1; /* Reloop for FIB change */
 163         }
 164
 165         /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */
 166         if (m->m_flags & M_FASTFWD_OURS) {
 167                 if (m->m_pkthdr.rcvif == NULL)
 168                         m->m_pkthdr.rcvif = V_loif;
 169                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 170                         m->m_pkthdr.csum_flags |=
 171                                 CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 172                         m->m_pkthdr.csum_data = 0xffff;
 173                 }
 174 #ifdef SCTP
 175                 if (m->m_pkthdr.csum_flags & CSUM_SCTP)
 176                         m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
 177 #endif
 178                 m->m_pkthdr.csum_flags |=
 179                         CSUM_IP_CHECKED | CSUM_IP_VALID;
 180
 181                 *error = netisr_queue(NETISR_IP, m);
 182                 return 1; /* Finished */
 183         }
 184         /* Or forward to some other address? */
 185         if ((m->m_flags & M_IP_NEXTHOP) &&
 186             ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
 187                 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
 188                 m->m_flags |= M_SKIP_FIREWALL;
 189                 m->m_flags &= ~M_IP_NEXTHOP;
 190                 m_tag_delete(m, fwd_tag);
 191
 192                 return -1; /* Reloop for CHANGE of dst */
 193         }
 194
 195         return 0;
 196 }
 197
 198 /*
 199  * IP output.  The packet in mbuf chain m contains a skeletal IP
 200  * header (with len, off, ttl, proto, tos, src, dst).
 201  * The mbuf chain containing the packet will be freed.
 202  * The mbuf opt, if present, will not be freed.
 203  * If route ro is present and has ro_rt initialized, route lookup would be
 204  * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL,
 205  * then result of route lookup is stored in ro->ro_rt.
 206  *
 207  * In the IP forwarding case, the packet will arrive with options already
 208  * inserted, so must have a NULL opt pointer.
 209  */
 210 int
 211 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
 212     struct ip_moptions *imo, struct inpcb *inp)
 213 {
 214         struct rm_priotracker in_ifa_tracker;
 215         struct ip *ip;
 216         struct ifnet *ifp = NULL;       /* keep compiler happy */
 217         struct mbuf *m0;
 218         int hlen = sizeof (struct ip);
 219         int mtu;
 220         int error = 0;
 221         struct sockaddr_in *dst;
 222         const struct sockaddr_in *gw;
 223         struct in_ifaddr *ia;
 224         int isbroadcast;
 225         uint16_t ip_len, ip_off;
 226         struct route iproute;
 227         struct rtentry *rte;    /* cache for ro->ro_rt */
 228         uint32_t fibnum;
 229         int have_ia_ref;
 230 #ifdef IPSEC
 231         int no_route_but_check_spd = 0;
 232 #endif
 233         M_ASSERTPKTHDR(m);
 234
 235         if (inp != NULL) {
 236                 INP_LOCK_ASSERT(inp);
 237                 M_SETFIB(m, inp->inp_inc.inc_fibnum);
 238                 if ((flags & IP_NODEFAULTFLOWID) == 0) {
 239                         m->m_pkthdr.flowid = inp->inp_flowid;
 240                         M_HASHTYPE_SET(m, inp->inp_flowtype);
 241                 }
 242         }
 243
 244         if (ro == NULL) {
 245                 ro = &iproute;
 246                 bzero(ro, sizeof (*ro));
 247         } else
 248                 ro->ro_flags |= RT_LLE_CACHE;
 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         if (rte == NULL) {
 286                 bzero(dst, sizeof(*dst));
 287                 dst->sin_family = AF_INET;
 288                 dst->sin_len = sizeof(*dst);
 289                 dst->sin_addr = ip->ip_dst;
 290         }
 291 again:
 292         /*
 293          * Validate route against routing table additions;
 294          * a better/more specific route might have been added.
 295          */
 296         if (inp)
 297                 RT_VALIDATE(ro, &inp->inp_rt_cookie, fibnum);
 298         /*
 299          * If there is a cached route,
 300          * check that it is to the same destination
 301          * and is still up.  If not, free it and try again.
 302          * The address family should also be checked in case of sharing the
 303          * cache with IPv6.
 304          * Also check whether routing cache needs invalidation.
 305          */
 306         rte = ro->ro_rt;
 307         if (rte && ((rte->rt_flags & RTF_UP) == 0 ||
 308                     rte->rt_ifp == NULL ||
 309                     !RT_LINK_IS_UP(rte->rt_ifp) ||
 310                           dst->sin_family != AF_INET ||
 311                           dst->sin_addr.s_addr != ip->ip_dst.s_addr)) {
 312                 RTFREE(rte);
 313                 rte = ro->ro_rt = (struct rtentry *)NULL;
 314                 if (ro->ro_lle)
 315                         LLE_FREE(ro->ro_lle);   /* zeros ro_lle */
 316                 ro->ro_lle = (struct llentry *)NULL;
 317         }
 318         ia = NULL;
 319         have_ia_ref = 0;
 320         /*
 321          * If routing to interface only, short circuit routing lookup.
 322          * The use of an all-ones broadcast address implies this; an
 323          * interface is specified by the broadcast address of an interface,
 324          * or the destination address of a ptp interface.
 325          */
 326         if (flags & IP_SENDONES) {
 327                 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst),
 328                                                       M_GETFIB(m)))) == NULL &&
 329                     (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
 330                                                     M_GETFIB(m)))) == NULL) {
 331                         IPSTAT_INC(ips_noroute);
 332                         error = ENETUNREACH;
 333                         goto bad;
 334                 }
 335                 have_ia_ref = 1;
 336                 ip->ip_dst.s_addr = INADDR_BROADCAST;
 337                 dst->sin_addr = ip->ip_dst;
 338                 ifp = ia->ia_ifp;
 339                 ip->ip_ttl = 1;
 340                 isbroadcast = 1;
 341         } else if (flags & IP_ROUTETOIF) {
 342                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
 343                                                     M_GETFIB(m)))) == NULL &&
 344                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst), 0,
 345                                                 M_GETFIB(m)))) == NULL) {
 346                         IPSTAT_INC(ips_noroute);
 347                         error = ENETUNREACH;
 348                         goto bad;
 349                 }
 350                 have_ia_ref = 1;
 351                 ifp = ia->ia_ifp;
 352                 ip->ip_ttl = 1;
 353                 isbroadcast = ifp->if_flags & IFF_BROADCAST ?
 354                     in_ifaddr_broadcast(dst->sin_addr, ia) : 0;
 355         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 356             imo != NULL && imo->imo_multicast_ifp != NULL) {
 357                 /*
 358                  * Bypass the normal routing lookup for multicast
 359                  * packets if the interface is specified.
 360                  */
 361                 ifp = imo->imo_multicast_ifp;
 362                 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
 363                 if (ia)
 364                         have_ia_ref = 1;
 365                 isbroadcast = 0;        /* fool gcc */
 366         } else {
 367                 /*
 368                  * We want to do any cloning requested by the link layer,
 369                  * as this is probably required in all cases for correct
 370                  * operation (as it is for ARP).
 371                  */
 372                 if (rte == NULL) {
 373 #ifdef RADIX_MPATH
 374                         rtalloc_mpath_fib(ro,
 375                             ntohl(ip->ip_src.s_addr ^ ip->ip_dst.s_addr),
 376                             fibnum);
 377 #else
 378                         in_rtalloc_ign(ro, 0, fibnum);
 379 #endif
 380                         rte = ro->ro_rt;
 381                 }
 382                 if (rte == NULL ||
 383                     (rte->rt_flags & RTF_UP) == 0 ||
 384                     rte->rt_ifp == NULL ||
 385                     !RT_LINK_IS_UP(rte->rt_ifp)) {
 386 #ifdef IPSEC
 387                         /*
 388                          * There is no route for this packet, but it is
 389                          * possible that a matching SPD entry exists.
 390                          */
 391                         no_route_but_check_spd = 1;
 392                         mtu = 0; /* Silence GCC warning. */
 393                         goto sendit;
 394 #endif
 395                         IPSTAT_INC(ips_noroute);
 396                         error = EHOSTUNREACH;
 397                         goto bad;
 398                 }
 399                 ia = ifatoia(rte->rt_ifa);
 400                 ifp = rte->rt_ifp;
 401                 counter_u64_add(rte->rt_pksent, 1);
 402                 rt_update_ro_flags(ro);
 403                 if (rte->rt_flags & RTF_GATEWAY)
 404                         gw = (struct sockaddr_in *)rte->rt_gateway;
 405                 if (rte->rt_flags & RTF_HOST)
 406                         isbroadcast = (rte->rt_flags & RTF_BROADCAST);
 407                 else if (ifp->if_flags & IFF_BROADCAST)
 408                         isbroadcast = in_ifaddr_broadcast(gw->sin_addr, ia);
 409                 else
 410                         isbroadcast = 0;
 411         }
 412
 413         /*
 414          * Calculate MTU.  If we have a route that is up, use that,
 415          * otherwise use the interface's MTU.
 416          */
 417         if (rte != NULL && (rte->rt_flags & (RTF_UP|RTF_HOST)))
 418                 mtu = rte->rt_mtu;
 419         else
 420                 mtu = ifp->if_mtu;
 421         /* Catch a possible divide by zero later. */
 422         KASSERT(mtu > 0, ("%s: mtu %d <= 0, rte=%p (rt_flags=0x%08x) ifp=%p",
 423             __func__, mtu, rte, (rte != NULL) ? rte->rt_flags : 0, ifp));
 424
 425         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 426                 m->m_flags |= M_MCAST;
 427                 /*
 428                  * IP destination address is multicast.  Make sure "gw"
 429                  * still points to the address in "ro".  (It may have been
 430                  * changed to point to a gateway address, above.)
 431                  */
 432                 gw = dst;
 433                 /*
 434                  * See if the caller provided any multicast options
 435                  */
 436                 if (imo != NULL) {
 437                         ip->ip_ttl = imo->imo_multicast_ttl;
 438                         if (imo->imo_multicast_vif != -1)
 439                                 ip->ip_src.s_addr =
 440                                     ip_mcast_src ?
 441                                     ip_mcast_src(imo->imo_multicast_vif) :
 442                                     INADDR_ANY;
 443                 } else
 444                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 445                 /*
 446                  * Confirm that the outgoing interface supports multicast.
 447                  */
 448                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 449                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 450                                 IPSTAT_INC(ips_noroute);
 451                                 error = ENETUNREACH;
 452                                 goto bad;
 453                         }
 454                 }
 455                 /*
 456                  * If source address not specified yet, use address
 457                  * of outgoing interface.
 458                  */
 459                 if (ip->ip_src.s_addr == INADDR_ANY) {
 460                         /* Interface may have no addresses. */
 461                         if (ia != NULL)
 462                                 ip->ip_src = IA_SIN(ia)->sin_addr;
 463                 }
 464
 465                 if ((imo == NULL && in_mcast_loop) ||
 466                     (imo && imo->imo_multicast_loop)) {
 467                         /*
 468                          * Loop back multicast datagram if not expressly
 469                          * forbidden to do so, even if we are not a member
 470                          * of the group; ip_input() will filter it later,
 471                          * thus deferring a hash lookup and mutex acquisition
 472                          * at the expense of a cheap copy using m_copym().
 473                          */
 474                         ip_mloopback(ifp, m, hlen);
 475                 } else {
 476                         /*
 477                          * If we are acting as a multicast router, perform
 478                          * multicast forwarding as if the packet had just
 479                          * arrived on the interface to which we are about
 480                          * to send.  The multicast forwarding function
 481                          * recursively calls this function, using the
 482                          * IP_FORWARDING flag to prevent infinite recursion.
 483                          *
 484                          * Multicasts that are looped back by ip_mloopback(),
 485                          * above, will be forwarded by the ip_input() routine,
 486                          * if necessary.
 487                          */
 488                         if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) {
 489                                 /*
 490                                  * If rsvp daemon is not running, do not
 491                                  * set ip_moptions. This ensures that the packet
 492                                  * is multicast and not just sent down one link
 493                                  * as prescribed by rsvpd.
 494                                  */
 495                                 if (!V_rsvp_on)
 496                                         imo = NULL;
 497                                 if (ip_mforward &&
 498                                     ip_mforward(ip, ifp, m, imo) != 0) {
 499                                         m_freem(m);
 500                                         goto done;
 501                                 }
 502                         }
 503                 }
 504
 505                 /*
 506                  * Multicasts with a time-to-live of zero may be looped-
 507                  * back, above, but must not be transmitted on a network.
 508                  * Also, multicasts addressed to the loopback interface
 509                  * are not sent -- the above call to ip_mloopback() will
 510                  * loop back a copy. ip_input() will drop the copy if
 511                  * this host does not belong to the destination group on
 512                  * the loopback interface.
 513                  */
 514                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 515                         m_freem(m);
 516                         goto done;
 517                 }
 518
 519                 goto sendit;
 520         }
 521
 522         /*
 523          * If the source address is not specified yet, use the address
 524          * of the outoing interface.
 525          */
 526         if (ip->ip_src.s_addr == INADDR_ANY) {
 527                 /* Interface may have no addresses. */
 528                 if (ia != NULL) {
 529                         ip->ip_src = IA_SIN(ia)->sin_addr;
 530                 }
 531         }
 532
 533         /*
 534          * Look for broadcast address and
 535          * verify user is allowed to send
 536          * such a packet.
 537          */
 538         if (isbroadcast) {
 539                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 540                         error = EADDRNOTAVAIL;
 541                         goto bad;
 542                 }
 543                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 544                         error = EACCES;
 545                         goto bad;
 546                 }
 547                 /* don't allow broadcast messages to be fragmented */
 548                 if (ip_len > mtu) {
 549                         error = EMSGSIZE;
 550                         goto bad;
 551                 }
 552                 m->m_flags |= M_BCAST;
 553         } else {
 554                 m->m_flags &= ~M_BCAST;
 555         }
 556
 557 sendit:
 558 #ifdef IPSEC
 559         switch(ip_ipsec_output(&m, inp, &error)) {
 560         case 1:
 561                 goto bad;
 562         case -1:
 563                 goto done;
 564         case 0:
 565         default:
 566                 break;  /* Continue with packet processing. */
 567         }
 568         /*
 569          * Check if there was a route for this packet; return error if not.
 570          */
 571         if (no_route_but_check_spd) {
 572                 IPSTAT_INC(ips_noroute);
 573                 error = EHOSTUNREACH;
 574                 goto bad;
 575         }
 576         /* Update variables that are affected by ipsec4_output(). */
 577         ip = mtod(m, struct ip *);
 578         hlen = ip->ip_hl << 2;
 579 #endif /* IPSEC */
 580
 581         /* Jump over all PFIL processing if hooks are not active. */
 582         if (PFIL_HOOKED(&V_inet_pfil_hook)) {
 583                 switch (ip_output_pfil(&m, ifp, inp, dst, &fibnum, &error)) {
 584                 case 1: /* Finished */
 585                         goto done;
 586
 587                 case 0: /* Continue normally */
 588                         ip = mtod(m, struct ip *);
 589                         break;
 590
 591                 case -1: /* Need to try again */
 592                         /* Reset everything for a new round */
 593                         RO_RTFREE(ro);
 594                         if (have_ia_ref)
 595                                 ifa_free(&ia->ia_ifa);
 596                         ro->ro_prepend = NULL;
 597                         rte = NULL;
 598                         gw = dst;
 599                         ip = mtod(m, struct ip *);
 600                         goto again;
 601
 602                 }
 603         }
 604
 605         /* 127/8 must not appear on wire - RFC1122. */
 606         if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET ||
 607             (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) {
 608                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 609                         IPSTAT_INC(ips_badaddr);
 610                         error = EADDRNOTAVAIL;
 611                         goto bad;
 612                 }
 613         }
 614
 615         m->m_pkthdr.csum_flags |= CSUM_IP;
 616         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
 617                 in_delayed_cksum(m);
 618                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 619         }
 620 #ifdef SCTP
 621         if (m->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
 622                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
 623                 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
 624         }
 625 #endif
 626
 627         /*
 628          * If small enough for interface, or the interface will take
 629          * care of the fragmentation for us, we can just send directly.
 630          */
 631         if (ip_len <= mtu ||
 632             (m->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) {
 633                 ip->ip_sum = 0;
 634                 if (m->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
 635                         ip->ip_sum = in_cksum(m, hlen);
 636                         m->m_pkthdr.csum_flags &= ~CSUM_IP;
 637                 }
 638
 639                 /*
 640                  * Record statistics for this interface address.
 641                  * With CSUM_TSO the byte/packet count will be slightly
 642                  * incorrect because we count the IP+TCP headers only
 643                  * once instead of for every generated packet.
 644                  */
 645                 if (!(flags & IP_FORWARDING) && ia) {
 646                         if (m->m_pkthdr.csum_flags & CSUM_TSO)
 647                                 counter_u64_add(ia->ia_ifa.ifa_opackets,
 648                                     m->m_pkthdr.len / m->m_pkthdr.tso_segsz);
 649                         else
 650                                 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
 651
 652                         counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len);
 653                 }
 654 #ifdef MBUF_STRESS_TEST
 655                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
 656                         m = m_fragment(m, M_NOWAIT, mbuf_frag_size);
 657 #endif
 658                 /*
 659                  * Reset layer specific mbuf flags
 660                  * to avoid confusing lower layers.
 661                  */
 662                 m_clrprotoflags(m);
 663                 IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
 664                 error = (*ifp->if_output)(ifp, m,
 665                     (const struct sockaddr *)gw, ro);
 666                 goto done;
 667         }
 668
 669         /* Balk when DF bit is set or the interface didn't support TSO. */
 670         if ((ip_off & IP_DF) || (m->m_pkthdr.csum_flags & CSUM_TSO)) {
 671                 error = EMSGSIZE;
 672                 IPSTAT_INC(ips_cantfrag);
 673                 goto bad;
 674         }
 675
 676         /*
 677          * Too large for interface; fragment if possible. If successful,
 678          * on return, m will point to a list of packets to be sent.
 679          */
 680         error = ip_fragment(ip, &m, mtu, ifp->if_hwassist);
 681         if (error)
 682                 goto bad;
 683         for (; m; m = m0) {
 684                 m0 = m->m_nextpkt;
 685                 m->m_nextpkt = 0;
 686                 if (error == 0) {
 687                         /* Record statistics for this interface address. */
 688                         if (ia != NULL) {
 689                                 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
 690                                 counter_u64_add(ia->ia_ifa.ifa_obytes,
 691                                     m->m_pkthdr.len);
 692                         }
 693                         /*
 694                          * Reset layer specific mbuf flags
 695                          * to avoid confusing upper layers.
 696                          */
 697                         m_clrprotoflags(m);
 698
 699                         IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
 700                         error = (*ifp->if_output)(ifp, m,
 701                             (const struct sockaddr *)gw, ro);
 702                 } else
 703                         m_freem(m);
 704         }
 705
 706         if (error == 0)
 707                 IPSTAT_INC(ips_fragmented);
 708
 709 done:
 710         if (ro == &iproute)
 711                 RO_RTFREE(ro);
 712         else if (rte == NULL)
 713                 /*
 714                  * If the caller supplied a route but somehow the reference
 715                  * to it has been released need to prevent the caller
 716                  * calling RTFREE on it again.
 717                  */
 718                 ro->ro_rt = NULL;
 719         if (have_ia_ref)
 720                 ifa_free(&ia->ia_ifa);
 721         return (error);
 722 bad:
 723         m_freem(m);
 724         goto done;
 725 }
 726
 727 /*
 728  * Create a chain of fragments which fit the given mtu. m_frag points to the
 729  * mbuf to be fragmented; on return it points to the chain with the fragments.
 730  * Return 0 if no error. If error, m_frag may contain a partially built
 731  * chain of fragments that should be freed by the caller.
 732  *
 733  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
 734  */
 735 int
 736 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
 737     u_long if_hwassist_flags)
 738 {
 739         int error = 0;
 740         int hlen = ip->ip_hl << 2;
 741         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
 742         int off;
 743         struct mbuf *m0 = *m_frag;      /* the original packet          */
 744         int firstlen;
 745         struct mbuf **mnext;
 746         int nfrags;
 747         uint16_t ip_len, ip_off;
 748
 749         ip_len = ntohs(ip->ip_len);
 750         ip_off = ntohs(ip->ip_off);
 751
 752         if (ip_off & IP_DF) {   /* Fragmentation not allowed */
 753                 IPSTAT_INC(ips_cantfrag);
 754                 return EMSGSIZE;
 755         }
 756
 757         /*
 758          * Must be able to put at least 8 bytes per fragment.
 759          */
 760         if (len < 8)
 761                 return EMSGSIZE;
 762
 763         /*
 764          * If the interface will not calculate checksums on
 765          * fragmented packets, then do it here.
 766          */
 767         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 768                 in_delayed_cksum(m0);
 769                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 770         }
 771 #ifdef SCTP
 772         if (m0->m_pkthdr.csum_flags & CSUM_SCTP) {
 773                 sctp_delayed_cksum(m0, hlen);
 774                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
 775         }
 776 #endif
 777         if (len > PAGE_SIZE) {
 778                 /*
 779                  * Fragment large datagrams such that each segment
 780                  * contains a multiple of PAGE_SIZE amount of data,
 781                  * plus headers. This enables a receiver to perform
 782                  * page-flipping zero-copy optimizations.
 783                  *
 784                  * XXX When does this help given that sender and receiver
 785                  * could have different page sizes, and also mtu could
 786                  * be less than the receiver's page size ?
 787                  */
 788                 int newlen;
 789
 790                 off = MIN(mtu, m0->m_pkthdr.len);
 791
 792                 /*
 793                  * firstlen (off - hlen) must be aligned on an
 794                  * 8-byte boundary
 795                  */
 796                 if (off < hlen)
 797                         goto smart_frag_failure;
 798                 off = ((off - hlen) & ~7) + hlen;
 799                 newlen = (~PAGE_MASK) & mtu;
 800                 if ((newlen + sizeof (struct ip)) > mtu) {
 801                         /* we failed, go back the default */
 802 smart_frag_failure:
 803                         newlen = len;
 804                         off = hlen + len;
 805                 }
 806                 len = newlen;
 807
 808         } else {
 809                 off = hlen + len;
 810         }
 811
 812         firstlen = off - hlen;
 813         mnext = &m0->m_nextpkt;         /* pointer to next packet */
 814
 815         /*
 816          * Loop through length of segment after first fragment,
 817          * make new header and copy data of each part and link onto chain.
 818          * Here, m0 is the original packet, m is the fragment being created.
 819          * The fragments are linked off the m_nextpkt of the original
 820          * packet, which after processing serves as the first fragment.
 821          */
 822         for (nfrags = 1; off < ip_len; off += len, nfrags++) {
 823                 struct ip *mhip;        /* ip header on the fragment */
 824                 struct mbuf *m;
 825                 int mhlen = sizeof (struct ip);
 826
 827                 m = m_gethdr(M_NOWAIT, MT_DATA);
 828                 if (m == NULL) {
 829                         error = ENOBUFS;
 830                         IPSTAT_INC(ips_odropped);
 831                         goto done;
 832                 }
 833                 /*
 834                  * Make sure the complete packet header gets copied
 835                  * from the originating mbuf to the newly created
 836                  * mbuf. This also ensures that existing firewall
 837                  * classification(s), VLAN tags and so on get copied
 838                  * to the resulting fragmented packet(s):
 839                  */
 840                 if (m_dup_pkthdr(m, m0, M_NOWAIT) == 0) {
 841                         m_free(m);
 842                         error = ENOBUFS;
 843                         IPSTAT_INC(ips_odropped);
 844                         goto done;
 845                 }
 846                 /*
 847                  * In the first mbuf, leave room for the link header, then
 848                  * copy the original IP header including options. The payload
 849                  * goes into an additional mbuf chain returned by m_copym().
 850                  */
 851                 m->m_data += max_linkhdr;
 852                 mhip = mtod(m, struct ip *);
 853                 *mhip = *ip;
 854                 if (hlen > sizeof (struct ip)) {
 855                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
 856                         mhip->ip_v = IPVERSION;
 857                         mhip->ip_hl = mhlen >> 2;
 858                 }
 859                 m->m_len = mhlen;
 860                 /* XXX do we need to add ip_off below ? */
 861                 mhip->ip_off = ((off - hlen) >> 3) + ip_off;
 862                 if (off + len >= ip_len)
 863                         len = ip_len - off;
 864                 else
 865                         mhip->ip_off |= IP_MF;
 866                 mhip->ip_len = htons((u_short)(len + mhlen));
 867                 m->m_next = m_copym(m0, off, len, M_NOWAIT);
 868                 if (m->m_next == NULL) {        /* copy failed */
 869                         m_free(m);
 870                         error = ENOBUFS;        /* ??? */
 871                         IPSTAT_INC(ips_odropped);
 872                         goto done;
 873                 }
 874                 m->m_pkthdr.len = mhlen + len;
 875 #ifdef MAC
 876                 mac_netinet_fragment(m0, m);
 877 #endif
 878                 mhip->ip_off = htons(mhip->ip_off);
 879                 mhip->ip_sum = 0;
 880                 if (m->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
 881                         mhip->ip_sum = in_cksum(m, mhlen);
 882                         m->m_pkthdr.csum_flags &= ~CSUM_IP;
 883                 }
 884                 *mnext = m;
 885                 mnext = &m->m_nextpkt;
 886         }
 887         IPSTAT_ADD(ips_ofragments, nfrags);
 888
 889         /*
 890          * Update first fragment by trimming what's been copied out
 891          * and updating header.
 892          */
 893         m_adj(m0, hlen + firstlen - ip_len);
 894         m0->m_pkthdr.len = hlen + firstlen;
 895         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
 896         ip->ip_off = htons(ip_off | IP_MF);
 897         ip->ip_sum = 0;
 898         if (m0->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
 899                 ip->ip_sum = in_cksum(m0, hlen);
 900                 m0->m_pkthdr.csum_flags &= ~CSUM_IP;
 901         }
 902
 903 done:
 904         *m_frag = m0;
 905         return error;
 906 }
 907
 908 void
 909 in_delayed_cksum(struct mbuf *m)
 910 {
 911         struct ip *ip;
 912         uint16_t csum, offset, ip_len;
 913
 914         ip = mtod(m, struct ip *);
 915         offset = ip->ip_hl << 2 ;
 916         ip_len = ntohs(ip->ip_len);
 917         csum = in_cksum_skip(m, ip_len, offset);
 918         if (m->m_pkthdr.csum_flags & CSUM_UDP && csum == 0)
 919                 csum = 0xffff;
 920         offset += m->m_pkthdr.csum_data;        /* checksum offset */
 921
 922         /* find the mbuf in the chain where the checksum starts*/
 923         while ((m != NULL) && (offset >= m->m_len)) {
 924                 offset -= m->m_len;
 925                 m = m->m_next;
 926         }
 927         KASSERT(m != NULL, ("in_delayed_cksum: checksum outside mbuf chain."));
 928         KASSERT(offset + sizeof(u_short) <= m->m_len, ("in_delayed_cksum: checksum split between mbufs."));
 929         *(u_short *)(m->m_data + offset) = csum;
 930 }
 931
 932 /*
 933  * IP socket option processing.
 934  */
 935 int
 936 ip_ctloutput(struct socket *so, struct sockopt *sopt)
 937 {
 938         struct  inpcb *inp = sotoinpcb(so);
 939         int     error, optval;
 940 #ifdef  RSS
 941         uint32_t rss_bucket;
 942         int retval;
 943 #endif
 944
 945         error = optval = 0;
 946         if (sopt->sopt_level != IPPROTO_IP) {
 947                 error = EINVAL;
 948
 949                 if (sopt->sopt_level == SOL_SOCKET &&
 950                     sopt->sopt_dir == SOPT_SET) {
 951                         switch (sopt->sopt_name) {
 952                         case SO_REUSEADDR:
 953                                 INP_WLOCK(inp);
 954                                 if ((so->so_options & SO_REUSEADDR) != 0)
 955                                         inp->inp_flags2 |= INP_REUSEADDR;
 956                                 else
 957                                         inp->inp_flags2 &= ~INP_REUSEADDR;
 958                                 INP_WUNLOCK(inp);
 959                                 error = 0;
 960                                 break;
 961                         case SO_REUSEPORT:
 962                                 INP_WLOCK(inp);
 963                                 if ((so->so_options & SO_REUSEPORT) != 0)
 964                                         inp->inp_flags2 |= INP_REUSEPORT;
 965                                 else
 966                                         inp->inp_flags2 &= ~INP_REUSEPORT;
 967                                 INP_WUNLOCK(inp);
 968                                 error = 0;
 969                                 break;
 970                         case SO_SETFIB:
 971                                 INP_WLOCK(inp);
 972                                 inp->inp_inc.inc_fibnum = so->so_fibnum;
 973                                 INP_WUNLOCK(inp);
 974                                 error = 0;
 975                                 break;
 976                         default:
 977                                 break;
 978                         }
 979                 }
 980                 return (error);
 981         }
 982
 983         switch (sopt->sopt_dir) {
 984         case SOPT_SET:
 985                 switch (sopt->sopt_name) {
 986                 case IP_OPTIONS:
 987 #ifdef notyet
 988                 case IP_RETOPTS:
 989 #endif
 990                 {
 991                         struct mbuf *m;
 992                         if (sopt->sopt_valsize > MLEN) {
 993                                 error = EMSGSIZE;
 994                                 break;
 995                         }
 996                         m = m_get(sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
 997                         if (m == NULL) {
 998                                 error = ENOBUFS;
 999                                 break;
1000                         }
1001                         m->m_len = sopt->sopt_valsize;
1002                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1003                                             m->m_len);
1004                         if (error) {
1005                                 m_free(m);
1006                                 break;
1007                         }
1008                         INP_WLOCK(inp);
1009                         error = ip_pcbopts(inp, sopt->sopt_name, m);
1010                         INP_WUNLOCK(inp);
1011                         return (error);
1012                 }
1013
1014                 case IP_BINDANY:
1015                         if (sopt->sopt_td != NULL) {
1016                                 error = priv_check(sopt->sopt_td,
1017                                     PRIV_NETINET_BINDANY);
1018                                 if (error)
1019                                         break;
1020                         }
1021                         /* FALLTHROUGH */
1022                 case IP_BINDMULTI:
1023 #ifdef  RSS
1024                 case IP_RSS_LISTEN_BUCKET:
1025 #endif
1026                 case IP_TOS:
1027                 case IP_TTL:
1028                 case IP_MINTTL:
1029                 case IP_RECVOPTS:
1030                 case IP_RECVRETOPTS:
1031                 case IP_RECVDSTADDR:
1032                 case IP_RECVTTL:
1033                 case IP_RECVIF:
1034                 case IP_ONESBCAST:
1035                 case IP_DONTFRAG:
1036                 case IP_RECVTOS:
1037                 case IP_RECVFLOWID:
1038 #ifdef  RSS
1039                 case IP_RECVRSSBUCKETID:
1040 #endif
1041                         error = sooptcopyin(sopt, &optval, sizeof optval,
1042                                             sizeof optval);
1043                         if (error)
1044                                 break;
1045
1046                         switch (sopt->sopt_name) {
1047                         case IP_TOS:
1048                                 inp->inp_ip_tos = optval;
1049                                 break;
1050
1051                         case IP_TTL:
1052                                 inp->inp_ip_ttl = optval;
1053                                 break;
1054
1055                         case IP_MINTTL:
1056                                 if (optval >= 0 && optval <= MAXTTL)
1057                                         inp->inp_ip_minttl = optval;
1058                                 else
1059                                         error = EINVAL;
1060                                 break;
1061
1062 #define OPTSET(bit) do {                                                \
1063         INP_WLOCK(inp);                                                 \
1064         if (optval)                                                     \
1065                 inp->inp_flags |= bit;                                  \
1066         else                                                            \
1067                 inp->inp_flags &= ~bit;                                 \
1068         INP_WUNLOCK(inp);                                               \
1069 } while (0)
1070
1071 #define OPTSET2(bit, val) do {                                          \
1072         INP_WLOCK(inp);                                                 \
1073         if (val)                                                        \
1074                 inp->inp_flags2 |= bit;                                 \
1075         else                                                            \
1076                 inp->inp_flags2 &= ~bit;                                \
1077         INP_WUNLOCK(inp);                                               \
1078 } while (0)
1079
1080                         case IP_RECVOPTS:
1081                                 OPTSET(INP_RECVOPTS);
1082                                 break;
1083
1084                         case IP_RECVRETOPTS:
1085                                 OPTSET(INP_RECVRETOPTS);
1086                                 break;
1087
1088                         case IP_RECVDSTADDR:
1089                                 OPTSET(INP_RECVDSTADDR);
1090                                 break;
1091
1092                         case IP_RECVTTL:
1093                                 OPTSET(INP_RECVTTL);
1094                                 break;
1095
1096                         case IP_RECVIF:
1097                                 OPTSET(INP_RECVIF);
1098                                 break;
1099
1100                         case IP_ONESBCAST:
1101                                 OPTSET(INP_ONESBCAST);
1102                                 break;
1103                         case IP_DONTFRAG:
1104                                 OPTSET(INP_DONTFRAG);
1105                                 break;
1106                         case IP_BINDANY:
1107                                 OPTSET(INP_BINDANY);
1108                                 break;
1109                         case IP_RECVTOS:
1110                                 OPTSET(INP_RECVTOS);
1111                                 break;
1112                         case IP_BINDMULTI:
1113                                 OPTSET2(INP_BINDMULTI, optval);
1114                                 break;
1115                         case IP_RECVFLOWID:
1116                                 OPTSET2(INP_RECVFLOWID, optval);
1117                                 break;
1118 #ifdef  RSS
1119                         case IP_RSS_LISTEN_BUCKET:
1120                                 if ((optval >= 0) &&
1121                                     (optval < rss_getnumbuckets())) {
1122                                         inp->inp_rss_listen_bucket = optval;
1123                                         OPTSET2(INP_RSS_BUCKET_SET, 1);
1124                                 } else {
1125                                         error = EINVAL;
1126                                 }
1127                                 break;
1128                         case IP_RECVRSSBUCKETID:
1129                                 OPTSET2(INP_RECVRSSBUCKETID, optval);
1130                                 break;
1131 #endif
1132                         }
1133                         break;
1134 #undef OPTSET
1135 #undef OPTSET2
1136
1137                 /*
1138                  * Multicast socket options are processed by the in_mcast
1139                  * module.
1140                  */
1141                 case IP_MULTICAST_IF:
1142                 case IP_MULTICAST_VIF:
1143                 case IP_MULTICAST_TTL:
1144                 case IP_MULTICAST_LOOP:
1145                 case IP_ADD_MEMBERSHIP:
1146                 case IP_DROP_MEMBERSHIP:
1147                 case IP_ADD_SOURCE_MEMBERSHIP:
1148                 case IP_DROP_SOURCE_MEMBERSHIP:
1149                 case IP_BLOCK_SOURCE:
1150                 case IP_UNBLOCK_SOURCE:
1151                 case IP_MSFILTER:
1152                 case MCAST_JOIN_GROUP:
1153                 case MCAST_LEAVE_GROUP:
1154                 case MCAST_JOIN_SOURCE_GROUP:
1155                 case MCAST_LEAVE_SOURCE_GROUP:
1156                 case MCAST_BLOCK_SOURCE:
1157                 case MCAST_UNBLOCK_SOURCE:
1158                         error = inp_setmoptions(inp, sopt);
1159                         break;
1160
1161                 case IP_PORTRANGE:
1162                         error = sooptcopyin(sopt, &optval, sizeof optval,
1163                                             sizeof optval);
1164                         if (error)
1165                                 break;
1166
1167                         INP_WLOCK(inp);
1168                         switch (optval) {
1169                         case IP_PORTRANGE_DEFAULT:
1170                                 inp->inp_flags &= ~(INP_LOWPORT);
1171                                 inp->inp_flags &= ~(INP_HIGHPORT);
1172                                 break;
1173
1174                         case IP_PORTRANGE_HIGH:
1175                                 inp->inp_flags &= ~(INP_LOWPORT);
1176                                 inp->inp_flags |= INP_HIGHPORT;
1177                                 break;
1178
1179                         case IP_PORTRANGE_LOW:
1180                                 inp->inp_flags &= ~(INP_HIGHPORT);
1181                                 inp->inp_flags |= INP_LOWPORT;
1182                                 break;
1183
1184                         default:
1185                                 error = EINVAL;
1186                                 break;
1187                         }
1188                         INP_WUNLOCK(inp);
1189                         break;
1190
1191 #ifdef IPSEC
1192                 case IP_IPSEC_POLICY:
1193                 {
1194                         caddr_t req;
1195                         struct mbuf *m;
1196
1197                         if ((error = soopt_getm(sopt, &m)) != 0) /* XXX */
1198                                 break;
1199                         if ((error = soopt_mcopyin(sopt, m)) != 0) /* XXX */
1200                                 break;
1201                         req = mtod(m, caddr_t);
1202                         error = ipsec_set_policy(inp, sopt->sopt_name, req,
1203                             m->m_len, (sopt->sopt_td != NULL) ?
1204                             sopt->sopt_td->td_ucred : NULL);
1205                         m_freem(m);
1206                         break;
1207                 }
1208 #endif /* IPSEC */
1209
1210                 default:
1211                         error = ENOPROTOOPT;
1212                         break;
1213                 }
1214                 break;
1215
1216         case SOPT_GET:
1217                 switch (sopt->sopt_name) {
1218                 case IP_OPTIONS:
1219                 case IP_RETOPTS:
1220                         if (inp->inp_options)
1221                                 error = sooptcopyout(sopt,
1222                                                      mtod(inp->inp_options,
1223                                                           char *),
1224                                                      inp->inp_options->m_len);
1225                         else
1226                                 sopt->sopt_valsize = 0;
1227                         break;
1228
1229                 case IP_TOS:
1230                 case IP_TTL:
1231                 case IP_MINTTL:
1232                 case IP_RECVOPTS:
1233                 case IP_RECVRETOPTS:
1234                 case IP_RECVDSTADDR:
1235                 case IP_RECVTTL:
1236                 case IP_RECVIF:
1237                 case IP_PORTRANGE:
1238                 case IP_ONESBCAST:
1239                 case IP_DONTFRAG:
1240                 case IP_BINDANY:
1241                 case IP_RECVTOS:
1242                 case IP_BINDMULTI:
1243                 case IP_FLOWID:
1244                 case IP_FLOWTYPE:
1245                 case IP_RECVFLOWID:
1246 #ifdef  RSS
1247                 case IP_RSSBUCKETID:
1248                 case IP_RECVRSSBUCKETID:
1249 #endif
1250                         switch (sopt->sopt_name) {
1251
1252                         case IP_TOS:
1253                                 optval = inp->inp_ip_tos;
1254                                 break;
1255
1256                         case IP_TTL:
1257                                 optval = inp->inp_ip_ttl;
1258                                 break;
1259
1260                         case IP_MINTTL:
1261                                 optval = inp->inp_ip_minttl;
1262                                 break;
1263
1264 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1265 #define OPTBIT2(bit)    (inp->inp_flags2 & bit ? 1 : 0)
1266
1267                         case IP_RECVOPTS:
1268                                 optval = OPTBIT(INP_RECVOPTS);
1269                                 break;
1270
1271                         case IP_RECVRETOPTS:
1272                                 optval = OPTBIT(INP_RECVRETOPTS);
1273                                 break;
1274
1275                         case IP_RECVDSTADDR:
1276                                 optval = OPTBIT(INP_RECVDSTADDR);
1277                                 break;
1278
1279                         case IP_RECVTTL:
1280                                 optval = OPTBIT(INP_RECVTTL);
1281                                 break;
1282
1283                         case IP_RECVIF:
1284                                 optval = OPTBIT(INP_RECVIF);
1285                                 break;
1286
1287                         case IP_PORTRANGE:
1288                                 if (inp->inp_flags & INP_HIGHPORT)
1289                                         optval = IP_PORTRANGE_HIGH;
1290                                 else if (inp->inp_flags & INP_LOWPORT)
1291                                         optval = IP_PORTRANGE_LOW;
1292                                 else
1293                                         optval = 0;
1294                                 break;
1295
1296                         case IP_ONESBCAST:
1297                                 optval = OPTBIT(INP_ONESBCAST);
1298                                 break;
1299                         case IP_DONTFRAG:
1300                                 optval = OPTBIT(INP_DONTFRAG);
1301                                 break;
1302                         case IP_BINDANY:
1303                                 optval = OPTBIT(INP_BINDANY);
1304                                 break;
1305                         case IP_RECVTOS:
1306                                 optval = OPTBIT(INP_RECVTOS);
1307                                 break;
1308                         case IP_FLOWID:
1309                                 optval = inp->inp_flowid;
1310                                 break;
1311                         case IP_FLOWTYPE:
1312                                 optval = inp->inp_flowtype;
1313                                 break;
1314                         case IP_RECVFLOWID:
1315                                 optval = OPTBIT2(INP_RECVFLOWID);
1316                                 break;
1317 #ifdef  RSS
1318                         case IP_RSSBUCKETID:
1319                                 retval = rss_hash2bucket(inp->inp_flowid,
1320                                     inp->inp_flowtype,
1321                                     &rss_bucket);
1322                                 if (retval == 0)
1323                                         optval = rss_bucket;
1324                                 else
1325                                         error = EINVAL;
1326                                 break;
1327                         case IP_RECVRSSBUCKETID:
1328                                 optval = OPTBIT2(INP_RECVRSSBUCKETID);
1329                                 break;
1330 #endif
1331                         case IP_BINDMULTI:
1332                                 optval = OPTBIT2(INP_BINDMULTI);
1333                                 break;
1334                         }
1335                         error = sooptcopyout(sopt, &optval, sizeof optval);
1336                         break;
1337
1338                 /*
1339                  * Multicast socket options are processed by the in_mcast
1340                  * module.
1341                  */
1342                 case IP_MULTICAST_IF:
1343                 case IP_MULTICAST_VIF:
1344                 case IP_MULTICAST_TTL:
1345                 case IP_MULTICAST_LOOP:
1346                 case IP_MSFILTER:
1347                         error = inp_getmoptions(inp, sopt);
1348                         break;
1349
1350 #ifdef IPSEC
1351                 case IP_IPSEC_POLICY:
1352                 {
1353                         struct mbuf *m = NULL;
1354                         caddr_t req = NULL;
1355                         size_t len = 0;
1356
1357                         if (m != NULL) {
1358                                 req = mtod(m, caddr_t);
1359                                 len = m->m_len;
1360                         }
1361                         error = ipsec_get_policy(sotoinpcb(so), req, len, &m);
1362                         if (error == 0)
1363                                 error = soopt_mcopyout(sopt, m); /* XXX */
1364                         if (error == 0)
1365                                 m_freem(m);
1366                         break;
1367                 }
1368 #endif /* IPSEC */
1369
1370                 default:
1371                         error = ENOPROTOOPT;
1372                         break;
1373                 }
1374                 break;
1375         }
1376         return (error);
1377 }
1378
1379 /*
1380  * Routine called from ip_output() to loop back a copy of an IP multicast
1381  * packet to the input queue of a specified interface.  Note that this
1382  * calls the output routine of the loopback "driver", but with an interface
1383  * pointer that might NOT be a loopback interface -- evil, but easier than
1384  * replicating that code here.
1385  */
1386 static void
1387 ip_mloopback(struct ifnet *ifp, const struct mbuf *m, int hlen)
1388 {
1389         struct ip *ip;
1390         struct mbuf *copym;
1391
1392         /*
1393          * Make a deep copy of the packet because we're going to
1394          * modify the pack in order to generate checksums.
1395          */
1396         copym = m_dup(m, M_NOWAIT);
1397         if (copym != NULL && (!M_WRITABLE(copym) || copym->m_len < hlen))
1398                 copym = m_pullup(copym, hlen);
1399         if (copym != NULL) {
1400                 /* If needed, compute the checksum and mark it as valid. */
1401                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1402                         in_delayed_cksum(copym);
1403                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1404                         copym->m_pkthdr.csum_flags |=
1405                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1406                         copym->m_pkthdr.csum_data = 0xffff;
1407                 }
1408                 /*
1409                  * We don't bother to fragment if the IP length is greater
1410                  * than the interface's MTU.  Can this possibly matter?
1411                  */
1412                 ip = mtod(copym, struct ip *);
1413                 ip->ip_sum = 0;
1414                 ip->ip_sum = in_cksum(copym, hlen);
1415                 if_simloop(ifp, copym, AF_INET, 0);
1416         }
1417 }